ImageSharp/HEIF_IMPLEMENTATION_PLAN.md

HEIF family implementation plan

Goal

Complete a production-quality, fully managed HEIF family image-format implementation for ImageSharp. The implementation must support still and animated HEIC, HIF/HEIF, and AVIF image files, interoperate with independent encoders and decoders, follow the existing ImageSharp architecture and code style, reuse existing ImageSharp infrastructure wherever its semantics match, and use SIMD for measured hot paths without maintaining a separate behavior model.

HEIF is the shared ISO BMFF-derived container. HEIC carries HEVC image items, AVIF carries AV1 image items, and .hif/.heif are container extensions whose payload codec must be determined from brands and item types rather than the filename. The completed implementation will support HEVC, AV1, and legacy JPEG image items. Other registered HEIF payload codecs must not be advertised unless they are implemented and independently verified.

Completion boundary

This plan has one PR completion gate. The phases below are dependency order and internal verification points only; none is a separately releasable or merge-complete subset. The PR is not complete until the complete matrix is implemented and independently verified.

Full completion includes:

• XML documentation covers every type and contract in the HEIF implementation, including the AV1 and HEVC codec internals, and inline comments explain non-obvious container layouts, bitstream rules, numerical algorithms, SIMD choices, and interoperability constraints.
• HEIF, HEIC, and AVIF brands and item types are detected without relying on file extensions.
• HEVC/HEIC and AV1/AVIF decode and encode are complete for every image behavior included in the final scope.
• Legacy JPEG image-item decode and encode are standards compliant for HEIF/HIF files.
• Every bit depth and chroma format permitted by the exposed HEVC and AV1 profiles is supported, including 8, 10, and 12-bit and monochrome, YUV 4:2:0, 4:2:2, and 4:4:4 paths.
• Full- and limited-range conversion supports every valid signaled color-primary, transfer-characteristic, matrix-coefficient, and chroma-sample-position combination, including identity RGB signaling.
• Every normative AV1 compression tool permitted by the final image scope decodes without an unsupported branch or silently skipped reconstruction.
• The AV1 encoder implements real lossy and lossless compression with working quality and effort controls, complete mode decision, prediction, transform, quantization, entropy coding, and legal in-loop filter decisions.
• Every normative HEVC compression tool permitted by the final image scope decodes across the exposed profiles, including the range-extension tools required for high bit depth and 4:2:2/4:4:4.
• The HEVC encoder implements real lossy and lossless compression with working quality and effort controls, complete coding-tree, prediction, transform, quantization, CABAC, deblocking, and sample-adaptive-offset decisions.
• Primary images, alpha auxiliary images, image grids, ICC and CICP color information, Exif, XMP, pixel aspect ratio, clean aperture, rotation, and mirroring are complete.
• AVIF interoperability is independently verified with libavif and libaom, and HEIC interoperability is independently verified with a separately selected HEVC/HEIF implementation.

Checkboxes may be marked complete only when the implementation and the verification required by the corresponding phase exit gate are both recorded. Source presence, compilation, self-roundtrip, or a temporary vertical slice is not completion evidence.

Delivery dashboard

Last reconciled with the source tree on 2026-08-30 against production checkpoint 640b64ee5 and the current local HEVC precision and presentation corrections. Committed checkpoints include the AV1 transform architecture, OBU framing, tile partitioning, mode information, segmentation, delta quantization, transform-size selection, coefficient decoding, inverse quantization, inverse transforms, complete intra prediction, intra-block copy, official extreme-displacement intra-block-copy conformance, the 12-profile reconstruction matrix, corrected twelve-bit inverse-transform SIMD arithmetic, layered-item properties, layered reference/header/CDF/motion-field state, inter-frame intra blocks, SIMD-first translational prediction, complete single-reference inter reconstruction, compound reference trees and modes, paired reference-MV derivation, reference-dependent bounded sequence decoding, allocation-free SIMD-first equal averaging, selected inter-intra prediction, selectable compound blending, OBMC, scaled-reference reconstruction, local warped prediction, non-translational global prediction, official motion-vector conformance, official spatial- and temporal-layer default-operating-point conformance, official active-film-grain and monochrome sequence conformance, selected spatial-layer presentation, progressive color and auxiliary-alpha conformance, pinned-HM HEVC CABAC verification, pinned-HM HEVC coding-tree traversal verification, official HEVC intra-prediction verification, official HEVC residual-reconstruction verification, HEVC parallel slice decoding, bounded HEVC supplemental presentation and metadata handling, and the complete independently coded HEVC Range Extensions native-plane profile/tool matrix. This dashboard is the authoritative delivery order. The detailed phase checklists below provide subsystem evidence; they do not override the current-stage marker or permit work to skip ahead.

Commit a0d1b5af56acb3a3a9aebb29c916d1b36cd49367 restores the established JPEG color-converter operator architecture throughout the HEIF implementation. Av1ForwardTransformer owns its nested one-dimensional contract and all twelve concrete forward transform operators; Av1Inverse2dTransformer owns the corresponding inverse contract and operators; and Av1InverseTransformer owns the byte and high-bit-depth reconstruction-output operators. Each contract lives in the owner's .Operator.cs, while every semantic transform with a shared traversal has its own <Owner>.<Semantic>Operator.cs partial-family file and implements concrete scalar and SIMD overload bodies without a forwarding semantic core. The same owner-and-partial-file rule now covers HEVC inverse transforms and intra prediction, AV1 and HEVC deblocking, AV1 CDEF, and HEVC transform-skip reconstruction. Vector-width and sample-storage adapters that are not semantic dispatch operators are named Operations; no operator type is named after a hardware width, no static contract uses CRTP, no source file groups multiple concrete semantic operator structs, and no semantic operator remains top-level. Roslynk reports zero compiler errors, both Release source targets build with zero warnings and errors, and 736 focused net10.0 transform, loop-filter, prediction, reconstruction, transfer-function, and film-grain cases pass without failures or skips through the existing FeatureTestRunner coverage. All 56 new C# paths resolve to the existing csharp diff driver without an attributes change, and git diff --check is clean.

Commit 35e65e9546dd78847d4de7cae082e26229b3a744 closes the Phase 4 HEVC CABAC checkpoint without a production correction. Exact comparison with HM commit 9c1f298659ab0cee9dc13d23d0304221575410b9 confirms the context-initialization equation, all 178 retained intra-picture initialization values, both 128-entry probability-state transition tables, the 64-by-4 least-probable-symbol range table, the 32-entry normalization table, arithmetic decision and refill paths, ordinary and aligned bypass paths, termination, alignment validation, and PCM restart. Focused tests cover QP 0, 22, and 51 initialization, every reachable packed context state, both arithmetic outcomes, bypass, termination, invalid alignment, and PCM suspension/restart. A fresh Release run passes all 26 CABAC and production complete-slice cases without failures or skips: ten official Sony GENERAL Range Extensions first pictures match their exact native planes and published digests, five real HEIC item payloads match HM output exactly, and both constructor-failure and constrained-allocation ownership cases remain balanced. Roslynk reports zero compiler errors, scoped StyleCop and whitespace verification are clean, and git diff --check is clean.

Commit 66d1f3c41b0baeaa98726b57d5feaeba50874f16 closes the Phase 4 HEVC coding-tree traversal checkpoint. Exact comparison with HM commit 9c1f298659ab0cee9dc13d23d0304221575410b9 covers boundary-forced coding-tree splits, split-context derivation, quantization-group QP references, intra partition-size syntax, PCM gating and CABAC restart, minimum and maximum transform-size split inference, chroma coded-block-flag inheritance including 4:2:2 sub-transform units, and first-residual ordering for luma and chroma QP updates. Official ITU RQT_A through RQT_E, STRUCT_A and STRUCT_B, and TUSIZE_A bitstreams match the archive's complete-sequence MD5 values through pinned HM; every RQT decoded-picture hash SEI also reports OK. Pinned-HM analyser output proves minimum-CU intra partition alternatives at 8, 16, and 32 samples, luma prediction sizes 4 through 64, coding-tree split syntax at 16 through 64, transform-subdivision syntax at 8 through 32, and luma transform blocks at 4 through 32, while TUSIZE_A covers 16- and 32-sample luma and 8-, 16-, and 32-sample chroma transforms. Eight production still-item tests compare the first picture's native Y, Cb, and Cr planes exactly with pinned HM. A fresh Release run passes all 150 focused HEVC decoder cases without failures or skips, the net11.0 source build completes with zero warnings and errors, Roslynk reports zero compiler errors, scoped formatting and StyleCop verification are clean, and git diff --check is clean.

Commit 9bc5b9842 closes the Phase 4 HEVC intra-prediction checkpoint without a production correction. Exact comparison with HM commit 9c1f298659ab0cee9dc13d23d0304221575410b9 confirms reference availability and substitution, normal three-tap and strong bilinear smoothing, filter-selection thresholds, planar, DC, and all angular equations, luma edge filters, negative-angle reference extension, 4:2:2 chroma-angle remapping, luma most-probable-mode syntax, chroma-mode syntax, and combined 4:4:4 NxN chroma traversal. The official one-picture IPRED_B_Nokia_3 Main Still Picture stream exercises all 35 modes at every published luma and chroma conformance size, for 245 mode-and-size combinations, and its complete native YUV output matches the published reference exactly. The independently coded first picture from official CIP_A_Panasonic_3 verifies the enabled constrained-intra PPS path and exact pinned-HM decoded-plane hashes; it does not claim coverage of constrained inter-neighbor exclusion in the following B picture, which is outside the bounded still decoder. FeatureTestRunner verifies the predictor oracle and both reference-filter paths through normal, AVX-512-disabled, AVX-disabled, and scalar execution. A fresh Release run passes all 149 focused HEVC decoder cases without failures or skips, both Release source targets build with zero warnings and errors, Roslynk reports zero compiler errors, scoped StyleCop and whitespace verification are clean, and git diff --check is clean.

Commit 97538abd29cb69df0ecf3e30a4ee3b40b586c070 closes the independently coded Phase 4 residual-reconstruction checkpoint. Five official HEVC v1 and Range Extensions pictures match the native Y, Cb, and Cr plane hashes produced by pinned HM commit 9c1f298659ab0cee9dc13d23d0304221575410b9. The retained pictures cover coding-unit luma-QP deltas, scaling lists, transform skip, transform-skip contexts and rotation, implicit residual DPCM, extended-precision processing at 12-bit 4:4:4, coding-unit chroma-QP adjustment, and complete transform-and-quantization bypass. The DELTAQP_A_BRCM_4 comparison exposed and corrected the PCM-neighbor invariant: PCM coding units skip intra-mode syntax but remain most-probable-mode neighbors, so the allocator-backed luma-mode map now starts at HM's required DC default. The five new exact-picture cases and all 169 focused HEVC decoder cases pass without failures or skips. Both Release source targets and the scoped Release test project build with zero warnings and errors, Roslynk reports zero compiler errors, the retained fixture hashes and existing LFS attributes match their recorded provenance without an attributes change, scoped whitespace verification is clean, and git diff --check is clean. Explicit inter residual-DPCM syntax remains unimplemented and is not claimed by this checkpoint.

Commit fcd16c862c437e3249c7e693012c717512985163 closes the bounded HEVC supplemental-presentation checkpoint. The item parser now consumes prefix-SEI messages in access-unit order, supports extended payload types and sizes, skips unknown payloads, rejects misplaced prefix messages and malformed trailing bits, and maps pinned-HM display orientation, no-display, mastering-display color volume, content light level, alternative transfer characteristics, ambient viewing environment, and content color volume into the bounded ImageSharp result. Codec orientation is applied after item scaling and auxiliary-alpha composition but before container crop, rotation, and mirroring, so a real public HEIC with alpha matches its independent reference PNG at every final RGBA sample after the injected transform. Metadata conflicts and no-display fail at the item boundary, while optional video timing, persistence, and post-processing messages without an ImageSharp still-image result or exposed metadata mapping remain intentionally ignored. FeatureTestRunner verifies normal and scalar presentation paths through an 8 KiB constrained tracked allocator with balanced exactly-once returns. All 195 focused HEVC cases pass without failures or skips, both Release source targets build with zero warnings and errors, Roslynk reports zero compiler errors, scoped StyleCop and whitespace verification are clean, .gitattributes is unchanged, and git diff --check is clean.

Commit 9d7348efd6496b54cef7ec27e42352641439c400 completes the independently coded HEVC Range Extensions native-plane profile and tool matrix. Twenty-five selected official H.265.1 pictures extend the existing ten Sony GENERAL pictures with Main RExt and High Throughput RExt coverage for cross-component prediction, unequal luma/chroma precision, persistent Rice adaptation, extended precision, PCM, transform-skip contexts, chroma-angle adjustment, coding-unit chroma-QP adjustment, tiles, wavefront entry points, and CABAC bypass alignment. Every selected Y, Cb, and Cr plane matches its normative decoded-picture-hash SEI or a first reference picture whose complete output matches the archive's published MD5. The matrix exposed and corrected PCM sample restoration from signaled PCM precision to reconstructed component precision and the chroma-deblocking rule that excludes slice and coding-unit chroma-QP adjustments from tc derivation; dead per-coding-unit deblocking-offset storage was removed. The exact 25-case matrix and all 220 focused HEVC cases pass in net10.0 Release, both Release source targets build with zero warnings and errors, Roslynk reports zero compiler errors, all 24 retained fixture SHA-256 values and staged LFS pointers match the documented existing attributes, .gitattributes is unchanged, and git diff --check is clean. This checkpoint proves native reconstruction for the independently coded Phase 4 scope; it does not prove the separate every-profile HEIC container-presentation exit gate.

Commit 1c58d855f70b024170ced9eb0a7005f0f9c955ad records the complete official motion-vector conformance checkpoint. The official IVF has SHA-1 F064290D7FCD3B3DE19020E8AEC6C43C88D3A505, matching the pinned libaom test-data manifest, and SHA-256 222A9050059B254DAB17CFB802FF829C778E3F93AF18961A622C8268576C1395; its pinned-libaom Y4M has SHA-256 D97AC78C81782CF1507549368047769DC677DBE205706458D1EE9C807DE6EC78. Both source targets build with zero warnings and errors, the net10.0 test-project analyzer build completes with zero errors and 1,014 pre-existing repository warnings, Roslynk reports zero compiler errors, 3,985 focused net10.0 cases pass without failures or skips, and git diff --check is clean.

Commit 019ac5648b4380a6ff4865072c96fde29ce09fad records the complete selected-spatial-layer presentation checkpoint. An essential lsel selecting the genuine 40x40 base layer of the committed 40x40-to-80x80 progressive AVIF returns that exact pinned-libaom native reconstruction, then scales its YUV444 planes to the 80x80 ispe extent before color conversion and matches pinned libavif's RGBA output exactly. Retained reconstructed planes remain at coded dimensions; presentation planes use group-safe allocator-owned rows and are returned exactly once under a 1,024-byte constrained allocator. Portable Vector128 and scalar two-times filters match pinned libyuv edge, quarter-sample, and rounding results at 8 and 12 bits through FeatureTestRunner. Both source targets build with zero warnings and errors, the net10.0 test-project analyzer build completes with zero errors and 1,014 pre-existing repository warnings, Roslynk reports zero compiler errors, 156 focused reconstruction and color cases pass without failures or skips, and git diff --check is clean.

Commit beb2ab86ce5e4052c667df30af8a48a860814031 records the complete progressive color and auxiliary-alpha conformance checkpoint. Both pinned progressive AVIF fixtures pass their exact native-plane and pinned-libavif presentation comparisons through normal dispatch, AVX-512-disabled, AVX-disabled, and scalar FeatureTestRunner configurations. The YUV444-alpha fixture also compares every composed alpha sample with the independent final native auxiliary plane. Both complete production presentation paths pass under a 1,024-byte constrained tracked allocator with every allocation returned exactly once. Both source targets build with zero warnings and errors, the net10.0 test-project analyzer build completes with zero errors and 1,014 pre-existing repository warnings, Roslynk reports zero compiler errors, 156 focused reconstruction and color cases pass without failures or skips, and git diff --check is clean.

Commit ef93d584511055f8e91e8d801662c44a5b0984d9 records the complete twelve-bit inverse-transform arithmetic checkpoint. ADST4 retains pinned libaom's signed 32-bit sine products and factorized sums and widens only its terminal rounding; Identity4 and Identity16 widen their fixed-point product and rounding bias only for the 20-bit twelve-bit row stage. The established 8/10-bit and twelve-bit column paths remain unchanged. Exact boundary vectors cover both 128-bit and 256-bit operators through normal, AVX-512-disabled, AVX-disabled, and scalar FeatureTestRunner configurations. Both source targets build with zero warnings and errors, the net10.0 test-project analyzer build completes with zero errors and 1,014 pre-existing repository warnings, Roslynk reports zero compiler errors, 510 focused inverse-transform cases and 44 production reconstruction cases pass without failures or skips, and git diff --check is clean.

Commit 5ba048be3667064f219ee230be7c8976927ea200 records the complete, verified official syntax-coverage and predictor-architecture checkpoint. SharedInfrastructure commit a835a9d74e82b2d32b580a7902eb2699ebc47098 is pushed to origin/main; the ImageSharp gitlink and synchronized root attributes assign IVF and Y4M fixtures to Git LFS. The official all-intra, CDF-update, and temporal motion-field IVF files have SHA-1 A9F7EA6312A533CC6426A6145EDD190D45813C37, AFCA5502A489692B0A3C120370B0F43B8FC572A1, and B48A717C7C003B8DD23C3C2CAED1AC673380FDB3, exactly matching the pinned libaom manifest. Their SHA-256 values are 5FCD265FD9F9BDD0D3179340B4C4532F1422CA5E5D97741C7481B84CB5DC122F, 14A3DBF537B6BF15EFC003182D9916D61438C93624A8BD26E6E3AE7EAF33EA82, and B59BF9586D8546DFDA81DFEC4EE4E32CEB502C9D22412AB0B63A2ABB534A1F14; the pinned-libaom Y4M references have SHA-256 1211EBEFBC9CCEF9ED19BE4CCE3F807D69FFFE338E95CCA1B5F4CA8023482175, 4FBFF73FF0DE2D9084DAE557D1D4BD677B0486516525BF4D327D2D795D5A7779, and F7DB607694818C19E62FD9A27F53E1A3E2D00B72C39C0430C1B26399CC76777D. Exact native-plane comparison covers all 39 all-intra frames, every intra mode, seven selected transform types, both tile-local and frame-end adaptive CDF publication, all four temporal motion-field frames under normal/scalar dispatch, and constrained tracked motion-field allocation with exactly-once returns. Every distinct AV1 predictor traversal now owns a family-named JPEG-style static operator contract and SIMD traversal instead of nesting separate predictors beneath broad intra/inter families. Compound reference convolution, equal averaging, distance weighting, alpha-mask blending, difference-weighted mask construction, and each intermediate reconstruction or mask traversal have separate family owners and matching .Operator.cs contracts; inter-intra mask construction has its own non-operator builder. Filtered intra-block-copy horizontal, vertical, and bilinear phases own separate family-named types, tap-exact static operator contracts, and family-owned SIMD/scalar traversals; the broad dispatcher retains only whole-sample copy and phase routing. Both Release source targets build with zero warnings and errors, Roslynk reports zero compiler errors, 102 final-tree focused predictor and reconstruction cases pass without failures or skips, and git diff --check is clean.

Commit ea2fa484716944b86fadac5eac3eef23874ac25b records the complete official extreme-displacement intra-block-copy checkpoint. The two-frame 1920x1080 YUV420 IVF has SHA-1 C7F336958E7AF6162C20DDC84D67C7DFA9826910, exactly matching the pinned libaom manifest, and SHA-256 0252716FE7A7C7F1EBB1EB23920A6737F45C10DFBA942304BD339AE42B5C576C; its pinned-libaom Y4M has SHA-256 8CF7851E112172C62799F8BFDF895505742A3CA77E339C12E547AD2C47AF41B3. The source corrects the spatial scan's four-mode-info-unit step threshold from 16-sample blocks to libaom's 64-sample boundary, clamps the ranked displacement stack before reference selection, and treats intra-block-copy neighbors as inter for transform-size context. Exact production native-plane comparison covers both complete official frames under normal and scalar FeatureTestRunner dispatch, while the existing independent 8/10/12-bit native-plane and presentation matrix remains clean. Both Release source targets build with zero warnings and errors, Roslynk reports zero compiler errors, 170 focused intra-block-copy, frame-buffer, reference-MV, inter-mode, and transform-size tests pass without failures or skips, and git diff --check is clean.

Commit 27ebcd54dc732aff8c665ee1ab61e72ed9621b42 records the official two-spatial-layer L2T1 default-operating-point checkpoint. The manifest-matching 159,037-byte IVF and pinned-libaom Y4M prove all eight 1280x720 YUV420 frames exactly under normal and scalar dispatch; constrained tracked allocation verifies balanced exactly-once returns. The existing decoder required no production correction. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, all five focused compact-sequence cases pass without failures or skips, and git diff --check is clean.

Commit a5e4cc6cc6cf87ee02b41cfe4ca71d1ef3411ed7 records the official one-spatial-layer, two-temporal-layer L1T2 checkpoint. The manifest-matching 48,883-byte IVF and pinned-libaom Y4M prove all eight 640x360 YUV420 frames exactly under normal and scalar dispatch; constrained tracked allocation verifies balanced exactly-once returns. The existing decoder required no production correction. The scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, both focused cases pass without failures or skips, and git diff --check is clean.

Commit 03737411e62155df8e998c37475ceeebf4f053bf records the official two-spatial-layer, two-temporal-layer L2T2 checkpoint. The manifest-matching 119,907-byte IVF and pinned-libaom Y4M prove all eight 1280x720 YUV420 frames exactly under normal and scalar dispatch; constrained tracked allocation verifies balanced exactly-once returns. The existing decoder required no production correction. The scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, both focused cases pass without failures or skips, and git diff --check is clean.

Commit 15613b01c0bc098fd9926b886216558e0d26eff9 records the official ten-frame active-film-grain checkpoint. The manifest-matching 28,634-byte IVF and pinned-libaom grained Y4M prove every displayed 352x288 YUV420 frame exactly under normal and scalar dispatch; the differing skip-grain digest and explicit coverage prove synthesis is active. The checkpoint corrected the AV1 default loop-filter reference-delta order so Golden is -1 and Backward is 0, restoring exact dependent-frame deblocking before display-only grain. Constrained tracked allocation returns every allocation exactly once. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, 28 focused cases pass without failures or skips, and git diff --check is clean.

Commit da9a46b3d72fd682ea9293af3f88fde26defb222 records the official eight-bit monochrome checkpoint. The manifest-matching 10,000-byte IVF and pinned-libaom raw luma reference prove all ten displayed 320x180 YUV400 frames exactly under normal and scalar dispatch. The shared compact-sequence oracle validates native monochrome without manufacturing chroma and keeps every prior YUV420 fixture exact. Constrained tracked allocation returns every allocation exactly once. The existing decoder required no production correction. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, all 17 official-sequence cases pass without failures or skips, and git diff --check is clean.

Commit e6f0ef84fc8273a7d50fdfc1489f54f058a2a424 records the OBU framing verification checkpoint. The bounded parser was re-audited against pinned libaom across length fields, operating points, header ordering, padding, alignment, tile-group boundaries, tile sizes, and ordered coverage. Corrected test oracles now address the actual item-relative tile bytes and avoid reflection invocation of span-valued properties. Both Release source builds remain clean, the scoped analyzer build and Roslynk report zero errors, all 119 focused cases pass, and git diff --check is clean.

Commit fc9320262cc08dd5daf41f231998403b804b98da closes the tile, mode, segmentation, delta-quantization, transform, coefficient, and inverse-quantization checkpoint. The production traversal and entropy order were re-audited against pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f. The stale coefficient round-trip oracle now compares the parser's retained entropy-scan order rather than raster order and disposes its encoded buffer owner. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, all 947 focused cases pass without failures or skips, documentation review is complete, and git diff --check is clean.

Commit 42384454eb45b6107608ed33379c7a69b101d533 closes the AV1 intra-prediction checkpoint. Pinned-source review confirms conventional mode and angle syntax, palette and filter-intra eligibility and ordering, chroma-from-luma eligibility and alpha mapping, reference availability, edge preparation, and predictor dispatch. Every predictor family retains the JPEG-style static operator architecture. All 93 focused cases pass without failures or skips, including the exact 39-frame official all-intra native sequence, exact palette native and presentation fixtures, and the exact 12-profile matrix. The existing production implementation required no correction.

Commit 6b3a5a5e1849de7881b929f46a5744fa3efe3fd6 closes the AV1 inter-frame prediction parent. Every completed reference-state, syntax, motion-derivation, translational, compound, inter-intra, OBMC, scaled, local-warp, global-warp, layered-presentation, film-grain, and monochrome child remains connected to the production path. All 342 focused inter-frame cases pass without failures or skips, including exact independent native-plane and presentation comparisons, normal and scalar dispatch, constrained allocation, and exactly-once ownership evidence. The existing production implementation required no correction.

Status meanings:

• Complete: the implementation and its phase exit evidence are recorded.
• In progress: usable implementation exists, but one or more required behaviors or verification gates remain open.
• Implemented locally; unverified: source exists in the working tree, but its checkbox remains open until the required Release and independent evidence pass.
• Not started: supporting primitives may exist, but the production format path is absent.
• Current: the only work item that should be advanced before taking the next queued item.

Current development stage: Stage 4 — complete decoded presentation and animation. The AV1 and independently coded HEVC still-image exit gates are complete across their exact profile, tool, constrained-allocation, SIMD-fallback, native-plane, and presentation matrices. The next dependency is the remaining shared color, ICC, alpha, grid, transform, metadata, and reference-dependent bounded-sequence decode matrix. Neither AV1 nor HEVC production encoding is implemented.

Immediate checkpoint: current — close shared decoded presentation and bounded animation. Complete the remaining color-signaling and ICC matrix, alpha, grid, transform, metadata, HEVC sequence, and reference-dependent AV1/HEVC presentation gates before starting either payload encoder.

Order	Delivery stage	State	Delivered state	Gate that remains open
1	Baseline, provenance, documentation, and public contract	In progress	Pinned codec references, a bounded image-only scope, encoder options, typed bit depth, decoder-option propagation, and extensive HEIF documentation exist.	Complete the all-file documentation audit, record a fresh Release baseline, finish distinct public HEIC/AVIF save boundaries, and close API review.
2	Bounded HEIF item and image-sequence container	In progress	Still-item parsing, grids, auxiliary alpha, metadata properties, bounded image-sequence tracks, Identify, and all-sync AV1 sequence presentation are connected.	Complete adversarial boundary coverage, remaining item/property behavior, reference-dependent sequence reconstruction, and the bounded sequence writer.
3	Still-image AV1 and HEVC decoding	Complete	AV1 still-image decoding has passed its complete exact native/presentation profile, layered-item, normative-tool, official-sequence, constrained-allocation, and SIMD-fallback exit gate. HEVC independently coded still-image decoding passes the complete Range Extensions profile/tool matrix with exact pinned-HM native planes, exact pinned presentation, constrained allocation, and scalar/SIMD evidence.	None for bounded still-image decoding.
4	Complete decoded presentation and animation	Current	Shared SIMD-first AV1/HEVC color conversion, ICC application, grids, transforms, direct planar alpha composition, frame metadata, repetition, and independently decodable AV1 sequence samples exist.	Close the full color/ICC cross-product, HEVC sequence decoding, AV1/HEVC reference-dependent samples, frame-local metadata/alpha behavior, and independent animated decode vectors.
5	AV1/AVIF encoding	Not started	RGB-to-planar conversion, forward transforms, OBU writer foundations, options, and container-writing infrastructure exist.	HeifEncoderCore still rejects AV1. Implement a real independently decodable lossy/lossless AV1 payload and the complete AVIF item/metadata matrix.
6	HEVC/HEIC encoding	Not started	Shared input color conversion, options, and HEIF writer infrastructure exist.	HeifEncoderCore still rejects HEVC. Implement a real independently decodable lossy/lossless HEVC payload and the complete HEIC item/metadata matrix.
7	Animated AVIF and HEIC encoding	Not started	Image-level repetition and per-frame duration contracts exist; bounded sequence parsing defines the required container surface.	Write the bounded sequence container and encode frame dependencies, timing, repetition, alpha, presentation, color, and metadata through both completed payload encoders.
8	Hardening and release readiness	Not started	Focused tests, independent fixtures, and representative SIMD benchmarks exist for completed slices.	Close the complete verification matrix, fuzz and malformed-input coverage, all Release/API/package checks, documentation and notices, and remove every placeholder or unsupported advertised path.

Immediate execution queue

• Finish the libaom-shaped AV1 forward-transform architecture, measured production dispatch, inverse-tier correction, suffix cleanup, FeatureTestRunner matrix, and focused Release verification recorded below.
• Close the base AV1 profile matrix with exact native-plane and presented-image comparisons for 8/10/12-bit monochrome, 4:2:0, 4:2:2, and 4:4:4 fixtures under normal, AVX2, 128-bit, and scalar dispatch.
• Accept the AV1-ISOBMFF final low-overhead OBU form that omits its payload-size field and uses the bounded image-item remainder; focused Release coverage reconstructs a valid combined frame in that form.
• Complete layered AV1 image-item dependency reconstruction and verify default final-layer output against the pinned libavif progressive fixtures.
  • Parse and associate a1op, lsel, and a1lx through the bounded image-item property model, including normative essential flags, duplicate handling, exact property lengths, and the four-layer limit.
  • Validate a1lx layer boundaries against the logical item size and restrict concrete lsel decoding to the cumulative payload through the selected spatial layer without copying item bytes.
  • Apply the selected a1op operating-point mask while consuming extended OBUs and validate the selected index against the parsed sequence header.
  • Store the eight fixed reference-validity, order-hint, and map-index tables inline on the frame header, retaining complete multi-bit order hints without per-header array allocations.
  • Enforce the bounded frame lifecycle across temporal delimiters, frame headers, tile coverage, and payload completion, publishing parser reference state only after successful decoder completion. The real two-layer libavif-derived lifecycle fixture verifies frame completion, retained-slot occupancy, reference resolution, and inherited dimensions through a fake tile lifecycle; it does not verify inter reconstruction or pixels.
  • Retain each successfully reconstructed reference as the sole owner of its ungrained sample planes, frame header, frame information, and published entropy snapshot; refresh only the selected slots, extend reference borders once after in-loop filtering, preserve a separate grained presentation buffer when required, and reset the store at an accepted sequence boundary.
  • Parse full and short reference signaling against retained-slot occupancy separately from frame-ID validity, resolve the primary-reference slot, implement frame_size_with_refs, and read high-precision-motion-vector, interpolation-filter, switchable-motion-mode, and reference-frame-motion-vector flags.
  • Initialize each frame's working CDF graph from the resolved primary snapshot, inherit segmentation feature data and an unchanged contiguous segmentation map, inherit loop-filter delta state, parse and inherit global-motion parameters, and initialize/project the retained per-8x8 temporal motion field.
  • Match libaom's bounded range-decoder consumed-bit accounting and tile trailing-bit validation so implicit zero refill cannot conceal truncated entropy data. Validate before publishing the selected CDF, reset decoder state after failure, and unwind every successful frame/block workspace rent when a later constructor allocation fails. A real truncated palette tile, decoder reuse, parser lifecycle, and allocator-identity tests cover these boundaries.
  • Consume the retained segmentation map through seg_id_predicted when temporal segmentation is enabled, including all three neighbor CDF contexts, pre-skip/post-skip ordering, skipped-block spatial inference, compatible retained-map geometry, and current-map updates.
  • Implement the complete inter-frame entropy, mode, motion-vector, compound-prediction, inter-prediction, and warped/global-motion paths permitted by the image profile.
    • Derive the frame-level skip-mode reference pair from mapped order hints, including modulo wraparound and the two-forward fallback, then decode the common inter-frame block prefix and intra-coded-block branch with retained CDF state and block-size luma contexts.
    • Implement allocation-free SIMD-first translational single-reference interpolation for regular, smooth, sharp, and bilinear filters across 8/10/12-bit samples. The predictor mirrors JPEG's closed static operator architecture, descends through Vector512, Vector256, and Vector128 before scalar fallback, and passes the exact independent convolution oracle through FeatureTestRunner.
    • Close the single-reference inter path before advancing to another mode.
      • Decode single-reference selection and own the exact adaptive reference and inter-mode distributions in the frame entropy context.
      • Derive the fixed-capacity spatial and temporal reference-MV stack with normative ordering, precision, clamping, global fallback, and DRL selection without per-block allocation.
      • Audit the disputed interpolation omission against pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f: av1_is_interp_needed() calls is_nontrans_global_motion(), whose global-motion-type test returns false only for TRANSLATION. An identity GLOBALMV block of sufficient size therefore omits switchable-filter symbols, matching the current != Translation predicate. The separate is_global_mv_block() > Translation test governs a different global-motion classification.
      • Audit the disputed spatial extension bound against the same pinned libaom: both loops in setup_ref_mv_list() stop at MAX_MV_REF_CANDIDATES, which is two. MAX_REF_MV_STACK_SIZE, which is eight, is the total stack capacity used by other candidate paths. The current spatial-loop bound matches the pinned source.
      • Decode NEW motion-vector differences with the independent normal-motion-vector entropy context and validate the final component range.
      • Decode or infer both directional interpolation filters after finalized motion-mode syntax, including fixed, switchable, dual-filter, and no-symbol paths.
      • Reconstruct one complete single-reference block through the retained padded plane and existing SIMD-first predictor before transform-unit residual traversal.
      • Consume inter-intra and motion-mode syntax in normative order when those modes are not selected, including the exact Block8x8 through Block32x32 inter-intra enum interval and binary/ternary motion-mode CDF selection.
      • Use counted frame/reference ownership for allocator-owned retained and temporal motion fields, with allocation tracking for initialization, retained-slot aliases, failure unwinding, presentation ownership, decoder-result ownership, and exactly-once final disposal.
      • Request the existing contiguous ImageSharp allocation contract for every padded AV1 frame plane. Constrained-allocator coverage verifies complete-plane block reconstruction without copying or per-block allocation.
      • Complete source review of the new motion-mode and single-reference tests.
      • Pass the exact single-reference checkpoint's Release verification: net10.0 and net11.0 source builds complete with zero warnings and errors; the net10.0 test-project analyzer build completes with zero errors and pre-existing repository warnings; 293 focused net10.0 entropy, candidate, motion, interpolation, lifecycle, reconstruction, ownership, and FeatureTestRunner cases pass with zero failures or skips; and git diff --check is clean.
      • Correct the progressive dependent-frame extraction and compare the final frame's first three native planes with pinned libaom output and its final RGBA presentation with pinned libavif exactly. The unmodified AVIF has the recorded SHA-256 and stores the primary color item's 55-byte base extent at offset 511 and 17-byte dependent extent at offset 583. The corrected logical .bit payload decodes as two YUV444 frames with pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f; the retained two-frame YUV444-alpha reference and final PNG come from pinned libavif linked to that build. The production test selects the second native frame, requires inter-coded blocks, and passes exact native and presentation comparisons through FeatureTestRunner.
    • Decode all compound reference-pair trees and inter modes, retain paired spatial/temporal reference-MV candidates through DRL selection, and reconstruct two translational predictors with allocation-free equal-weight averaging across 8/10/12-bit samples without changing the single-reference predictor contract or rounding model.
    • Verify the equal-average production branch with the pinned 19-frame libavif YUV444 sequence. The final native Y, U, and V planes match pinned libaom exactly, final RGBA presentation matches pinned libavif exactly, production DecodeBlock() is covered at 8/10/12 bit, constrained plane allocation passes, and allocator tracking returns every retained and temporal motion-field buffer exactly once.
    • Decode selected inter-intra prediction and selectable distance-weighted, wedge, and difference-weighted compound blending. Normative-order syntax, direct 8/10/12-bit production reconstruction, SIMD/scalar arithmetic, four pinned bounded sequences, exact final native planes and presentation, constrained contiguous planes, and balanced allocator returns pass. Both source targets build with zero warnings and errors; the test project builds with zero errors and pre-existing repository warnings; 25 direct focused cases and six FeatureTestRunner cases pass without failures or skips; and git diff --check is clean.
    • Decode and reconstruct above and left OBMC through allocator-owned prediction workspaces, exact pinned masks, and the existing SIMD-first masked blender. Commit 59a4fbad5ed3bc7c34331c4bdc65acc7acde3f04 records direct 8/10/12-bit production reconstruction, 4:2:0 and 4:2:2 overlap geometry, a real 19-frame YUV444 sequence containing more than one hundred traced OBMC blocks, exact final pinned-libaom native planes, exact pinned-libavif presentation through normal and scalar FeatureTestRunner configurations, constrained tracked allocation, zero-warning net10.0 and net11.0 source builds, a zero-error analyzer build with 1,013 pre-existing warnings, 14 direct focused cases plus the feature-isolated comparison, and clean staged git diff --check.
    • Decode and independently verify scaled-reference prediction through exact Q14/Q10 arithmetic, allocation-free SIMD/scalar convolution, 8/10/12-bit production reconstruction, scaled OBMC, active inter deblocking, constrained allocation, and a genuine size-changing bounded AVIF.
    • Decode local warped motion through pinned integer projection and allocation-free SIMD-first affine filtering. A genuine two-frame bounded AVIF selects traced multi-sample warped blocks and passes exact final native-plane and presentation comparisons, constrained allocation, and normal/scalar dispatch. Commit 0b2db86809e26757460e719a6b59c0fe58bb0586 records the checkpoint.
    • Decode non-translational global motion through the committed affine filter. A genuine two-frame bounded AVIF selects seven traced rotation/zoom GLOBALMV blocks and passes exact final native-plane and presentation comparisons, constrained allocation, and normal/scalar dispatch. Direct 8/10/12-bit production reconstruction covers both matrices of compound GLOBAL_GLOBALMV. Commit c5637ea0187df35b385bf43e2fe85cd955f01099 records the checkpoint.
    • Decode the official av1-1-b8-05-mv.ivf sequence through every ordinary inter mode, all three motion modes, and every regular/smooth/sharp dual-filter pair; reconstruct sub-8x8 chroma and allocation-free no-round compound intermediates; and compare all four shown 352x288 YUV420 frames exactly with pinned-libaom output under normal/scalar dispatch and constrained tracked allocation. The official IVF SHA-1 matches the pinned libaom manifest. Both Release source targets build with zero warnings and errors; the net10.0 test-project analyzer build completes with zero errors and 1,014 pre-existing repository warnings; Roslynk reports zero compiler errors; 3,983 focused decoder/predictor cases plus two complete official-sequence cases pass without failures or skips; and git diff --check is clean. Commit 1c58d855f70b024170ced9eb0a7005f0f9c955ad records the checkpoint.
    • Verify every connected inter mode and filter with exact independent content. The official pinned-libaom sequence supplies complete bit-mask coverage for all twelve ordinary inter modes, all three motion modes, and all nine regular/smooth/sharp dual-filter pairs. The committed average, selectable compound, inter-intra, OBMC, scaled-reference, local-warp, and global-warp AVIF fixtures separately require every connected reconstruction variant and pass exact native-plane comparison; the presentation-bearing fixtures also pass exact pinned-libavif output.
  • Return the explicitly selected spatial layer or the final displayed layer, keeping reference reconstruction separate from display-only film grain. The essential-lsel production fixture reconstructs the selected 40x40 YUV444 base exactly, scales native component planes to the 80x80 item extent with pinned-libyuv integer rounding, and matches pinned-libavif RGBA presentation under normal/scalar dispatch. The committed final-layer fixture and film-grain matrix remain exact. Constrained tracked allocation returns every short-lived presentation plane exactly once. Commit 019ac5648b4380a6ff4865072c96fde29ce09fad records the checkpoint.
  • Verify color and auxiliary-alpha output exactly against both pinned libavif progressive fixtures under normal SIMD dispatch and all required FeatureTestRunner fallbacks. Both fixtures pass exact public presentation under normal, AVX-512-disabled, AVX-disabled, and scalar dispatch; the alpha-bearing fixture additionally matches every composed alpha sample with its pinned native auxiliary plane. Both full production paths pass under a 1,024-byte constrained tracked allocator with exactly-once returns. Both Release source targets build with zero warnings and errors, the test-project analyzer build completes with zero errors and 1,014 pre-existing warnings, Roslynk reports zero compiler errors, and all 156 focused reconstruction and color cases pass without failures or skips.
• Correct the audited 12-bit inverse ADST4, Identity4, and Identity16 SIMD arithmetic by widening only the libaom-widened multiply/accumulate operations, with exact conformant-range vectors and FeatureTestRunner coverage. Both 128-bit and 256-bit operators match pinned outputs through every required hardware fallback; 510 focused inverse-transform cases, 44 production reconstruction cases, both zero-warning source builds, the zero-error analyzer build, zero Roslynk compiler errors, and clean git diff --check complete the verification.
• Verify the official libaom all-intra, CDF-update, and temporal motion-field sequences through exact native output, selected syntax coverage, normal/scalar dispatch, constrained tracked allocation, and the family-owned JPEG-style predictor operator architecture.
• Complete the AV1 still-image unsupported-branch inventory and extend the profile-matrix gate with exact independent compression-tool fixtures.
  • Add the missing official ten-bit sequence evidence from libaom's pinned test-data manifest. All ten active-film-grain YUV420 frames and all ten monochrome frames match the pinned generic libaom native Y4M samples exactly under normal and scalar FeatureTestRunner dispatch, and both complete sequences pass through a 2 KiB constrained tracked allocator with balanced exactly-once returns. The three new focused cases and all 20 official-sequence regression cases pass without failures or skips, the scoped Release test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and git diff --check is clean. The existing production decoder required no correction. Commit deefeba2340932909aa3e27516d965a294cf6faa records the checkpoint.
  • Add official minimum- and maximum-quantizer dependent-frame evidence for the eight- and ten-bit libaom matrices. The four retained streams and pinned generic-libaom Y4M references pass exact native comparison under normal and scalar FeatureTestRunner dispatch, and the combined sequences pass through a 2,560-byte row-aligned constrained tracked allocator with balanced exactly-once returns. All three new focused cases and all 23 official-sequence regression cases pass without failures or skips, the scoped Release test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and git diff --check is clean. The existing production decoder required no correction. Commit 68cbe2fb9483215db36977a06a2dcac04e2d90bc records the checkpoint.
  • Add exact official frame-size evidence at all four corners of libaom's 196-to-226-pixel matrix. Both frames from each retained stream match the pinned generic-libaom native Y4M samples under normal and scalar FeatureTestRunner dispatch, and all four sequences pass through a 1 KiB constrained tracked allocator with balanced exactly-once returns. Both new focused cases and all 25 official-sequence regression cases pass without failures or skips, the scoped Release test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and git diff --check is clean. The existing production decoder required no correction. Commit ccdcddc92e3f7c45eb0a964569a9c8d2756c0526 records the checkpoint.
• Verify HEVC CABAC arithmetic and every required context transition against pinned HM tables and complete production slices.
• Complete coding-tree, coding-unit, prediction-unit, and transform-unit traversal across every permitted size and partition mode. Eight official ITU first pictures compare exactly with pinned-HM native planes, pinned HM matches every archive complete-sequence MD5 and every RQT decoded-picture hash SEI, analyser output proves the recorded coding and transform sizes and partition alternatives, and all 150 focused HEVC decoder cases pass in Release. Commit 66d1f3c41b0baeaa98726b57d5feaeba50874f16 records the checkpoint.
• Complete HEVC intra prediction for every luma and chroma mode, including strong intra smoothing and constrained prediction rules. The official IPRED_B_Nokia_3 native output and independently coded first CIP_A_Panasonic_3 picture match pinned HM, all 149 focused HEVC decoder cases pass in Release, and commit 9bc5b9842 records the checkpoint.
• Correct and reverify the genuine-HEIC presentation matrix across every exposed HEVC profile and required Range Extensions tool, then close Phase 4 still-image decoding.
• Current: close shared decoded presentation, ICC, alpha, grid, transform, metadata, and animated AV1/HEVC decode gates.
• Implement and independently verify real AV1/AVIF still encoding.
• Implement and independently verify real HEVC/HEIC still encoding.
• Implement and independently verify animated AVIF and HEIC encoding through the bounded image-sequence surface.
• Complete the public API, documentation, provenance, hardening, performance, and release gates, then run the full Release matrix.

Performance, allocation, documentation, and independent test work are part of each delivery stage. They are not deferred cleanup phases, and the next queued stage must not begin while the current stage has unverified code or an open correctness gate.

Detailed implementation ledger

• Implemented and profile/tool presentation verified; complete color-signaling conformance remains: complete HEVC still-picture reconstruction and connect its native planes to the verified shared SIMD-first HEIF color pipeline.
  • Correct AV1 clipped-edge partition entropy handling and verify every block-mode cell from a real libavif AVIF against libaom 3.14.1.
  • Preserve the exact ICC payload from an independently encoded AVIF primary item.
  • Prove that a genuine non-sRGB AVIF profile changes decoded pixels and matches the independently converted source image within the documented AV1 tolerance.
  • Verify canonical-sRGB compaction and metadata skipping independently from ICC preservation and color conversion.
  • Replace the AV1 RGB/YUV arithmetic layer with JPEG's closed-generic static operator pattern in both directions: frame-scoped operator selection, in-place planar decode, RGB inputs with out component planes for encode, Vector512/Vector256/Vector128 traversal, one scalar tail, and JPEG's optimized byte-plane pack/unpack contracts for eight-bit pixels.
  • Keep codec row scheduling sequential and reuse frame-scoped allocator-backed buffers. Reuse JPEG's optimized UnpackIntoRgbPlanes contract for 8-bit encoding and pooled Rgb48 staging for 10/12-bit encoding so high-precision input is not truncated through an eight-bit or Vector4 intermediate.
  • Implement SIMD range expansion, chroma reconstruction, matrix/transfer conversion, clamping, and RGB packing for coefficient YCbCr, identity GBR, YCgCo, constant luminance, SMPTE ST 2085, and PQ/HLG ICtCp, with focused SIMD-lane/scalar-tail parity coverage.
  • Implement the H.273 V4 IPT-C2 and reversible YCgCo-Re/YCgCo-Ro transforms in both directions as closed SIMD operators, including their reduced RGB precision, signed integer lifting, full- and limited-range behavior, and 4:4:4 constraint.
  • Verify IPT-C2 against independently calculated matrix values and verify YCgCo-Re/YCgCo-Ro against exact 8/10/12-bit code values, limited-range black/white values, and SIMD-lane/scalar-tail parity.
  • Implement SIMD 4:2:0 and 4:2:2 encoder downsampling with odd-width and odd-height tails, and verify the stored 8/12-bit chroma samples against independently encoded full-resolution planes.
  • Move the H.273 parameter resolver, closed color operators, transfer functions, sample loading/storage, chroma filtering, and RGB packing into one shared HEIF color pipeline used by both AV1 and HEVC rather than maintaining codec-specific arithmetic copies.
  • Drive AV1 and HEVC through the same IHeifPlanarSampleBuffer<TSample> and HeifPlanarColorConverter architecture. Codec namespaces now adapt only native rows, sample precision, subsampling, crop windows, chroma positions, and codec color signaling; pooled row scheduling, chroma reconstruction/downsampling, fixed-point specialization, SIMD traversal, scalar tails, and RGB packing are selected by the shared HEIF driver.
  • Remove redundant internal member visibility from internal HEIF types. Existing assembly-visible methods, properties, accessors, and the cross-type constructor now use public member visibility, while type-local private members and JPEG-style internal nested operator contracts remain unchanged. Both Release source builds complete with zero warnings and zero errors, Roslynk reports zero compiler errors, and all 266 focused HEVC cases pass. Commit 6e906fed4 records the correction.
  • Implement sequential pooled HEVC conversion in both directions for monochrome, 4:2:0, 4:2:2, and 4:4:4; independent luma/chroma precision; 8/10/12-bit full/limited ranges; and all six progressive 4:2:0 chroma sample locations. Eight-bit paths use JPEG's optimized RGB plane contracts, while high-bit-depth paths retain 16-bit packed pixels.
  • Implement allocation-free SIMD-first HEVC intra prediction for all 35 modes, 4/8/16/32 blocks, and 8/10/12-bit samples, including three-tap and strong-bilinear reference filtering, negative-angle reference extension, luma boundary filters, and SIMD horizontal transposition. Verify the normal and forced-scalar paths against one scalar oracle.
    • Match HM's channel-specific smoothing rules: combined 4:4:4 chroma uses the normal three-tap filter when selected, while strong bilinear smoothing remains restricted to luma syntax. Exact official RExt reconstruction passes at 8, 10, and 12 bits.
  • Implement allocation-free reconstructed-plane reference collection and normative substitution for complete, empty, and partially available borders, including asymmetric 4:2:2 availability units and the HEVC mode/size/component filter-selection rules.
  • Implement allocation-free SIMD-first HEVC inverse quantization for flat and custom scaling lists, 4/8/16/32 transforms, transform-skip scaling rules, extended precision, signed normalization shifts, and transform-range clipping. Store compact and transform-sized scaling matrices in one contiguous parameter-set allocation so the hot path can widen consecutive weights without coordinate division.
  • Derive each reference-unit availability flag from coding-tree, slice, tile, constrained-intra, and picture-boundary state, then connect prediction to transform-unit traversal and the bounded item decoder lifecycle.
  • Register the bounded hvc1 still-image decoder, reconstruct the selected PPS/SPS picture, crop the conformance window directly in the pooled row converter, and pass the native planes through the shared SIMD-first H.273 pipeline.
  • Compare native HEVC planes byte-for-byte with HM 18 output for four item payloads extracted from independently produced HEIC files: two 8-bit 4:2:0 grid tiles, a cropped 8-bit 4:2:0 grid tile, and an 8-bit 4:4:4 wavefront tile.
  • Compare complete HEIC presentation output exactly with libheif 1.23.1 and libde265 1.1.1 for single images, multi-tile grids, an auxiliary alpha image, ICC-bearing images, 4:2:0, and 4:4:4 output. Keep the exact HM native-plane tests as the independent codec-reconstruction oracle.
  • Add permanent official HEVC 8/10/12-bit monochrome, 4:2:0, 4:2:2, and 4:4:4 fixtures with exact native-plane references and published per-plane digests. All 10 first pictures from the Sony GENERAL RExt conformance streams match every Y, Cb, and Cr sample with normal SIMD dispatch and with all hardware intrinsics disabled.
  • Extend the permanent HEVC matrix to every exposed profile and each Range Extensions tool not individually isolated by the Sony GENERAL streams, with exact native-plane references from HM or another independent decoder. Commit 9d7348efd6496b54cef7ec27e42352641439c400 records 25 additional official-picture cases, exact profile/tool signaling, exact native-plane hashes, the PCM precision and chroma-deblocking corrections, both zero-warning Release source builds, all 220 focused HEVC cases, verified existing LFS attributes and fixture hashes, zero Roslynk compiler errors, and clean git diff --check.
  • Retain 35 real HEIC containers and verify the complete independently coded profile/tool presentation matrix. Exact native-plane reconstruction remains pinned to HM; direct pinned-libheif presentation is used only where its HEVC backend reproduces those planes. Rejected or incorrectly reconstructed libheif profiles use pinned HM planes passed through an independent reproduction of libheif 1.23.1's float32 H.273 conversion, including nearest chroma traversal and source-precision clipping. FFmpeg raw output is accepted only as a byte-exact HM cross-check and its PNG conversion is not a reference. High-bit-depth conversion now retains source precision through Rgba64 instead of an eight-bit intermediate. The exact 25-case native matrix, 29 direct presentation cases, and three FeatureTestRunner presentation cases pass; the latter cover 8-bit subsampled, 12-bit full-resolution, and 12-bit Rgba64 output across the available SIMD widths and scalar fallback. Five constrained tracked-allocator cases retain balanced exactly-once returns. Both net10.0 and net11.0 Release source builds pass with zero warnings and errors. Commit 640b64ee5 records the original fixture matrix and the now-superseded weak comparison; the current local correction replaces that tolerance with exact pinned presentation evidence.
  • Implement the shared SIMD-first YUV/CICP paths in both directions for AV1 and HEVC bit depths, chroma formats, ranges, matrix coefficients, transfer functions, color primaries, and chroma positions, with scalar fallback only after the available vector widths are exhausted.
  • Complete the independently generated AV1 and HEVC conformance matrix for every supported color-signaling combination before closing the Phase 5 color exit gate.
  • Apply ICC conversion only after the SIMD YUV/CICP stage, alpha composition, grid assembly, and presentation transforms have produced the presented RGB image; retain ImageSharp's shared ICC converter and optimize reusable bulk kernels rather than creating a HEIF-specific color-management implementation.
  • Verify ICC preservation, conversion, compaction, and metadata skipping for grids, alpha-composited images, every presented sequence frame, and the completed HEVC path.
    • Independently encode AVIF primary, grid, alpha-auxiliary, and two-frame sequence vectors with libavif 1.4.2/libaom 3.14.1; verify exact profile preservation, non-sRGB conversion, canonical-sRGB compaction, non-sRGB compact preservation, metadata skipping, every presented frame, and unchanged composed alpha.
    • Repeat the complete ICC preservation, conversion, compaction, and skipping matrix for HEVC primary, grid, alpha-auxiliary, and sequence paths. Exact ICC-bearing still-image presentation is covered, but it does not close these separate color-management contracts.
  • Benchmark representative 8/10/12-bit end-to-end AVIF and HEIC color pipelines with and without ICC conversion, recording absolute timings and allocations before marking Phase 5 complete.
    • Add permanent frame-wide Av1ColorConversionBenchmarks coverage for sequential 1920x1080 YUV 4:2:0 conversion in both directions at 8, 10, and 12 bits, including managed-allocation reporting.
    • Record the .NET 10 short-run AV1 baseline on 2026-08-25. SIMD decode measured 4.025/2.803/2.742 ms and encode measured 3.750/2.555/2.450 ms at 8/10/12 bits with 488-744 B allocated per frame. Forced-scalar decode measured 31.15/29.47/28.40 ms and encode measured 19.19/15.77/15.14 ms, making the normal SIMD pipeline 7.7-10.5 times faster for decode and 5.1-6.2 times faster for encode on this machine.
    • Add permanent frame-wide HevcColorConversionBenchmarks coverage and record the .NET 10 short-run plane-conversion baseline on 2026-08-26. SIMD decode measured 1.141/2.640/2.668 ms and encode measured 2.089/2.481/2.412 ms at 8/10/12 bits with 504-712 B allocated per frame. Forced-scalar decode measured 13.42/28.35/27.60 ms and encode measured 14.82/14.67/14.45 ms, making the normal SIMD pipeline 10.3-11.8 times faster for decode and 5.9-7.1 times faster for encode on this machine.
    • Re-verify the unified AV1/HEVC driver with all 127 focused color and exact HEIC presentation cases under normal SIMD dispatch and with all hardware intrinsics disabled. Re-run all 12 permanent 1920x1080 plane-conversion benchmarks; BenchmarkDotNet could not hold the configured power plan and the CPU changed performance state during warmup, so do not replace the controlled baselines above with those throttled sustained means.
    • Add representative ICC-enabled AVIF cases and the equivalent HEVC cases after the HEVC reconstruction path is complete.
• Completed: remove production ToArray() materializations from the HEIF implementation and make every retained copy an explicit ownership decision.
  • Store AV1 palette colors and HEVC short-term reference-picture state inline at their normative fixed bounds.
  • Retain only the AV1 configuration sequence-header payload required for item/sample equivalence instead of every configuration OBU.
  • Allocate HEVC emulation-prevention positions once at the exact validated count without a growing list plus array copy.
  • Use ImageSharp's pooled ChunkedMemoryStream and direct chunk writes for the legacy JPEG item payload.
  • Keep exact owned arrays only where ICC, Exif, XMP, or retained property state must outlive pooled parser buffers.
• Completed: eliminate per-call scratch allocation across every AV1 forward and inverse 1-D transform, replace shared mutable transform scratch with operation-owned workspace, and complete the libaom-shaped SIMD-first transform paths with scalar fallback.
  • Inventory the existing hot-path scratch use: 15 forward/inverse transformer types contain 30 stackalloc int[...] sites.
  • Refactor every affected forward transform to reuse caller-owned workspace and its output buffer for staged ping-pong storage.
  • Refactor every affected inverse transform to reuse caller-owned workspace and its output buffer for staged ping-pong storage.
  • Remove the shared mutable Av1ForwardTransformer.TemporaryCoefficientsBuffer and give each concurrent transform operation exclusive workspace ownership.
  • Replace the instance-transformer interface with stateless forward and inverse operator structs following the JPEG color-transform static-generic operator pattern.
  • Select transform type, size, bit depth, and ISA once at the 2-D block boundary rather than dispatching through an interface for every row and column.
  • Port the DCT4/8/16/32/64, ADST4/8/16, and identity4/8/16/32 stage networks from the pinned libaom scalar and Highway sources into one static-generic operator architecture.
  • Implement paired add/subtract and whole-butterfly primitives for scalar, Vector128, Vector256, and Vector512 values, including saturated packed arithmetic and shared widening work at each supported SIMD width.
  • Implement the complete libaom two-dimensional load, flip, shift, axis-transform, transpose, rectangle-normalization, promotion, and 64-point coefficient-retention pipeline without per-block allocation. The axis driver passes independent input and output strides directly to the two fixed stage buffers, and the 64x64 row transform retires only the retained 32 coefficients without a transform-sized copy pass.
  • Port the applicable libaom bulk inverse-transform kernels using the same tables, rounding, saturation, and clipping rules as the scalar oracle.
  • Use normal ImageSharp descending-width dispatch and require the actual packed arithmetic ISA when selecting packed Vector512<short> traversal.
  • Document scratch ownership, stage-buffer alternation, fixed-point rounding, lane layout, transposition, and scalar fallback decisions at their implementation points.
  • Remove the separate SIMD files, width-specific forward contracts, and sample-representation suffixes so each transform operator owns one behavior model.
  • Verify every 1-D operator representation and every valid 2-D size/type/bit-depth combination through FeatureTestRunner with normal hardware, AVX-512 disabled, AVX disabled, and all hardware intrinsics disabled.
    • The focused Release run passes all 511 forward and inverse transform cases. The suite covers DCT, ADST, and identity operators, packed overflow-sensitive inputs, padded 2-D input strides, all valid transform configurations, 8/10/12-bit dispatch, inverse reconstruction, and the zero-allocation block contract.
  • Verify DCT4/8/16/32/64, ADST4/8/16, and identity4/8/16/32 independently against the analytical transform oracle and coefficient-error bound used by the pinned libaom forward-transform tests.
  • Benchmark the production 32x32 DCT path after the complete paired stage port with preferred 256-bit and 512-bit widths.
    • On the measured .NET 10 AVX-512 host, the packed 8-bit path measured 856.6 nanoseconds at 256 bits and 639.8 nanoseconds at 512 bits, making AVX-512BW 25.3% faster. The expanded 12-bit path measured 976.2 nanoseconds at 256 bits and 1019.8 nanoseconds at 512 bits, making the runtime-preferred 256-bit width 4.5% faster. BenchmarkDotNet reported no managed allocation. Production therefore follows Highway's AVX-512BW capability dispatch for packed stages and the runtime-preferred width for expanded stages.
• Queued: restore bounded animated HEIC and AVIF image-sequence scope, including the required image-level and per-frame metadata contracts, without introducing unrelated ISO BMFF surfaces.
  • Reconcile the top-level still-image-only scope with the required animated HEIC and AVIF completion boundary before sequence implementation begins.
  • Define the ImageSharp image-level sequence metadata and per-frame metadata contracts, including observable timing, repetition, color, alpha, orientation, and profile behavior.
    • HeifMetadata now carries repetition and root-frame animation behavior through FormatConnectingMetadata. HeifFrameMetadata carries frame duration through FormatConnectingFrameMetadata; frame-local color, alpha, orientation, and profile state remain owned by the existing ImageFrameMetadata contract. The focused Release suite passes all 40 encoder, image metadata, and frame metadata tests.
  • Identify and document the minimum normative ISO BMFF track, sample-description, sample-location, dependency, and timing syntax required by conforming HEIC and AVIF image sequences.
    • The bounded syntax matrix below is derived from ISO/IEC 23008-12:2022 clauses 7 through 10, the codec-specific HEVC Annex B, AVIF 1.2 section 3, AV1-ISOBMFF 1.2 section 2, and the pinned libavif read.c and write.c sequence paths.
  • Parse and write only that bounded image-sequence syntax; do not add audio, arbitrary media, editing, fragmentation, streaming, or general presentation APIs.
    • Share one bounded 32/64-bit box-header reader between the existing item path and the image-sequence path, including short-read handling and parent-boundary validation.
    • Select one enabled pict master track without materializing unrelated tracks, accepting conforming image-sequence tracks that omit track_in_movie, then parse its mvhd/tkhd/mdhd/hdlr, self-contained dref, av01 or hvc1 sample entry, codec configuration, mandatory ccst, and bounded repetition edit.
    • Resolve stsc, stco/co64, stsz/stz2, stts, and stss into one exact value-type descriptor array capped by DecoderOptions.MaxFrames; validate complete run/count syntax through one allocator-owned sequential scratch buffer.
      • Release verification passes the libavif-shaped two-sample parser fixture, the one-frame retention boundary, and a sample offset/length beyond the file. Pinned upstream libavif files independently verify 8-bit and 12-bit sequence identification, alpha-track and Exif/XMP retention, infinite and finite repetition, and safe omission of an unrelated audio track. The parser performs no per-entry allocation and does not buffer moov, mdat, unrelated track payloads, or complete attacker-sized tables.
    • Parse HEVC composition offsets and hidden samples from ctts/cslg, while rejecting the ctts box prohibited for AV1 tracks.
      • Release verification covers signed and unsigned composition-offset syntax, hidden-sample visibility, composition-time calculation, required cslg and edit-list signaling, complete run counts, and the AV1 prohibition without buffering either table.
    • Parse optional direct dependencies from refs sample groups into a compact retained-sample dependency graph.
      • The parser validates complete sbgp runs and sgpd entry lengths, versions, indices, positive identifier uniqueness, decode order, sync-sample independence, and ccst reference limits. It uses pooled value-type sort indexes plus one exact flat reference array, with no per-sample collections or attacker-sized table materialization.
    • Parse track presentation, color, and HDR properties through the same typed property parser used by still-image items.
      • Selected visual sample entries now retain ICC and CICP color profiles, pasp, clap, irot, imir, clli, mdcv, cclv, amve, reve, and ndwt. Fixed-size properties use the parser's existing scratch buffer; only retained ICC bytes use bounded pooled staging before entering ImageSharp's existing profile model.
    • Parse bounded sequence metadata items.
      • Selected image tracks now retain implicitly associated Exif and unencoded application/rdf+xml XMP items from bounded iinf/infe and iloc declarations, including file-relative and idat-relative extents. Item identifiers and extent descriptors use allocator-owned temporary storage; only the exact retained profile payloads survive parsing. DecoderOptions.SkipMetadata bypasses metadata parsing, SegmentIntegrityHandling follows the shared ancillary/image-data policy, and structural errors that prevent safe track parsing remain fatal. The focused Release sequence-parser suite passes all 15 tests.
    • Connect the parsed sequence index to HEIF detection, Identify, frame decode, alpha matching, and frame metadata without changing still-image source selection.
      • Recognize supported avis, hevc, and hevx sequence brands while continuing to reject layered HEVC and JPEG sequence brands.
      • Identify bounded sequence dimensions, frame count, timing, repetition, codec precision, color, HDR, pixel aspect ratio, Exif, and XMP state.
      • Decode all-sync independently decodable AV1 samples into directly adopted ImageSharp frames without cloning complete pixel buffers.
      • Match auxiliary alpha samples by exact decode duration, visibility, and presentation time, and validate premultiplication track identity.
      • Require unity movie and track matrices so image presentation remains on the optimized clap/irot/imir path without a movie compositor.
      • Apply the shared DecoderOptions contract consistently to still items, nested payload codecs, grids, metadata properties, and sequence samples.
        • Strict rejects recoverable ancillary and image-data errors, IgnoreAncillary suppresses only ancillary failures, and IgnoreImageData additionally permits failed image properties or samples to be omitted. SkipMetadata avoids optional property and item-payload validation, while cancellation and the caller configuration flow into nested JPEG and AV1 decoders. Target scaling and ICC conversion remain presentation-level operations after item or grid composition. The focused Release matrix passes all 19 new still-image policy cases, all 3 new sequence-sample cases, and the complete 37-test sequence-parser suite; the Release test-project build completes with zero errors.
        • Apply the same recovery boundary to still-image item relationships, coded payloads, primary-item thumbnail fallback, and optional alpha composition.
          • Unknown item-reference types are skipped within their validated child boundaries. cdsc failures follow ancillary policy and are not parsed when metadata is skipped. dimg, auxl, prem, and thmb failures follow image-data policy. IgnoreImageData can omit an unreadable alpha plane or recover from a failed primary payload through a valid registered thumbnail, but decoding still fails when no color presentation remains. Real AVIF fixtures cover corrupt alpha payloads, corrupt alpha relationships, malformed Exif relationships, and strict, ancillary-only, image-data, and metadata-skipping behavior.
      • Complete reference-dependent AV1 and HEVC sample reconstruction and independent sequence vectors.
    • Write the same bounded movie, track, sample-description, location, dependency, timing, repetition, alpha, and metadata syntax from ImageSharp frames.
  • Decode frame dependencies, durations, repetition, frame-local auxiliary images, and frame-local metadata into the existing ImageSharp multi-frame model.
  • Encode ImageSharp frames, durations, repetition, frame-local auxiliary images, and frame-local metadata as independently decodable HEIC and AVIF image sequences.
  • Add malformed-sequence boundary coverage and independently verify animated inputs and outputs with pinned HEIC and AVIF implementations.
• Queued: complete the AV1 and HEVC encoding paths and remove the explicit HeifEncoderCore encoding stubs only after each emitted payload passes independent decode.
  • Complete every AV1 encoder task and exit gate in Phase 6; a smallest-valid fixed coding subset is only an intermediate milestone.
  • Complete every HEVC encoder task and exit gate in Phase 7; a smallest-valid fixed coding subset is only an intermediate milestone.
  • Connect both payload encoders to the bounded HEIF writer with the selected bit depth, chroma layout, range, color signaling, alpha, metadata, and animation state.
  • Replace each NotSupportedException branch only when the corresponding payload is accepted by the pinned independent decoder and the ImageSharp decoder.
  • Verify that every public quality, effort, lossless, bit-depth, chroma, alpha, and metadata option changes or constrains the encoded output exactly as documented.
• Completed: define and document the HEIF encoder option contract using the established ImageSharp encoder pattern.
  • Confirm that IHeifEncoderOptions has only one concrete implementation and remove the unnecessary interface.
  • Document the default, valid range, special values, invalid-value behavior, and format-dependent restrictions of every retained option using observable API behavior only.
  • Pass HeifEncoder directly to HeifEncoderCore, matching the JPEG, PNG, and WebP encoder-core contracts and avoiding interface dispatch.
  • Verify construction-time range validation and legacy-JPEG codec-boundary restrictions with focused tests before the Phase 1 API-review gate is marked complete.
• Completed: lock public HEIF image metadata to the supported component bit-depth contract.
  • Replace the unrestricted integer bit depth with HeifBitDepth and preserve the 8-bit default.
  • Resolve format-connecting component precision to the nearest supported 8/10/12-bit output without widening the public value domain.
  • Reject HEVC configuration records outside the exposed 8/10/12-bit profile matrix at the external parse boundary.
  • Verify defaults, cloning, format-connecting conversion, pixel-type projection, and current HEIC/HIF/AVIF Identify results.
    • Release build: 0 errors. Focused metadata and Identify tests: 42 passed, 0 failed.

Gain maps, progressive/layered images, sample transforms, and experimental extension brands require explicit conformance and API decisions. They do not create permission to omit any valid color, compression, or bit-depth path from the PR. The container reader must skip unsupported optional extensions safely and reject an unsupported essential property with a useful error.

Container scope

The container implementation is a deliberately narrow HEIF image-format reader and writer, not a general ISO BMFF framework. Implement only the box syntax and relationships required for supported HEIF image items and bounded HEIC/AVIF image sequences.

In scope are the file type, metadata, item location/data, item information, item properties, item references, primary-item selection, idat/mdat payload storage, grids, auxiliary alpha, presentation transforms, color properties, and Exif/XMP paths required by still images. Image sequences additionally include only the brands, tracks, sample descriptions, sample locations, decode dependencies, timing, repetition, and frame-local metadata required to map HEIC and AVIF sequences to ImageSharp frames.

Still files produce one presented primary image. Supported image sequences produce one ordered ImageSharp frame collection with bounded duration, repetition, frame dependency, alpha, color, orientation, and profile state. Supporting items are decoded only when required to construct or describe those presented frames. The implementation does not expose arbitrary HEIF image collections, bursts, pages, audio, or non-image media.

Out of scope are audio, arbitrary video, edit lists, fragments, streaming profiles, general presentation APIs, and reusable movie, track, or sample-table object models. Track, sample, timing, and inter-frame codec state may exist only inside the bounded HEIC/AVIF sequence decoder or encoder that owns it. Unknown optional boxes remain bounded and skippable; unsupported essential image or sequence syntax must fail with a useful image-format error.

Implementation rule: do not introduce a reusable general-purpose ISO BMFF box hierarchy, track model, or media parser. Add box syntax directly to the bounded HEIF model only when a supported still image, image-sequence frame, relationship, property, metadata path, or conformance fixture requires it. Each addition must name the image behavior it enables and have a focused image-format test.

An ISO BMFF construct may be added only when all of the following are true:

1. A conforming supported still image or HEIC/AVIF image sequence requires it to produce or describe presented ImageSharp frames.
2. Its owning image item and its effect on the decoded or encoded image are explicit.
3. It can be parsed or written as bounded state owned by the HEIF image-item or image-sequence path without adding a general media abstraction.
4. Independent still-image or animated-image fixtures exercise the behavior it enables.

Encountering a box in libavif, ISO BMFF, or a third-party file is not by itself a reason to port it. Constructs not required by the supported still-image or bounded image-sequence presentation must be skipped when optional or rejected when essential.

Codec-configuration rule: parse av1C and hvcC as properties of coded image items and as the bounded decoder configurations referenced by supported image-sequence sample descriptions. Validate profile, level, bit depth, chroma layout, and parameter-set/OBU declarations against the associated item or sequence samples. Do not expose visual sample entries, sample descriptions, layer selection, sample groups, or timing as general public or reusable ISO BMFF models.

AV1 sequence headers and HEVC VPS/SPS/PPS structures remain in scope because they are codec syntax required by image items and image-sequence frames. Retained reference-frame state is permitted only for the lifetime of a bounded HEIC/AVIF sequence decode or encode operation.

Bounded image-sequence syntax

The sequence reader and writer may retain only the following syntax and the resolved image behavior listed beside it:

Syntax	Required image behavior
ftyp sequence and structural brands	Recognize avis AV1 sequences and the non-layered hevc/hevx HEVC sequence profiles. avio can additionally signal an all-sync AV1 sequence. Layered hevm/hevs, JPEG jpgs, arbitrary video brands, and brands for unimplemented codecs remain unsupported until their image payload and presentation requirements are implemented.
moov/mvhd, trak/tkhd, and mdia/mdhd/hdlr	Select an enabled pict master image-sequence track without requiring the optional track_in_movie flag; retain its displayed dimensions, media time scale, media duration, and movie-time-scale track duration. Require unity movie and track matrices because arbitrary movie-canvas composition is outside image-format scope; use clap/irot/imir for supported image presentation. Inspect unrelated tracks only far enough to identify and skip their non-image handlers.
minf/dinf/dref and stbl	Accept only self-contained sample data references and own the bounded sample-table state for one selected image sequence plus an optional linked auxiliary-alpha sequence. No reusable data-reference, media-information, or sample-table API is created.
stsd and one supported visual sample entry	Require exactly one av01 entry for AVIF or one non-layered hvc1 entry for HEIC. Retain only its dimensions, codec configuration (av1C or hvcC), image presentation/color/HDR properties, and mandatory version-zero ccst coding constraints. Reject an unsupported essential configuration rather than treating it as generic video.
stsc, stco/co64, and stsz/stz2	Resolve each declared image sample directly to a validated file offset and length. Expand run tables once into a compact frame-owned descriptor array bounded by DecoderOptions.MaxFrames; never buffer the movie or complete mdat.
stts	Expand decode-order sample durations against the media time scale into exact per-frame rational delays. A sequence with advisory timing still maps every visible sample to an ImageSharp frame.
optional ctts version 0/1 and cslg	Retain only composition offsets and the HEIF hidden-sample condition required to determine presented frame order and visibility. No general composition timeline is exposed.
optional stss	Mark the one-based random-access samples; absence means every sample is a sync sample. The first retained frame must have a valid decode path from a sync sample.
optional sgpd/sbgp with grouping type refs	Resolve direct-reference sample identifiers for bounded dependency validation and random access. Other sample-group types are ignored unless a later image-format requirement explicitly brings them into scope.
optional edts/elst	Interpret only normal-rate image-sequence presentation and the HEIF repetition flag. Reject edits that require arbitrary media splicing, rate changes, or a general edit timeline.
optional tref entries auxl, prem, thmb, and cdsc	Link only image alpha, premultiplication, image thumbnails, and image metadata. Auxiliary samples are matched to color frames by their decoded time; audio and every unrelated reference type are ignored.
optional track-level meta and image metadata linkage	Apply sequence-wide Exif/XMP profiles from the bounded track metadata model. Per-frame metadata is limited to supported metadata items/groups or codec metadata that can be mapped to ImageFrameMetadata; generic timed-metadata tracks are not exposed.

The implementation must validate uniqueness, versions, entry counts, run totals, one-based indices, integer products, cumulative durations, offsets, and lengths before creating retained arrays. Temporary box payloads use the configured MemoryAllocator; persistent sample descriptors are allocated once at the validated retained-frame count. Parsing must skip unrelated boxes by their bounded length and must not copy an entire moov or mdat payload.

Reference hierarchy

Use the references in this order when behavior differs:

1. The published ISO BMFF, HEIF, HEVC, AV1, AV1-ISOBMFF, and AVIF requirements are normative.
2. The official AOM AV1 Codec Library is the primary implementation reference for AV1 decode, encode, high-bit-depth behavior, tests, and optimized scalar/SIMD algorithms. Pin one reviewed commit before porting. Use its architecture-specific paths to shape the managed SIMD-first production architecture and its scalar C paths to define the equivalent fallback behavior.
3. The local D:\GitHub\AOMediaCodec\libavif checkout is the AVIF container, metadata, color-conversion, grid, alpha, and interoperability oracle. At inspection time it identifies itself as 1.4.2-devel.
4. The official HM reference software at commit 9c1f298659ab0cee9dc13d23d0304221575410b9 is the primary scalar implementation reference for HEVC syntax, reconstruction, and the complete Range Extensions profile matrix. Android libhevc v1.6.0 at commit c83a76b084498d55f252f48b2e3786804cdf24b7 is a secondary production-C comparison for its supported common 8-bit 4:2:0 paths; its decoder rejects other chroma formats and bit depths and therefore is not the feature-coverage authority. The HEVC specification remains normative, and an independent HEIC container implementation is still required as an interoperability oracle. Do not copy from GPL or otherwise incompatible sources.
5. Existing ImageSharp codecs are the authority for ImageSharp API shape, memory ownership, stream behavior, cancellation, resource limits, pixel conversion, tests, and SIMD dispatch.

The linked ImageSharp discussion establishes the project constraint: the shipped implementation is purely managed and other codec libraries are references, not native runtime dependencies. libaom is the official encoder/decoder implementation reference for AV1, but it does not parse the HEIF container or implement HEVC. libavif dispatches AV1 work to external codec libraries, so it remains an observable AVIF/container oracle rather than the source for every AV1 algorithm.

The AOM source is distributed under the BSD 2-Clause License and the Alliance for Open Media Patent License 1.0. HM is distributed under its BSD license with an explicit patent-rights disclaimer, while the reviewed Android libhevc source is distributed under Apache-2.0. Before porting further code, record the exact upstream file, commit, applicable license/patent notice, and corresponding managed file or method. Audit the existing SVT-AV1-attributed WIP separately rather than relabeling it as libaom-derived. Update THIRD-PARTY-NOTICES.TXT before any referenced implementation code is merged. A pure managed HEVC implementation does not remove HEVC patent or licensing obligations, so those must be resolved before the HEIC work is considered releasable.

Pinned reference and baseline snapshot

The initial post-merge snapshot was established on 2026-08-24:

• the official libaom reference is tag v3.14.1, commit 03087864cf4bea6abb0d28f95cf7843511413d8f, matching the revision selected by the local libavif ext/aom.cmd dependency script;
• the local libavif container, color-conversion, and interoperability oracle is commit 092276ce89098ead06db80975173191e5fee1826, described as v1.4.2-66-g092276ce;
• the official HM reference software is commit 9c1f298659ab0cee9dc13d23d0304221575410b9; its decoder and profile tables cover the HEVC Range Extensions chroma, bit-depth, intra, and still-picture profiles required for the final HEIC feature matrix;
• the official Android libhevc reference is v1.6.0 at commit c83a76b084498d55f252f48b2e3786804cdf24b7; it remains a secondary common-path comparison because its SPS parser accepts only 8-bit YUV 4:2:0;
• the official ITU-T H.274 (V4) recommendation from January 2026 is the normative semantics reference for mastering-display and content color-volume fields reused by still-image item properties; its video-SEI persistence and cancellation behavior is outside this container scope;
• an independent HEIC container interoperability oracle and the complete external HEIC fixture-generation matrix remain to be pinned before HEIC support can pass its final integration gate;
• dotnet build ImageSharp.sln -c Release --no-restore -m:1 -v minimal succeeds with no errors after the upstream compatibility fixes; and
• the existing HEIF-focused test run executes 8,198 cases, with 8,184 passing and 14 failing. Thirteen failures are isolated to the WIP AV1 YUV conversion tests, and one is the existing legacy JPEG HIF reference-image mismatch. Golden artifacts have not been changed.

This snapshot pins or classifies the available references and failures; it does not complete Phase 0. The full WIP provenance map, disabled-test inventory, HEIC container interoperability selection, and feature-state matrix remain required.

Provenance map in progress

Managed implementation	Normative behavior	Reviewed implementation reference	Use
HeifColorConversionParameters, HeifColorConverter, IHeifPlanarSampleBuffer, HeifPlanarColorConverter, Av1YuvConverter, and HevcYuvConverter	ITU-T H.273 (V4) equations 14-16 and 27-84, including limited/full-range scaling, chromaticity-derived equations 39-47, YCgCo equations 51-57, constant-luminance equations 66-75, IPT-C2, YCgCo-Re/YCgCo-Ro, and the PQ/HLG ICtCp matrices; AV1 section 6.4.2 and HEVC Annex E chroma sample positions	Official ITU-T H.273 (V4) (07/2024); libavif src/reformat.c and src/colr.c at 092276ce89098ead06db80975173191e5fee1826; libaom aom/aom_image.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	SIMD-first closed operators provide full/limited-range conversion at 8, 10, and 12 bits, with an equivalent scalar fallback and tail. Decode and encode cover every H.273 V4 matrix code point through 17: identity, YCgCo, coefficient-based NCL, both fixed and chromaticity-derived constant/non-constant-luminance systems, SMPTE ST 2085, ICtCp, IPT-C2, YCgCo-Re, and YCgCo-Ro. One closed generic planar driver owns pooled sequential row traversal, chroma reconstruction/downsampling, fixed-point specialization, SIMD dispatch, scalar tails, and packed-pixel conversion for both AV1 and HEVC adapters. Shared paths cover monochrome, YUV 4:2:0, 4:2:2, and 4:4:4; AV1 centered/co-sited placement; and all six HEVC progressive 4:2:0 sample locations. Identity and reversible lifting enforce their full-chroma constraint. Limited-range YCgCo retains the 219-code scale inherited from its R/G/B inputs instead of applying YCbCr's unrelated 224-code chroma range. YCgCo-Re and YCgCo-Ro apply range adjustment at reduced-precision RGB code values before signed integer lifting. Chromaticity derivation uses every defined H.273 primary and matches libavif's BT.709 fallback for unspecified or reserved primaries. ICtCp and IPT-C2 inverses come from the exact H.273 matrices rather than unrelated display conversions. Closed operators provide Vector512, Vector256, and Vector128 implementations with one scalar tail. Eight-bit conversion uses JPEG's RGB plane pack/unpack contracts; high-bit-depth conversion retains native 16-bit packed pixels.
HeifTransferFunctions	ITU-T H.273 (V4) Table 3 transfer characteristics 1-18	Official ITU-T H.273 (V4) (07/2024); libavif src/colr.c at 092276ce89098ead06db80975173191e5fee1826	Apply every HEVC/AV1-signallable transfer function required by constant-luminance, ICtCp, and IPT-C2 conversion. Retain the H.273 normalized PQ and HLG definitions; do not import libavif's display-oriented 203-nit scaling or HLG OOTF into codec sample interpretation. Use libavif's midpoint convention only for the non-bijective zero code of the two logarithmic curves.
ObuReader.ReadSequenceHeader, ReadUncompressedFrameHeader, decoder-model parsing, and operating-parameter consumption	AV1 sections 5.5.2 through 5.5.4 sequence timing and decoder-model syntax, section 5.9.2 uncompressed frame-header syntax, and section 5.9.31 temporal-point syntax	libaom av1/decoder/decodeframe.c functions av1_read_decoder_model_info, av1_read_op_parameters_info, read_temporal_point_info, and read_uncompressed_header, plus common/av1_config.c, at 03087864cf4bea6abb0d28f95cf7843511413d8f	Read the normative 32-bit decoding-tick field; consume operating-point buffer delays and the layer-applicable frame removal times needed to keep a non-reduced still-image sequence aligned; and read presentation time only under the normative decoder-model condition. Retain no scheduling or playback behavior from those values and introduce no ISO BMFF timing, track, sample-table, or sequence surface.
Av1FrameBuffer high-bit-depth sample layout and Av1YuvConverter 10/12-bit packed-pixel conversion	AV1 section 6.4.1 bit depth and H.273 sample-range scaling	libaom aom_scale/yv12config.h, av1/common/idct.c, and av1/common/reconintra.c at 03087864cf4bea6abb0d28f95cf7843511413d8f; libavif src/avif.c and src/reformat.c at 092276ce89098ead06db80975173191e5fee1826	Establish two-byte native sample storage with sample-unit strides for 10/12-bit reconstruction and use ImageSharp's existing Rgb48 pixel-operation paths in both directions so packed-pixel staging does not reduce high-bit-depth samples to eight bits.
Av1PredictionDecoder, Av1IntraPredictor, Av1ChromaFromLumaContext, Av1ChromaFromLumaPredictor, Av1PartitionInfo, and the DC, directional, Paeth, smooth, filter-intra, and chroma-from-luma predictor operators	AV1 sections 7.11.2 and 7.11.2.3 intra prediction	libaom aom_dsp/intrapred.c, av1/common/reconintra.c, av1/common/av1_common_int.h, av1/common/blockd.h, av1/common/cfl.c, and av1/common/cfl.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Design the shared predictor contract, sample layout, reference preparation, and scratch ownership for widest-to-narrowest SIMD traversal, then provide the equivalent scalar fallback through that contract. Preserve luma/chroma mode-neighbor addressing, directional upsampling, Paeth selection, smooth normalization, filter-intra taps, high-bit-depth clipping, chroma-from-luma storage/subsampling, and chroma-from-luma row strides. The WIP rectangular byte-pipeline smooth digest expectations encode width/height-swapped weights and must be replaced only from an independently generated oracle, not regenerated from this implementation.
Av1ReferenceScale, Av1InterPredictor scaled operators, Av1BlockDecoder scaled-reference traversal, and Av1FrameBuffer prediction border	AV1 sections 7.11.3.3 reference scaling and 7.11.3.4 block inter prediction	libaom av1/common/scale.c, av1/common/scale.h, av1/common/reconinter.h, av1/decoder/decodeframe.c, av1/common/convolve.c, av1/common/filter.h, and aom_scale/yv12config.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Preserve Q14 reference scale factors, Q10 variable phases and steps, signed coordinate rounding, scaled UMV margins, the 288-sample reference border, reduced filter selection, bit-depth-dependent two-pass rounding, and byte/high-bit-depth clipping. A four-output Vector128 gather-shaped traversal and exact scalar tail share one behavior model and caller-owned scratch. The production path covers ordinary, compound, and OBMC prediction without per-block allocation.
Av1TileReader palette mode/color-map parsing, Av1SymbolDecoder palette distributions, Av1BlockModeInfo palette state, and Av1PredictionDecoder palette reconstruction	AV1 sections 5.11.46, 5.11.49, and 7.11.2 palette prediction	libaom av1/decoder/decodemv.c, av1/decoder/detokenize.c, av1/decoder/decoder.h, av1/common/pred_common.c, av1/common/pred_common.h, and av1/common/entropymode.c at 03087864cf4bea6abb0d28f95cf7843511413d8f	Reuse the existing tile range decoder and frame-owned mode map while porting the normative palette cache merge, high-bit-depth color deltas, tile-adaptive mode/size/index distributions, diagonal color-map traversal, edge padding, and direct palette-sample reconstruction. This is AV1 still-image compression syntax and does not add retained video reference state or any ISO BMFF surface.
Av1DeblockingFilter, Av1LoopFilterContext, and Av1LoopFilterDecoder	AV1 section 7.14 deblocking loop filter	libaom aom_dsp/loopfilter.c and av1/common/av1_loopfilter.c at 03087864cf4bea6abb0d28f95cf7843511413d8f	Preserve the 8- and high-bit-depth edge layout and 4-, 6-, 8-, and 14-tap kernels for SIMD-first traversal, with sharpness thresholds, per-block intra/inter reference and mode deltas, current and previous skipped-transform decisions, transform and prediction-unit edge selection, and plane traversal shared by the exact scalar fallback. This is normative AV1 image reconstruction and adds neither generic ISO BMFF models nor retained video reference state.
Av1CdefDecoder, Av1CdefFilter, and CDEF-unit strength storage	AV1 sections 7.15.2 through 7.15.4 constrained directional enhancement filtering	libaom av1/common/cdef.c, av1/common/cdef_block.c, av1/common/cdef.h, and av1/common/cdef_block.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Design direction search and constrained primary/secondary filtering around vector-friendly block storage and SIMD-first kernels, with variance adjustment, subsampling direction conversion, skipped-8x8 selection, and frame-edge sentinel behavior shared by the scalar fallback. Use a frame-owned source snapshot so filtering never consumes already modified samples. This is normative AV1 still-image reconstruction and introduces no ISO BMFF, track, timing, or sequence-playback surface.
Av1SuperResolutionDecoder, Av1SuperResolutionFilter, frame-size derivation, and decoded-image dimensions	AV1 section 7.16 normative super-resolution upscaling	libaom av1/common/resize.c, av1/common/resize.h, av1/common/convolve.c, av1/common/x86/av1_convolve_horiz_rs_sse4.c, and aom_dsp/aom_filter.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Port the fixed 64-phase, 8-tap horizontal filter, phase/step derivation, replicated frame edges, chroma width rounding, signed rounding, and 8/10/12-bit clipping. Follow libaom's four-output multiply/add, two-stage horizontal reduction, packed-store traversal with exact scalar fallback. Generic image resizing is not normative AV1 super-resolution. This adds no track, timing, fragment, animation, or generic ISO BMFF model.
Av1TileReader loop-restoration unit syntax, Av1SymbolDecoder restoration distributions/subexponential codes, and Av1FrameInfo unit storage	AV1 section 5.11.57 read_lr and read_lr_unit syntax	libaom av1/decoder/decodeframe.c, av1/common/restoration.c, av1/common/restoration.h, av1/common/entropymode.c, aom_dsp/binary_codes_reader.c, and aom_dsp/recenter.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Decode tile-local switchable/Wiener/self-guided selections, finite reference-subexponential coefficients, chroma Wiener windows, self-guided parameter sets, super-resolution-adjusted unit corners, and the AV1 nearest-unit-count rule into frame-owned per-plane grids. This is compressed still-image syntax and adds no movie, track, timing, fragment, audio, or sequence surface.
Av1WienerFilter	AV1 sections 7.17.4 and 7.17.5 Wiener restoration filtering and coefficient derivation	libaom av1/common/restoration.c, av1/common/restoration.h, av1/common/convolve.c, and av1/common/convolve.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Preserve the implicit center-sample contribution, separable horizontal/vertical rounding, bit-depth-dependent 16-bit intermediate range, and final 8/10/12-bit clipping. Reuse Vector128_.MultiplyAddAdjacent for the contiguous horizontal eight-tap product with an exact scalar fallback. Keep the restoration stage disabled until stripe boundaries and self-guided filtering are both complete.
Av1SelfGuidedFilter	AV1 sections 7.17.2 and 7.17.3 self-guided and box-filter processes	libaom av1/common/restoration.c and av1/common/restoration.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Design the window layout, caller-owned scratch, local statistics, and projection traversal from libaom's architecture-specific kernels so the production path is SIMD-first. Implement the equivalent scalar fallback through the same contract. Preserve the sixteen normative radius/variance parameter sets, local mean and variance normalization, alternating-row radius-two optimization, decoded projection-coefficient behavior, signed rounding, and 8/10/12-bit clipping. Keep this image-reconstruction stage disabled until restoration stripe boundaries are complete.
Av1LoopRestorationBoundary, Av1LoopRestorationDecoder, and Av1FrameDecoder restoration-stage ordering	AV1 section 7.17 loop restoration, including striped boundary semantics	libaom av1/common/restoration.c, av1/common/restoration.h, av1/common/resize.c, and av1/decoder/decodeframe.c at 03087864cf4bea6abb0d28f95cf7843511413d8f	Preserve two deblocked rows at internal 64-luma stripe boundaries before CDEF, apply the existing normative SIMD-backed super-resolution kernel to saved rows when scaled, use post-CDEF/super-resolution samples at frame edges, extend the final restoration unit up to 150 percent of nominal size, and filter from immutable plane snapshots into separate output planes. This is bounded still-image reconstruction state, not retained reference-frame, track, timing, or playback state.
Av1FilmGrainDecoder, Av1FilmGrainGaussianSequence, Av1FilmGrainNoise, Av1FilmGrainOverlap, and Av1FilmGrainSampleOperations<TSample>	AV1 section 7.18 film-grain synthesis	libaom av1/decoder/grain_synthesis.c, av1/decoder/grain_synthesis.h, and aom_dsp/grain_params.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Preserve the normative 2,048-sample Gaussian sequence as compile-time span data, the serial linear-feedback shift register and luma/chroma autoregressive templates, scaling lookup interpolation, 32x32 block selection, boundary overlap, restricted-range clipping, monochrome and 4:2:0/4:2:2/4:4:4 layouts, and 8/10/12-bit arithmetic. Use allocator-owned scratch. Apply scaling and noise through AVX2 gather arithmetic at every bit depth, retain the measured cross-platform 128-bit high-bit-depth path, and use the exact scalar path for 8-bit machines without AVX2 and for vector tails. Process horizontal overlap through preferred-native 512-bit, then 256-bit, 128-bit, and scalar tiers; retain scalar vertical overlap because each output row exposes only one or two strided samples. Keep SIMD width, ISA, sample storage, and bit depth out of folders, namespaces, files, and type names. Apply grain only to displayed samples after all in-loop filters, inherit reference-selected parameters through the bounded image-layer decoder state, and keep the ungrained reconstruction in every refreshed reference slot. This does not add sequence playback or video scope.
Av1FrameInfo, Av1TileReader, and Av1BlockDecoder transform/coefficient storage	AV1 section 5.11.39 coefficient syntax and section 7.11.2 reconstruction	libaom av1/decoder/decodetxb.c and av1/decoder/decoder.h at 03087864cf4bea6abb0d28f95cf7843511413d8f	Preserve separate luma and chroma transform coefficients at monotonically advancing per-plane offsets within each superblock so reconstruction consumes the same transform-block order produced by tile parsing.
Av1InverseQuantizer and Av1InverseQuantizationLookup	AV1 section 7.12.3 inverse quantization	libaom aom_dsp/aom_dsp_common.h, av1/common/quant_common.c, and av1/decoder/decodetxb.c at 03087864cf4bea6abb0d28f95cf7843511413d8f	Select the per-segment matrix level, alias 64-pixel transform dimensions to their adjusted matrices, retain a flat level-15 matrix, and apply the five-bit inverse-matrix weight scale. The large managed lookup remains a single process-wide table.
Av1ForwardTransformer, Av1Inverse2dTransformer, Av1Transform2dFlipConfiguration, their nested forward/inverse 1-D operator structs, and Av1Transform1dMath	AV1 forward transform definitions and section 7.11.2 inverse transform and reconstruction	libaom av1/encoder/av1_fwd_txfm2d_hwy.h, its AVX-512 and AVX2 instantiations, av1/encoder/av1_fwd_txfm1d.c, av1/common/av1_inv_txfm1d_cfg.h, av1/common/av1_inv_txfm1d.c, av1/common/av1_inv_txfm2d.c, the x86 AVX2/SSE4 implementations, and the corresponding Neon implementations at 03087864cf4bea6abb0d28f95cf7843511413d8f	Preserve the normative staged DCT, ADST, and identity arithmetic, direction-specific stage ranges and shifts, transposition, clipping, and high-bit-depth sample addition. Stateless static-generic operators follow ImageSharp's JPEG color-transform pattern and remain nested in the family that owns their shared traversal. Forward operators share one behavior model across scalar, Vector128, Vector256, and Vector512, selecting packed or expanded lane storage at the 2-D boundary. Inverse production traversal retains the verified scalar, Vector128, and Vector256 tiers until a wider upstream shape and complete-block benefit are both established.
HeifDecoderCore box extension handling and HeifDecoderCore/HeifEncoderCore item-property associations	ISO/IEC 14496-12 box extensibility and section 8.11.14 item properties and ipma syntax	libavif src/read.c and src/write.c at 092276ce89098ead06db80975173191e5fee1826	Skip unrecognized top-level and metadata child boxes, preserve the position of every property in ipco, reject an unrecognized property only when its item association marks it essential, associate properties by item ID, and read or write the essential bit plus one-based 7-bit or 15-bit property index according to the full-box flags. Independent HEIC, HIF, and AVIF fixtures provide the reader oracle; container-level identification of encoded output guards the writer independently of pixel roundtripping.
HeifCleanAperture, HeifItem presentation state, and HeifDecoderCore transformative-property parsing and application	ISO/IEC 14496-12 section 12.1.4 clean aperture; HEIF image rotation and mirror properties; MIAF section 7.3.6.7 presentation order and section 7.3.9 essential transformative properties	libavif src/avif.c clean-aperture conversion, src/read.c property parsers and alpha-property validation, and apps/shared/avifutil.c transform application at 092276ce89098ead06db80975173191e5fee1826	Resolve fractional clean-aperture dimensions and center offsets to exact bounded integer pixels, validate the registered rotation/mirror reserved bits, require essential associations, crop after auxiliary-alpha composition, map counter-clockwise HEIF quarter turns to ImageSharp's optimized clockwise rotate modes, then mirror around the signaled axis. Reuse ImageSharp's existing crop, rotation, and flip processors for every pixel type. Retain only the three image-item property values; do not add a generic transform-box or ISO BMFF model.
HeifConstants.IsSupportedFileType, HeifImageFormatDetector, and HeifDecoderCore.CheckFileTypeBox	ISO/IEC 14496-12 FileTypeBox syntax and the MP4 Registration Authority HEIF/AVIF still-image and sequence brand registrations	libavif src/read.c functions avifParseFileTypeBox, avifFileTypeHasBrand, and avifFileTypeIsCompatible at 092276ce89098ead06db80975173191e5fee1826	Apply one rule to the major and compatible brands, accept implemented still-image brands and the bounded avis, hevc, and hevx image-sequence brands, and distinguish item and sequence presentation before parsing their payloads. Layered HEVC and JPEG sequence brands remain unsupported. The decoder validates the complete ftyp payload; the fixed-size format detector inspects the available prefix.
HeifSequenceParser, HeifTrackMetadataParser, and the bounded HEIF image-sequence writer	ISO/IEC 23008-12:2022 clauses 7 through 10 and HEVC Annex B; ISO/IEC 14496-12 sample tables, normal edit lists, and track-level metadata; AVIF 1.2 section 3; AV1-ISOBMFF 1.2 section 2	libavif src/read.c track/sample-table and track-metadata paths and src/write.c sequence path at 092276ce89098ead06db80975173191e5fee1826; independently produced HEIC sequence fixtures remain required	Resolve only pict master tracks and linked image alpha/metadata state into validated sample descriptors, exact frame durations, sync/dependency state, repetition, presentation transforms, color/HDR properties, and frame metadata. Track metadata is limited to implicitly associated Exif and unencoded XMP items declared by bounded iinf/infe, iloc, and optional idat syntax. DecoderOptions controls metadata retention and recoverable ancillary/image-data errors through the same rules as other decoders. Consume the required container boxes listed in the bounded syntax matrix without exposing a movie, track, edit, or sample-table API.
HeifDecoderCore.ReadBoxHeader and HeifDecoderCore.ParseBoxHeader	ISO/IEC 14496-12 section 4.2.2 basic box syntax	libavif src/stream.c functions avifROStreamReadBoxHeaderPartial and avifROStreamReadBoxHeader at 092276ce89098ead06db80975173191e5fee1826	Resolve 32-bit, 64-bit, UUID, and top-level size-zero boxes into content lengths only after validating the complete variable-sized header and the remaining parent boundary. Nested size-zero boxes are invalid; large skips retain 64-bit offsets.
HeifDecoderCore.ParseMetadata	ISO/IEC 14496-12 MetaBox and HEIF item declarations, locations, properties, and associations	libavif src/read.c functions avifParseMetaBox, avifMetaFindOrCreateItem, avifParseItemLocationBox, and avifParseItemPropertiesBox at 092276ce89098ead06db80975173191e5fee1826	Index unique recognized metadata children by type and payload location, then parse them in dependency order so physical placement does not control item lookup or property association. Duplicate unique children and truncated full-box headers are invalid.
HeifDecoderCore.ApplyAssociatedMetadata	HEIF Annex A Exif item data, MIME metadata items, and cdsc item references	libavif src/read.c function avifDecoderFindMetadata, src/exif.c function avifGetExifTiffHeaderOffset, and the Exif/XMP item writing paths in src/write.c at 092276ce89098ead06db80975173191e5fee1826	Resolve only metadata items whose cdsc reference identifies the decoded primary image, validate the Exif TIFF-header offset, and attach Exif or application/rdf+xml XMP through ImageSharp's existing profile types. This is a bounded still-image metadata path; it does not introduce a generic ISO BMFF metadata, media, or track model.
HeifPropertyParser, HeifDecoderCore color-property association, HeifItem and HeifSequenceTrack color profiles, and Av1Decoder effective color validation	ISO/IEC 14496-12 section 12.1.5 color information; HEIF section 6.5.5.1 color-information properties; AV1-ISOBMFF section 2.3.4 configuration semantics	libavif src/read.c functions avifParseColourInformationBox, avifReadColorNclxProperty, and avifReadColorProperties, plus src/write.c function avifEncoderWriteColorProperties, at 092276ce89098ead06db80975173191e5fee1826	Associate at most one ICC and one nclx property with each presented color image item or selected image-sequence sample entry, validate ICC payloads and CICP reserved bits, expose them through ImageSharp's existing profile types, and inherit a grid's CICP description only for tiles that do not declare one. For AV1 images, require every specified sequence-header CICP field and the range flag to match nclx; use the container property only to supply sequence-header fields coded as unspecified before color conversion, then expose the effective CICP values even when no property is present. Retain only the two image color profiles; do not add a reusable color-box or media model.
HeifPropertyParser, HeifPixelAspectRatio, item/track presentation state, and HeifDecoderCore.ApplyItemPixelAspectRatioMetadata	ISO/IEC 14496-12 section 12.1.4.3 pixel aspect ratio	libavif src/read.c function avifParsePixelAspectRatioBox, src/write.c functions avifEncoderWritePaspProperty and avifEncoderWriteTransformativeProperties, and the presented-image property selection in src/read.c at 092276ce89098ead06db80975173191e5fee1826	Preserve the registered pasp, clap, irot, and imir presentation properties on an associated image item or selected image-sequence sample entry, reject invalid lengths, reserved bits, duplicates, and non-integral clean apertures, and map the displayed pixel ratio and transforms through ImageSharp's existing metadata and processing model. No generic transform or display model is introduced.
Av1CodecConfiguration, HeifItem.Av1CodecConfiguration, Av1HeifItemDecoder, and AV1 grid configuration checks	AV1-ISOBMFF sections 2.3.3 and 2.3.4 codec-configuration record syntax and semantics; AVIF sections 2.1, 2.2.1, and 2.2.3 AV1 image-item, item-configuration, and HDR metadata constraints; AV1 sections 5.8.3, 5.8.4, 6.7.3, and 6.7.4 HDR metadata syntax and semantics; ISOBMFF mastering-display and content-light image properties; ITU-T H.274 section 8.9 mastering-display field semantics; MIAF section 7.3.11.4.1 grid input constraints	libavif src/read.c functions avifParseCodecConfiguration, avifDecoderItemValidateProperties, avifReadCodecConfigProperty, avifParseContentLightLevelInformation, and avifSkipMasteringDisplayColourVolume at 092276ce89098ead06db80975173191e5fee1826; libaom av1/decoder/obu.c functions read_metadata, read_metadata_hdr_cll, and read_metadata_hdr_mdcv at 03087864cf4bea6abb0d28f95cf7843511413d8f	Associate exactly one av1C property with each decoded av01 image item, validate the fixed record and its bit depth/chroma fields against the item's AV1 sequence header and optional pixi channel depths, require matching configurations across grid tiles, and report the encoded image precision and monochrome shape through HeifMetadata. Validate low-overhead OBU framing, require exactly one sequence header in the image item, allow at most one first-position sequence header in configOBUs, and compare a repeated header's extension and payload exactly while ignoring only its legal size-field representation. Decode clli and mdcv as still-image item properties, validate matching HDR CLL and HDR MDCV metadata OBUs from the combined configuration/item sequence, and account for the different primary order and fixed-point precision of the ISOBMFF and AV1 MDCV representations. Expose the effective HDR values without adding sample groups, tracks, or media metadata. Related still-image HDR properties remain required. Consume but do not retain presentation-delay syntax, and introduce no sample entry, sample description, track, timing, or generic decoder-configuration model.
HevcCodecConfiguration, HevcNalUnitArray, and HeifItem.HevcCodecConfiguration	ISO/IEC 14496-15 HEVCDecoderConfigurationRecord, HEVC NAL-unit header syntax, and HEVC image-item configuration requirements	HM source/Lib/TLibDecoder/NALread.cpp and parameter-header paths at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc v1.6.0 decoder NAL and parameter-header paths at c83a76b084498d55f252f48b2e3786804cdf24b7; independently produced HEIC fixtures for hvcC record layout	Associate exactly one hvcC property with each hvc1 image item; validate the fixed profile, level, chroma, bit-depth, reserved, length-size, array, and NAL-header fields within the property boundary; and retain only the image description plus bounded configuration NAL units required to decode that item. Consume but do not retain average-frame-rate or temporal-layer presentation fields. Do not add visual sample entries, sample descriptions, tracks, timing, access-unit timelines, or a generic decoder-configuration model. BSD and Apache-2.0 notice preservation and separate HEVC patent/release clearance remain final integration gates.
HevcNalUnit, HevcNalUnitHeader, HevcRbspDecoder, and HevcBitReader	HEVC sections 7.3.1.1 and 7.3.1.2 NAL-unit and header syntax, section 7.3.2.11 RBSP trailing bits, and section 9.1 Exp-Golomb parsing	HM source/Lib/TLibDecoder/NALread.cpp and SyntaxElementParser.cpp at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc NAL extraction and bitstream paths at c83a76b084498d55f252f48b2e3786804cdf24b7	Validate the fixed NAL header, remove only legal emulation-prevention bytes, and provide bounded fixed-width, flag, unsigned/signed Exp-Golomb, and RBSP-trailing-bit reads. Configuration arrays immediately use the shared parser. The primitives consume only NAL units delivered by a supported HEIF image item or its hvcC property and introduce no Annex B API, access-unit model, timeline, track, sample table, or generic ISO BMFF surface.
HevcProfileTierLevel, HevcVideoParameterSet, and HevcCodecConfiguration.VideoParameterSets	HEVC sections 7.3.2.1 video parameter sets, 7.3.3 profile-tier-level syntax, and E.2.2 hypothetical-reference-decoder syntax; ISO/IEC 14496-15 HEVCDecoderConfigurationRecord matching semantics	HM source/Lib/TLibDecoder/TDecCAVLC.cpp functions parseVPS, parsePTL, parseProfileTier, and parseHrdParameters at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoder/ihevcd_parse_headers.c profile-tier-level and VPS paths at c83a76b084498d55f252f48b2e3786804cdf24b7	Retain the base-layer VPS identifier, temporal-sublayer count, nesting flag, and exact general profile/tier/constraint/level fields. Validate profile, tier, compatibility, level, temporal count, and nesting against hvcC, while tolerating the independently observed legacy-muxer practice of zeroing some or all profile-specific constraint bits in the configuration record; SPS validation remains authoritative for chroma and bit depth. Consume but do not retain sublayer ordering, timing, and HRD syntax. Reject multilayer and layer-set state because the supported contract is one independently coded image item; add no layer-selection, access-unit, DPB, scheduling, track, or sample model.
HevcSequenceParameterSet, HevcVideoUsabilityInformation, HevcScalingList, HevcShortTermReferencePictureSet, and HevcParameterSetSyntax	HEVC sections 7.3.2.2 sequence parameter sets, 7.3.4 scaling-list data, 7.3.7 short-term reference-picture sets, E.2.1 VUI syntax, and Range Extensions SPS syntax	HM source/Lib/TLibDecoder/TDecCAVLC.cpp functions parseSPS, parseScalingList, xDecodeScalingList, parseShortTermRefPicSet, parseVUI, and parseHrdParameters, plus source/Lib/TLibCommon/TComRom.cpp default matrices and diagonal scans, at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoder/ihevcd_parse_headers.c SPS, VUI, scaling-list, and reference-set paths at c83a76b084498d55f252f48b2e3786804cdf24b7	Retain coded/display dimensions, conformance crop, monochrome/4:2:0/4:2:2/4:4:4 and separate-plane layout, 8-through-16-bit precision, coding/transform/PCM geometry, effective scaling matrices, compression-tool flags, bounded reference-set declarations, Range Extensions tools, and still-image VUI aspect/color/range/chroma-location/display-window fields. Consume timing, HRD, ordering, and bitstream-restriction syntax without scheduling state. Reject interlaced fields and multilayer extensions because they do not represent the one progressive independently coded image-item contract; add no DPB pictures, playback state, tracks, or sample tables.
HevcPictureParameterSet and HevcCodecConfiguration.PictureParameterSets	HEVC sections 7.3.2.3 picture parameter sets, 7.3.4 scaling-list data, and Range Extensions PPS syntax	HM source/Lib/TLibDecoder/TDecCAVLC.cpp function parsePPS, source/Lib/TLibDecoder/TDecConformance.cpp PPS range checks, and source/Lib/TLibCommon/TComPicSym.cpp tile derivation at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoder/ihevcd_parse_headers.c PPS and tile paths at c83a76b084498d55f252f48b2e3786804cdf24b7	Link each PPS to a configuration-local SPS and retain the CABAC, quantization, transform-skip, scaling-list, tile/wavefront, deblocking, SAO, slice-header, and Range Extensions controls required to reconstruct the independently coded picture. Derive the complete CTB tile grid and validate every SPS-dependent depth, size, precision, and offset bound during parsing. Reject multilayer extensions and add no access-unit, reference-picture lifetime, playback, track, timing, or sample-table model.
HevcImageItemBitstream, HevcSliceSegmentHeader, and HevcSliceType	HEVC sections 7.3.1.2 NAL-unit syntax, 7.3.6 slice-segment headers, and 7.4.7 slice-segment-header semantics; HEVC image-item length delimiting	HM source/Lib/TLibDecoder/TDecCAVLC.cpp function parseSliceHeader, source/Lib/TLibDecoder/NALread.cpp, and slice byte-alignment and entry-point handling at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoder/ihevcd_nal.c, ihevcd_parse_slice_header.c, and ihevcd_parse_headers.c at c83a76b084498d55f252f48b2e3786804cdf24b7	Split only the selected hvc1 image item's one-through-four-byte length-delimited NAL units, reject in-band parameter-set replacement and end-of-sequence state, and require exactly one base-layer IDR picture. Retain the intra slice, separate-plane, SAO, quantization, deblocking, cross-slice filtering, tile/wavefront entry-point, extension, byte-alignment, and entropy-payload fields required by reconstruction. Keep emulation-prevention positions only to translate slice substream byte boundaries; add no Annex B scanner, access-unit collection, non-IDR reference lifetime, playback, timing, track, or sample model.
HevcHeifItemDecoder, HeifCompressionFactory, and HevcYuvConverter coded-window output	HEVC sections 6.1 through 6.3 decoded picture and conformance-window semantics; ISO/IEC 14496-12 section 12.1.5 color information and ISO/IEC 23008-12 HEVC image-item binding	HM source/Lib/TLibDecoder/TDecTop.cpp decoded-picture output and source/Lib/TLibCommon/TComPicYuv.cpp conformance-window handling at 9c1f298659ab0cee9dc13d23d0304221575410b9; ImageSharp's shared HEIF H.273 SIMD operators, pooled row conversion, and item-decoder lifecycle	Register only hvc1 still-image items, bind the selected slice PPS/SPS to one picture decoder, convert the conformance-window region directly from native component planes without an intermediate packed image, and preserve the effective container-or-VUI CICP description used by conversion. Reuse the shared descending-width SIMD color operators and pixel packers with one pooled row workspace. Add no Annex B entry point, decoded-picture queue, inter-picture state, track, timing, or general-purpose HEVC API.
HevcCabacContext and HevcCabacDecoder	HEVC sections 9.3.3.2 through 9.3.3.4 context initialization, arithmetic decoding, bypass decoding, and termination	HM source/Lib/TLibCommon/ContextModel.cpp, source/Lib/TLibCommon/TComCABACTables.cpp, and source/Lib/TLibDecoder/TDecBinCoderCABAC.cpp at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc common/ihevc_cabac_tables.c, common/ihevc_cabac_tables.h, decoder/ihevcd_cabac.c, and decoder/ihevcd_cabac.h at c83a76b084498d55f252f48b2e3786804cdf24b7	Implement the bit-exact scalar probability-state transitions, range normalization, bounded byte refills, bypass modes, termination, and alignment validation for one slice tile or wavefront substream. CABAC decisions are serial and adaptive, so this primitive is the scalar oracle rather than a SIMD target; later prediction, transform, filtering, and color-conversion stages must reuse ImageSharp's compatible SIMD kernels. The decoder owns no container boxes, access units, reference pictures, tracks, timing, playback, or general-purpose bitstream surface.
HevcCabacContexts	HEVC section 9.3.2.2 initialization of context variables for an intra-coded slice and Range Extensions context additions	HM source/Lib/TLibCommon/ContextTables.h and source/Lib/TLibDecoder/TDecSbac.cpp function resetEntropy at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc common/ihevc_cabac_tables.c and decoder/ihevcd_cabac.c at c83a76b084498d55f252f48b2e3786804cdf24b7	Initialize only the context groups reachable from the independently coded intra-picture syntax, including transform skip, chroma quantization adjustment, and cross-component prediction. Keep one contiguous mutable context set per entropy substream so tile and wavefront synchronization can copy the normative adaptive state without retaining reference-picture or playback state. Omit skip, merge, motion, reference-index, and other inter-picture-only groups.
HevcCabacSyntaxReader	HEVC sections 9.3.3 through 9.3.4 CABAC binarization and context selection for intra coding units, transform trees, residual coefficients, sample-adaptive offset, and Range Extensions tools	HM source/Lib/TLibDecoder/TDecSbac.cpp, source/Lib/TLibCommon/TComChromaFormat.h, and source/Lib/TLibCommon/TComTrQuant.cpp at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoder/ihevcd_cabac.c, decoder/ihevcd_parse_slice.c, and decoder/ihevcd_parse_residual.c at c83a76b084498d55f252f48b2e3786804cdf24b7	Decode only syntax values reachable while reconstructing the selected independently coded still picture: intra partition and mode selectors, transform splitting and coded-block flags, quantization adjustments, transform skip, coefficient significance and levels, cross-component prediction, SAO, bypass alignment, and termination. Keep neighbor-dependent context selection in the owning coding-tree/reconstruction layer and expose no inter prediction, reference-picture, access-unit, track, timing, or playback syntax.
HevcCoefficientScanOrder, HevcCoefficientCodingParameters, and HevcCoefficientDecoder	HEVC sections 9.3.4.2.4 through 9.3.4.2.8 last-position, significance-map, level, sign-data-hiding, and Rice-adaptation processes; Range Extensions transform-skip context and bypass-alignment rules	HM source/Lib/TLibCommon/TComRom.cpp grouped scan generation, source/Lib/TLibCommon/TComChromaFormat.cpp transform-unit entropy parameters, source/Lib/TLibCommon/TComTrQuant.cpp significance-context derivation, and source/Lib/TLibDecoder/TDecSbac.cpp function parseCoeffNxN at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoder/ihevcd_parse_residual.c at c83a76b084498d55f252f48b2e3786804cdf24b7; ImageSharp MemoryAllocator and fixed inline buffers	Generate diagonal, horizontal, and vertical grouped scans into entropy-substream-owned scratch; derive directional scan selection, last-significant positions, right/lower group contexts, significant coefficient flags, greater-than-one/two levels, escape remainders, sign hiding, and the four persistent Rice states. Allocate one pooled scratch block per entropy substream and perform no allocation or dynamic stack growth per transform block. This serial adaptive stage intentionally remains scalar and feeds the existing SIMD-first inverse quantization, inverse transform, transform-skip, and residual-reconstruction kernels.
HevcPlane and HevcPictureBuffer	HEVC sections 6.2 and 6.3 source and decoded picture sample-array dimensions for monochrome, 4:2:0, 4:2:2, 4:4:4, and separate-color-plane coding	HM source/Lib/TLibCommon/TComPicYuv.cpp at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoded-picture buffer layout paths at c83a76b084498d55f252f48b2e3786804cdf24b7; ImageSharp MemoryAllocator and Buffer2D<T>	Allocate one allocator-owned native ushort plane set for the selected still picture so the same reconstruction path preserves every supported 8-through-16-bit sample without byte-backed reinterpretation. Derive chroma dimensions by ceiling division for each sampling layout and treat separate color planes as full-resolution independently coded arrays. The buffer contains one picture only and introduces no decoded-picture buffer, reference lifetime, frame queue, or playback state.
HevcCodingTreeState	HEVC sections 6.4 coding-tree block and coding-block availability plus section 9.3 split-flag context derivation	HM source/Lib/TLibCommon/TComDataCU.cpp function getCtxSplitFlag and source/Lib/TLibDecoder/TDecCu.cpp coding-tree traversal at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc decoder/ihevcd_parse_slice.c coding-tree paths at c83a76b084498d55f252f48b2e3786804cdf24b7; ImageSharp MemoryAllocator and Buffer2D<T>	Store only leaf depth, effective luma QP, transquant-bypass, and PCM state at minimum-coding-block resolution for the selected picture. Derive split contexts from caller-approved left and above availability so slice and tile boundaries remain owned by traversal rather than hidden in a general block graph. Clip edge writes to the coded still-picture state and add no prediction units, motion fields, references, decoded-picture queue, or sequence lifetime.
HevcIntraPredictionState	HEVC sections 8.4.2 and 9.3 intra luma/chroma prediction-mode derivation and binarization	HM source/Lib/TLibCommon/TComDataCU.cpp functions getIntraDirPredictor and getAllowedChromaDir, plus source/Lib/TLibDecoder/TDecSbac.cpp functions parseIntraDirLumaAng and parseIntraDirChroma, at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc intra-mode parsing and neighbor derivation paths at c83a76b084498d55f252f48b2e3786804cdf24b7; ImageSharp MemoryAllocator and Buffer2D<T>	Preserve the normative two-pass luma flag/suffix order, spatial most-probable-mode derivation, omitted-mode reinsertion, four-way minimum-CU partition order, explicit chroma candidate substitution, and derived-chroma mode at 4x4 luma resolution. Accept slice/tile availability from traversal, fill only the selected still-picture map, and add no inter prediction unit, motion, reference, or sequence state.
HevcIntraPredictor	HEVC section 8.4.4.2 intra sample prediction and section 8.4.4.2.3 reference-sample substitution and filtering	HM source/Lib/TLibCommon/TComPrediction.cpp functions predIntraGetPredValDC, xPredIntraAng, xPredIntraPlanar, and xDCPredFiltering, plus source/Lib/TLibCommon/TComPattern.cpp function fillReferenceSamples and its reference-filter selection at 9c1f298659ab0cee9dc13d23d0304221575410b9; Android libhevc intra-prediction kernels at c83a76b084498d55f252f48b2e3786804cdf24b7	Predict from caller-prepared references with planar, DC, and all 33 angular modes; negative-angle extension; three-tap and strong-bilinear smoothing; and the normative luma boundary filters. Traverse Vector512, Vector256, and Vector128 widths before one scalar tail, use caller-owned reusable scratch for horizontal transposition, and retain no generic video, inter-picture, or reference-picture state.
HeifContentColorVolume, HeifItem.ContentColorVolume, and HeifDecoderCore content color-volume parsing and presentation	HEIF content color-volume item property; AVIF 1.2 content color-volume requirements; ITU-T H.274 (V4) content colour volume syntax and semantics	libavif src/read.c function avifSkipContentColourVolume at 092276ce89098ead06db80975173191e5fee1826; official ITU-T H.274 (V4), January 2026	Decode only the bounded per-image cclv property: require zero cancellation, persistence, and reserved bits; preserve optional signed G/B/R primary coordinates and normalized minimum, maximum, and average luminance values; and validate their registered ranges and ordering. Expose the effective grid-or-tile still-image value through HeifMetadata. Do not add SEI persistence, retained video state, tracks, samples, timing, or a generic ISO BMFF color-volume box model.
HeifAmbientViewingEnvironment, HeifReferenceViewingEnvironment, HeifNominalDiffuseWhite, and their per-item presentation metadata	ISOBMFF ambient viewing environment; ITU-T H.274 (V4) section 8.13; HEIF Amendment 1 sections 6.5.44 and 6.5.45; AVIF 1.2 image-item box requirements	libavif src/read.c functions avifSkipAmbientViewingEnvironment, avifSkipReferenceViewingEnvironment, and avifSkipNominalDiffuseWhite at 092276ce89098ead06db80975173191e5fee1826; official ITU-T H.274 (V4), January 2026	Decode the fixed-size amve, version-zero reve, and version-zero ndwt properties only when associated with the presented still-image item. Preserve physical illuminance and luminance units, distinct surround/periphery chromaticities, and the coded zero that requests the standard nominal diffuse-white default. Validate registered coordinate ranges and duplicate associations. Retain no video-SEI persistence, visual sample entry, display pipeline, track, timing, or generic viewing-environment box model.
GridHeifItemDecoder and HeifDecoderCore grid/thumbnail selection	ISO/IEC 23008-12 section 6.6.2.3 image-grid syntax and MIAF grid-cell constraints	libavif src/read.c functions avifParseImageGridBox, avifDecoderDataAllocateImagePlanes, and avifDecoderDataCopyTileToImage at 092276ce89098ead06db80975173191e5fee1826	Parse version-zero 16-bit and 32-bit grid descriptors, preserve row-major dimg order, require the declared tile count and one coding format, validate canvas coverage and edge overlap, and crop only the rightmost column and bottom row while copying through ImageSharp pixel buffers. A primary grid whose tile codec is unavailable may use only a decodable thumbnail that explicitly references that grid.
HeifDecoderCore alpha auxiliary selection/composition and GridHeifItemDecoder auxiliary tile ordering	ISO/IEC 23008-12 alpha auxiliary image semantics, auxC, auxl, prem, and per-grid-tile alpha relationships	libavif src/read.c functions avifParseAuxiliaryTypeProperty, avifDecoderItemIsAlphaAux, avifMetaFindAlphaItem, and avifDecoderCheckAlphaProperties, plus src/scale.c box-filter scaling at 092276ce89098ead06db80975173191e5fee1826	Recognize both registered alpha URNs, decode a direct alpha image/grid or the complete row-major set of per-color-tile alpha auxiliaries, normalize through L16, box-resample differing auxiliary dimensions, compose through Rgba64 and PixelOperations<TPixel>, and unassociate prem color samples with transparent-black handling. This remains an image-item relationship only; no track or generic media-reference model is introduced.

This table is intentionally incomplete. Add a row before each additional AV1 or HEVC algorithm is ported or materially reshaped.

Current implementation assessment

This assessment was reconciled with the source tree on 2026-08-30, including production checkpoint 9d7348efd6496b54cef7ec27e42352641439c400. Unless a result is stated explicitly, each item is a source-inspection finding rather than a verified interoperability claim.

Public integration

• HeifFormat combines the HEIF, HEIC, HIF, and AVIF identities and extensions, but the implementation does not yet decode all payloads that contract implies.
• HeifDecoder now defaults to Rgba32, preserving decoded auxiliary alpha for non-generic loads.
• HeifMetadata now reports alpha presence and the corresponding 24/32-bit RGB pixel shape, but complete decoded HEVC/AV1 bit depth, monochrome/chroma layout, color signaling, and profiles remain absent.
• HeifEncoder defines quality, alpha quality, effort, lossless, chroma-subsampling, and bit-depth contracts directly, without a single-implementation options interface. The legacy JPEG path applies its supported quality, bit-depth, and chroma options and rejects unsupported combinations; AV1 and HEVC must implement the same public contracts before the Phase 1 API-review gate can pass.
• HEIF/HEIC/AVIF is absent from the format source-generation list in _Formats.ttinclude, so the standard ImageSharp save extensions are not generated.
• Configuration registration exists, but it currently registers capabilities broader than the implementation provides.

HEIF/ISO BMFF container

• The bounded reader now handles basic, extended-size, UUID, and permitted top-level to-end boxes, skips unknown optional top-level and metadata children, and rejects child boxes that escape their parent.
• File type handling evaluates supported still-image major and compatible brands and recognizes the bounded avis, hevc, and hevx image-sequence brands. The sequence parser retains only the selected picture and auxiliary-alpha tracks, timing, sample locations, dependencies, presentation properties, and image metadata required by animated HEIF; unrelated media tracks and general ISO BMFF playback state are not materialized.
• Item IDs are resolved as keys rather than list indices; metadata children are indexed and parsed in dependency order rather than physical order.
• Item property associations now preserve physical ipco indices, apply one-based 7-bit or 15-bit indices and essential flags, associate by item ID, and reject arbitrary unknown essential properties.
• Item locations now support bounded file-relative and idat-relative storage, multiple ordered extents, versioned item IDs, 0/4/8-byte registered field sizes, and 64-bit offsets. Referenced-item construction method two and external data references remain explicitly unsupported.
• Grid derived-image decoding now parses both registered descriptor widths, resolves the ordered dimg cells, validates tile count, coding format, dimensions, canvas coverage, and edge overlap, then composes the output through ImageSharp row buffers. Unsupported grid tile codecs can select only a decodable thumbnail of the same primary grid. HEVC grids have exact complete-image libheif references and exact HM tile-plane references; independent AV1 and JPEG grid fixtures are still required.
• Alpha auxiliary decoding recognizes auxC, auxl, and prem, supports direct auxiliary images, auxiliary grids, and per-color-grid-tile alpha, and reports alpha presence. The common-size path normalizes the native luma plane and composes each row directly into the destination; the scaled path uses allocator-owned row staging and the bounded ImageSharp resize window instead of materializing a complete Image<L16>. HEVC auxiliary-alpha presentation matches the independent libheif reference exactly, and independently encoded AVIF alpha fixtures cover composed alpha and ICC interaction.
• Clean aperture, image rotation, and image mirror properties now validate their registered payloads, exact integer crop geometry, and essential associations; affect Identify dimensions; and reuse ImageSharp's optimized crop/rotate/flip processors after auxiliary alpha composition in the MIAF-defined order. Independent transform vectors must still verify every crop/rotation/mirror/alpha combination.
• Decode now resolves cdsc-associated Exif and application/rdf+xml XMP items for the primary still image, validates the declared Exif TIFF-header offset, and attaches the payloads through ImageSharp's existing profile types before presentation transforms. Independent AVIF, HEIC, and HIF metadata fixtures and Identify-time profile reporting remain required.
• ICC and nclx CICP color properties are now associated with the presented color item instead of global parser state, validated, and exposed on Decode and Identify through the existing ImageSharp profiles. AV1 decode requires specified sequence-header fields and the range flag to match nclx, uses the property only for fields the bitstream leaves unspecified, and exposes the effective sequence-header CICP description when no property is present. Independent libavif fixtures verify exact ICC byte preservation, observable non-sRGB conversion, canonical-sRGB compaction, non-sRGB compact preservation, and metadata skipping for primary, grid, alpha-composited, and every presented sequence frame; the alpha vector also verifies that ICC conversion does not alter composed alpha. Identify-time bitstream fallback, independent CICP vectors, and HEVC integration remain required.
• Pixel aspect ratio now preserves the complete unsigned spacing pair, affects Decode and Identify through ImageSharp's existing unitless resolution metadata, and follows quarter-turn presentation rotation. Independent grid, rotation, and maximum-spacing fixtures remain required.
• AV1 codec configuration is now retained per av01 image item rather than in decoder-global state. Decode requires the property, validates its fixed record against the item's sequence header and any associated pixi channel depths, and requires matching configurations across grid tiles. Identify now reports the configuration's 8/10/12-bit precision and monochrome shape. The optional configOBUs sequence is bounded and validated, including its mandatory size fields, first-position/at-most-one sequence-header rule, the image item's exactly-one sequence-header rule, and exact comparison of a repeated configuration header with the item header. Content light-level and mastering-display color-volume information are decoded from the bounded clli and mdcv image properties and matching AV1 HDR metadata OBUs, with the representations' distinct fixed-point precision, grid/property precedence, and SkipMetadata behavior preserved. The still-image cclv, amve, reve, and ndwt properties preserve content-volume, ambient-viewing, mastering-environment, and diffuse-white descriptions without importing their video-SEI or sample-entry state models. Independent fixtures and encoder preservation remain required. HEVC hvcC is associated with each hvc1 image item, structurally validated through its VPS/SPS/PPS arrays, and used to report encoded precision and monochrome shape. The bounded decoder reconstructs exact native planes and exact genuine-HEIC presentation for the complete independently coded 8/10/12-bit monochrome, 4:2:0, 4:2:2, and 4:4:4 Range Extensions profile/tool matrix.
• Several image-item properties and relationships remain missing or parsed without fully affecting output.
• Identify and decode now use the same bounded metadata parser and both validate the complete leading file type box. The parsed state is still mutable and Identify does not yet report the complete bit depth, color, profile, or transform model.

HEVC decoder and encoder

• Heif4CharCode recognizes hvc1 image items, HevcCodecConfiguration validates and associates the bounded hvcC property, Identify reports its HEVC precision and monochrome shape, and HeifCompressionFactory registers the bounded HEVC still-image decoder.
• The HEVC path validates NAL headers and RBSP escaping, links hvcC VPS/SPS/PPS structures through Range Extensions, parses one length-delimited base-layer IDR picture, and reconstructs its complete intra-coded coding tree. It includes CABAC, transform traversal, coefficient decoding, intra prediction, inverse quantization and transform, PCM, deblocking, sample-adaptive offset, and allocator-owned native ushort planes for monochrome, 4:2:0, 4:2:2, 4:4:4, separate-color-plane, and 8-through-16-bit samples. The item adapter converts the SPS conformance window directly through the shared pooled SIMD-first H.273 pipeline and preserves the effective container-or-VUI CICP description. Four real item payloads match HM 18 native planes exactly, ten official Sony GENERAL pictures cover the 8/10/12-bit chroma-layout boundary, and 25 additional official pictures cover every exposed independently coded Range Extensions profile and required tool with exact native-plane hashes. Complete images covering grids, auxiliary alpha, ICC, and the complete independently coded Range Extensions profile/tool matrix match pinned presentation references exactly. The complete color-signaling matrix remains open.
• There is no HEVC encoder. HeifEncoderCore rejects HeifCompressionMethod.Hevc; the separately selected legacy JPEG path emits a generic HEIF/HIF-style JPEG image item and does not provide HEIC output.
• Permanent HEVC conformance tests compare extracted item planes exactly with HM output or normative decoded-picture hashes and complete presented images exactly with pinned references. Direct libheif/libde265 output is retained where its native reconstruction is exact; other profile/tool cases use pinned HM planes and independently reproduced pinned-libheif conversion. Native-plane and genuine-HEIC presentation fixtures cover the complete independently coded exposed profile/tool matrix.

AV1 decoder

• The bounded Av1Decoder session parses all tile state before allocating and reconstructing each coded image layer or sequence sample. After successful completion it retains the ungrained reference planes, frame header, frame information, and published entropy snapshot in the refreshed slots, while presentation-only ownership remains separate. A new accepted sequence header resets both parser and retained-owner state. Bounded image-sequence decoding accepts show_existing_frame, including its key-frame reference reset and separate film-grain presentation ownership, without exposing a general animation/video reference model.
• Committed inter-frame support parses single-reference selection, builds the fixed-capacity spatial and temporal reference-MV stack, decodes NEAREST/NEAR/NEW/GLOBAL and DRL syntax, decodes or infers interpolation filters, checks inter-intra and motion-mode eligibility, and invokes simple translational prediction before residual reconstruction. Commit 8b9aa4b2cfe59ff0f8c348d9bb06678a5ff12f2a records exact corrected dependent-frame native-plane and final-presentation comparisons plus the complete focused Release matrix and ownership/allocation evidence.
• Commit e1a764b8f2e8a8db98a003fb3b913213f40084db owns every compound reference-selection and inter-mode distribution through the retained CDF lifecycle, derives primary and secondary candidate vectors as inseparable pairs, reconstructs both retained references through the existing translational predictor, and combines them through one SIMD-first byte/ushort averaging operator with scalar parity. The pinned 19-frame libavif sequence exercises the complete production block branch and matches pinned libaom native planes and pinned libavif presentation exactly after decoding all preceding reference samples.
• Commit 59a4fbad5ed3bc7c34331c4bdc65acc7acde3f04 reconstructs eligible above and left OBMC neighbors through the retained primary reference, exact libaom masks, and the existing SIMD-first masked blender without per-block allocation. Direct production tests cover 8/10/12-bit storage plus 4:2:0 and 4:2:2 plane geometry. A pinned 19-frame sequence contains more than one hundred traced OBMC blocks and passes exact final native-plane, final-presentation, constrained-allocation, normal-dispatch, and scalar-dispatch comparison.
• Commit 8f57c29b250b4291f034b31c1ea3de3fa3345467 maps current-frame Q4 positions into retained-frame Q10 coordinates with pinned-libaom Q14 scale factors and signed rounding, then applies allocation-free variable-phase two-pass convolution across byte and high-bit-depth storage. The production block path selects scaling from actual retained/current dimensions, covers both primary and secondary workspaces plus scaled OBMC neighbors, and retains libaom's 288-sample prediction border. The real layered fixture also exposed and now verifies per-block inter loop-filter reference/mode deltas and skipped-transform edge decisions that the earlier intra-only deblocking integration did not model.
• Commit 0b2db86809e26757460e719a6b59c0fe58bb0586 derives AV1's local affine matrix from fixed-capacity spatial samples with pinned integer least squares, reciprocal lookup, coefficient clamping, and shear validation. Production reconstruction applies the pinned 193-phase Q7 warped filter in allocation-free 8x8 tiles through Vector128 convolution with scalar fallback for 8/10/12-bit planes; an invalid selected model falls back to translational prediction as libaom does.
• Commit c5637ea0187df35b385bf43e2fe85cd955f01099 selects the full frame-header rotation/zoom or affine matrix independently for each GLOBAL_GLOBALMV reference, while identity, translation, invalid shear, scaled references, integer-only frames, OBMC, and sub-8x8 plane predictions retain their normative translational path. A pinned bounded sequence proves seven actual single-reference GLOBALMV blocks through exact native and presentation output, and a direct FeatureTestRunner production test proves both compound matrices at 8, 10, and 12 bits.
• Source inspection against pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f resolves two disputed audit claims in favor of the current predicates. The interpolation call path uses is_nontrans_global_motion(), which returns false only for TRANSLATION, so an identity GLOBALMV block omits filter symbols. The single-reference spatial extension loops use MAX_MV_REF_CANDIDATES, which is two, while the full reference-MV stack capacity is eight. These loops are spatial extension, not temporal extension.
• Every padded AV1 frame plane requests ImageSharp's contiguous allocation contract, and constrained-allocator reconstruction passes without copying. Motion fields use allocator-owned storage and counted leases; allocation tracking passes for initialization, retained aliases, failure unwinding, presentation ownership, decoder-result ownership, and exactly-once disposal. The corrected logical progressive color payload reaches the production inter branch and its final native planes match pinned libaom exactly.
• Transform coefficient entropy derivation and updates now address the above contexts relative to the tile column and the left contexts relative to the current superblock row, preserve luma coordinates independently of chroma subsampling, and test every packed context entry for the libaom any-nonzero rule. Extended vertical partition updates advance the mode-information column rather than the row. The existing multi-superblock 4:4:4 AVIF fixture now completes tile parsing; independent coefficient-context vectors across tile boundaries, chroma layouts, bit depths, and edge-clipped transforms remain required.
• The reconstruction pipeline now records plane-relative transform geometry, preserves tile-local delta-Q and delta-LF predictors, derives segmentation and reference-adjusted filter levels, and runs the exact AV1 4-, 6-, 8-, and 14-tap deblocking kernels in normative vertical-then-horizontal order. Deblocking uses the same closed edge-operator architecture as the HEVC filter, with operators specialized by sample storage and orientation, Vector128<int> lanes representing the four rows or columns along an edge, and an allocation-free scalar fallback for disabled intrinsics. Exact native-plane comparison with pinned scalar libaom output now verifies active deblocking and complete reconstruction for real 8-bit 4:2:0, 10-bit 4:4:4, and 12-bit 4:4:4 content; genuine AVIF containers separately verify presentation and public bit-depth metadata. The pipeline then applies CDEF through one semantic filter architecture: paired AVX2 and single-block Vector128 direction analysis, closed primary/secondary strength operators, packed 4x4/4x8/8x4/8x8 constrained filtering, byte/16-bit output operators, and an exact allocation-free scalar fallback. Decoder orchestration now owns the immutable plane snapshots and clean direction/variance maps through ImageSharp's memory allocator, widens 8-bit source rows with the same AVX2/128-bit/scalar tiers as libaom, lists each unit's non-skipped blocks in fixed inline storage, analyzes listed blocks in pairs, and writes filtered bytes or 16-bit samples directly to the frame planes. Independently encoded active-CDEF 8-bit 4:2:0 and 10/12-bit 4:4:4 streams match every visible native sample produced by pinned scalar libaom with restoration disabled. Independently encoded AVIF containers at the same three bit depths also match pinned scalar-libavif presentation exactly under normal, 256-bit, 128-bit, and scalar color-conversion dispatch. Active super-resolution derives the Appendix A bounded coded width and applies the exact 64-phase, 8-tap horizontal filter with aligned reconstruction-edge input, 8/10/12-bit clipping, and the existing cross-platform Vector128_.MultiplyAddAdjacent helper. Independently encoded active-super-resolution AV1 streams at 8, 10, and 12 bits match every visible native sample produced by pinned scalar libaom under normal and forced-scalar dispatch. Independently packaged AVIF containers at the same bit depths retain matching libavif profile, dimensions, chroma, and CICP properties, require active super-resolution in their actual AV1 item, and match pinned scalar-libavif presentation exactly under normal, 256-bit, 128-bit, and scalar color-conversion dispatch. Loop restoration follows super-resolution, preserves the required pre-CDEF deblocked context at internal stripes, and applies decoded Wiener or self-guided units from immutable plane snapshots. Independently encoded active-restoration streams at 8, 10, and 12 bits now match every native sample from pinned scalar libaom across AVX2, 128-bit, and scalar dispatch, with the fixture matrix proving both Wiener and self-guided unit selection. The matching AVIF containers also match pinned scalar-libavif presentation exactly, and an independent direct-window definition verifies all sixteen self-guided parameter sets at each supported bit depth across vector and scalar dispatch. Combined 8-bit 4:2:0, 10-bit 4:2:2, and 12-bit 4:4:4 streams additionally verify restoration-unit boundaries after super-resolution, including clipped chroma transform traversal at a coded-frame edge.
• The visible still-image path applies the complete self-contained film-grain parameter set after all in-loop filters. Independently encoded pinned-libaom vectors match every native sample at 8, 10, and 12 bits across monochrome, 4:2:0, 4:2:2, and 4:4:4 layouts, full and restricted ranges, identity-matrix signaling, overlap, and odd 33x11 frame extension. FeatureTestRunner verifies normal, AVX-disabled, and fully scalar dispatch. A full-HD-equivalent 4:2:0 benchmark reports zero allocation: 8-bit AVX2 is 2.335 ms versus 5.806 ms scalar, while 12-bit AVX2 is 3.195 ms, cross-platform 128-bit is 7.382 ms, and scalar is 8.614 ms on the measured Ryzen platform. The slower 8-bit 128-bit lookup construction is deliberately not dispatched.
• Palette mode now reads the normative luma/chroma mode and size CDFs, neighbor color caches, high-bit-depth color syntax, diagonal color-index maps, and clipped-edge padding. Reconstruction retains libaom's exact indexed-palette contract while using repeated eight-entry tables and native lane-local shuffles across runtime-selected Vector512, Vector256, and Vector128 paths before an exact scalar fallback. FeatureTestRunner verifies palette sizes 2-8, transform widths 4-64, padded rows, and 8/10/12-bit samples. An independently encoded 33x11 8-bit YUV 4:4:4 AVIF fixture selects both luma and chroma palettes, matches every native sample from pinned scalar libaom, and matches scalar-libavif presentation exactly across the available vector widths and scalar fallback.
• Non-reduced still-image sequence parsing now consumes decoder-model operating parameters, temporal presentation fields, and OBU-layer-applicable buffer-removal fields only to preserve AV1 bit alignment. The scheduling values are not retained, and no movie, track, timing, playback, or generic ISO BMFF surface has been introduced. Existing focused sequence-header coverage exercises only reduced-still syntax, so an independent non-reduced still AVIF vector remains required.
• Loop-restoration unit parsing records tile-local switchable/Wiener/self-guided filter selections and coefficients in frame-owned plane grids, including super-resolution-adjusted unit corners and the corrected conditional 64x64-superblock unit-size bit. The active restoration stage implements the normative unit geometry, striped deblocked boundaries, Wiener filtering, self-guided projection, and 8/10/12-bit clipping. Self-guided filtering now follows libaom's summed-area-table, coefficient-grid, alternating-row radius-two, full radius-one, and projection stages through AVX2 and cross-platform 128-bit traversals with one exact scalar fallback and caller-owned scratch. Independently encoded fixtures cover active Wiener and self-guided reconstruction and exact AVIF presentation at every supported bit depth. An independent direct-window definition covers all sixteen self-guided parameter sets, narrow and odd processing units, both vector-tail widths, padded strides, and the scalar fallback. Combined active-restoration and super-resolution fixtures cover 4:2:0, 4:2:2, and 4:4:4 at 8, 10, and 12 bits, including restoration-unit boundaries and clipped chroma transform traversal. Other normative independently decodable still-image syntax paths still contain NotImplementedException or equivalent unsupported branches. Tile-local palette CDF adaptation is present; the remaining still-image frame-context behavior requires a separate source audit without introducing sequence playback state.
• The frame buffer now establishes two-byte native sample storage, logical plane rows, and sample-unit block strides for 10/12-bit frames. The active intra-prediction, inverse-transform, and block-reconstruction path selects native 16-bit samples for 10/12-bit frames and has focused pipeline wiring coverage. Chroma-from-luma storage, 4:4:4/4:2:2/4:2:0 subsampling, rounded mean subtraction, U/V sharing, and 8/10/12-bit prediction now traverse AVX2 and cross-platform Vector128 paths before an exact scalar fallback. FeatureTestRunner verifies every tier against independent fixed-point definitions across every supported block width; independently encoded high-bit-depth and chroma-from-luma AVIF conformance files are still required.
• Av1YuvConverter now consumes the signaled full or limited range, every non-reserved AV1 H.273 matrix coefficient, transfer characteristics where the matrix definition requires them, subsampling, and chroma sample position for 8, 10, and 12-bit output. Its high-bit-depth decode and encode paths use allocator-backed Rgb48 rows and the existing PixelOperations<TPixel> conversions, avoiding the former eight-bit intermediate. Encoder conversion covers monochrome, YUV 4:2:0, 4:2:2, and 4:4:4 with libavif-compatible box averaging. Identity, full/limited-range YCgCo, the fixed non-constant-luminance matrices, both fixed and chromaticity-derived constant/non-constant-luminance systems, SMPTE ST 2085, and PQ/HLG ICtCp are active in both directions. Independent vectors for every matrix, transfer, range, bit depth, sampling layout, and chroma position remain required before the complete color matrix is externally verified.
• Forward and inverse transforms use operation-owned workspace, explicit sequential fixed storage, and stateless static-generic operator structs for every valid DCT, ADST, and identity size. Named configuration factories keep the encoder's three shifts and variable cosine precision separate from the decoder's two shifts, fixed 12-bit cosine precision, and 8/10/12-bit clamp ranges. Forward traversal shares one libaom-shaped stage network across scalar, Vector128, Vector256, and Vector512; inverse production traversal uses the verified scalar, Vector128, and Vector256 tiers. Lossless segments now bypass the DCT pipeline and apply the reversible four-by-four inverse Walsh-Hadamard transform through the same byte/high-bit-depth output operators, with a Vector128 production path and caller-workspace scalar fallback. The focused Release matrix passes across the FeatureTestRunner hardware configurations, and the production transform benchmarks report zero allocation. Commit ef93d584511055f8e91e8d801662c44a5b0984d9 records the separately audited 12-bit inverse ADST4, Identity4, and Identity16 widening correction and closes the Phase 3 transform correctness gate.
• Core intra prediction, chroma-from-luma, palette reconstruction, and nonlinear self-guided restoration now use SIMD-first operator or packed-kernel architectures with exact scalar fallbacks. Self-guided restoration dispatches portably through Vector256, Vector128, and scalar tiers; AVX2-only gather and scan operations remain local fast paths within the 256-bit implementation. FeatureTestRunner verifies AVX2, portable AVX-only 256-bit, 128-bit, and scalar execution. No namespace, file, or type name exposes SIMD width, ISA, storage, or bit depth. Transform traversal is already SIMD-first, while normative super-resolution and Wiener horizontal products reuse ImageSharp's cross-platform adjacent multiply/add SIMD helper with exact scalar fallbacks.

AV1 encoder

• HeifEncoderCore.Encode() is synchronous, matching the ImageSharp encoder contract.
• The production switch writes only an explicitly selected legacy JPEG image item. It throws for HeifCompressionMethod.Av1 and HeifCompressionMethod.Hevc, so it produces neither AVIF nor HEIC.
• Av1FrameEncoder.Encode() is an outline of an SVT-style pipeline rather than an implementation.
• Required mode-decision, block-geometry, forward-transform, token-writing, neighbor-context, palette, intra-block-copy, transform-size, quantization/rate-control, and OBU-writing paths are absent or throw.
• Encoder tests currently cover the public option contract and legacy JPEG boundaries. There is no AV1 or HEVC payload output to cross-decode.

Legacy JPEG image items

• A JPEG item decoder exists, and the Fujifilm .hif fixture is identified and decoded as a legacy JPEG image item.
• The encoder uses the existing ImageSharp JPEG codec only when HeifCompressionMethod.LegacyJpeg is explicitly selected and emits a jpeg primary item.
• The legacy JPEG path still needs independent container interoperability coverage, but its quality, bit-depth, lossless, and chroma restrictions are explicit at the public option and codec boundaries.

Tests

• HEVC coverage includes exact native-plane comparison with HM output or normative decoded-picture hashes, exact complete-image comparison with pinned presentation references, ten official Sony GENERAL pictures across the 8/10/12-bit chroma-layout boundary, and 25 additional official Range Extensions pictures across every exposed independently coded profile and required tool. Genuine HEIC presentation covers that complete independently coded profile/tool matrix; dependent-picture decoding and encoding remain separate Phase 4 work.
• AV1 has focused bitstream, prediction, entropy, reconstruction, filter, film-grain, color, and transform coverage, plus real libavif inputs. A real two-layer libavif-derived OBU stream verifies the bounded frame lifecycle, retained-slot occupancy, resolved inter references, and frame_size_with_refs dimensions through a fake tile lifecycle. The corrected progressive color item then verifies its production single-reference inter branch against exact pinned-libaom native planes and pinned-libavif presentation. The pinned 19-frame YUV444 sequence separately reaches production equal-average compound reconstruction after every preceding reference sample and matches its final native and presented references exactly. A real palette stream truncated inside its tile entropy payload verifies libaom-equivalent overflow/trailing-bit rejection and decoder-session recovery. The committed equal-average checkpoint passes 119 direct focused cases plus three FeatureTestRunner cases, zero-warning net10.0 and net11.0 source builds, a zero-error test-project analyzer build with pre-existing repository warnings, allocator lifetime validation, and git diff --check. Four additional pinned sequences reach distance-weighted, both wedge signs, both difference-mask orientations, and smooth and wedge inter-intra reconstruction with exact final native and presented output. Commit 864b11ad9791d3254a38691246270c569001b5a1 records the selectable-compound checkpoint after both source targets, the test-project analyzer build, 25 direct focused cases, six FeatureTestRunner cases, constrained allocation, exactly-once tracked returns, and git diff --check passed. Commit 59a4fbad5ed3bc7c34331c4bdc65acc7acde3f04 adds exact traced OBMC sequence evidence, subsampled direct reconstruction, constrained allocation, both source builds, the analyzer build, 14 direct cases, the normal/scalar FeatureTestRunner comparison, and clean staged git diff --check. Commit 8f57c29b250b4291f034b31c1ea3de3fa3345467 adds a genuine size-changing layered AVIF, exact retained-base and final native planes, exact final presentation, active inter deblocking, constrained allocation, normal/scalar dispatch, direct 8/10/12-bit block reconstruction, and 87 passing focused cases. The local-warp checkpoint adds a genuine two-frame 256x256 bounded AVIF with many traced WARPED_CAUSAL blocks, exact pinned multi-sample projection, exact final native and presentation comparisons, constrained allocation, normal/scalar FeatureTestRunner dispatch, direct 8/10/12-bit predictor parity, zero-warning source builds, and 86 passing focused cases. Valid still-image syntax paths still contain explicit unsupported branches, so the independent AV1 decode matrix is not complete.
• Local-warp fixture provenance is pinned to libavif 062e582e8afda88e6baf988fdcf046a801efa0f5 and libaom 03087864cf4bea6abb0d28f95cf7843511413d8f. A deterministic two-frame limited-range YUV444 grid sequence was encoded with avifenc -j 1 -s 0 -q 60 -a color:enable-warped-motion=1 -a color:enable-global-motion=0 -a color:enable-obmc=0. SHA-256 values are 990BAC4AD443005C217B0DA4FCCFA9ADFB3AA147AD06C85F9A655A4433E9E8A7 for the AVIF, 984B2815CEE0C05FDE26430F150A21B5C993141E25BA1ED4FDE09372DB64AC13 for the pinned-libaom frame-1 Y4M, and 4490D62FB6679378E92CACA48427359091AD2106BE49FC1A3848F78BE03BEEB1 for the pinned-libavif frame-1 PNG.
• Scaled-reference fixture provenance is pinned to libavif 062e582e8afda88e6baf988fdcf046a801efa0f5 and libaom 03087864cf4bea6abb0d28f95cf7843511413d8f. The source is libavif tests/data/webp_logo_animated.y4m, encoded as two layered YUV444 inputs with avifenc -j 1 -q 90 --layered --scaling-mode:u 1/2 <input> --scaling-mode:u 1/1 <input> <output>. The resulting item contains a 701-byte 40x40 base layer followed by a 1,494-byte 80x80 dependent layer. SHA-256 values are EB239F31EC8DBF5E97AD6F52670FCA6497AE2A933822CFE724C75F66AAA2520B for the AVIF, B7E30E04A935414A517BAA2DF06AB756DA18BA7C291220D9D7C063297761AE82 for the 2,195-byte logical AV1 payload, 873DC1AB5623910FBF8053CBF684399F4F3BDDBA4110492E3E11DE86E01FEC1B for the 4,800-byte pinned-libaom base YUV, 964F4D85A79FC10EEB3722E2C9DA3415083E1EB8C2611FE154946FA55E2171ED for the pinned-libaom final Y4M, and DC4C6DBE6BD92C5FCE1E3E23700AFA603EF04ED02EDD336213EBBA1E3BD84BA0 for the pinned-libavif final PNG.
• The selected-layer derivative adds one essential lsel association selecting spatial layer zero without changing the logical AV1 payload. Its absolute iloc extent offsets move by eleven bytes with the enlarged property containers, and pinned libavif linked to the same libaom commit presents the selected 40x40 base at the item's 80x80 ispe extent. The 2,505-byte AVIF has SHA-256 E6E1F53F8DD558C5309E0C80B0635FA4EFCBB636541E4E38206985A91E0FBA66; its exact pinned-libavif PNG has SHA-256 678C5E5D4650EA6F0C590302E7DB9E3C6608851BC577453DA4A6837BDB4D3AF3.
• The AV1 transform matrix verifies scalar, Vector128, Vector256, and Vector512 forward representations plus the production inverse tiers across every valid size/type combination and supported bit depth. All 511 focused forward and inverse cases pass in Release; FeatureTestRunner isolates hardware tiers, every two-dimensional configuration exercises production dispatch, and the complete-block benchmark records zero managed allocation.
• Independent libavif fixtures cover primary, grid, auxiliary-alpha, ICC, metadata-skipping, and all-sync image-sequence presentation. Reference-dependent AV1 and HEVC sequence reconstruction and independent HEVC ICC sequence coverage remain open.
• Focused decoder-option tests cover strict, ancillary-only, image-data, and metadata-skipping behavior for still items and sequence samples. Complete adversarial dimension, allocation, malformed-container, and resource-limit coverage remains open.
• No cross-codec encode test exists for AV1 or HEVC because both production encoding branches still throw before producing a payload.

Architectural direction

Public format identities, shared internal HEIF container

Expose HEIC and AVIF as distinct public format identities backed by one internal HEIF/ISO BMFF container implementation. Provide format-appropriate decoders, encoders, metadata, options, MIME types, extensions, and generated save methods. Generic .heif and .hif input must be dispatched by brands and primary item type to HEVC, AV1, or legacy JPEG rather than by extension. Avoid a broad internal namespace rename until the container model is stable.

Name folders, namespaces, and types for the image-format behavior they own. SIMD width, ISA, sample storage, and bit depth remain internal operator or dispatch details; do not create .Simd, .LowBitDepth, width-specific, or equivalent implementation-mechanic namespaces, files, or type suffixes.

Generate the standard sync and async SaveAsHeic and SaveAsAvif overloads through the existing format templates. If a generic SaveAsHeif API is retained, its encoder options must require an explicit supported payload codec rather than infer one from an extension. Public XML documentation must describe observable behavior and option effects only.

Parsed container model

Parse the meta box into an immutable logical model before decoding payloads:

• dictionaries keyed by item ID, not collection position;
• a one-based property table with explicit association records and essential flags;
• resolved item references (dimg, auxl, thmb, and metadata relationships);
• item locations represented as validated extents and construction methods;
• typed properties for ispe, pixi, hvcC, av1C, colr, auxC, pasp, clap, irot, and imir;
• bounded views over item payloads, independent of source box order; and
• explicit primary, alpha, grid-tile, thumbnail, Exif, and XMP roles.

Use existing ImageSharp buffered stream and allocation abstractions. Do not copy the whole file into an unbounded byte array. Every length, offset, multiplication, allocation, tile count, and image dimension coming from the file is an external boundary and must be checked against the enclosing box and ImageSharp limits. Internal decode stages should rely on the validated model rather than repeat defensive checks.

Codec state and sample storage

Keep HEVC and AV1 bitstream state in separate codec implementations. Within each codec, separate parameter/sequence state, frame or picture headers, tile/slice entropy state, and reconstructed images. Still-image operations retain no reference pictures; bounded image-sequence operations retain only the reference frames required by their declared dependencies. Give each allocation one owner and a deterministic disposal point.

Represent 8-bit samples with bytes and high-bit-depth samples with unsigned 16-bit storage. Plane dimensions and strides must reflect monochrome and chroma subsampling instead of pretending every plane is full-resolution 4:4:4. Design storage and reconstruction operators for SIMD-first traversal, with the scalar fallback sharing the same layout and contract for behavioral parity.

Decode directly into ImageSharp pixels

Perform chroma upsampling, range expansion, matrix conversion, alpha composition, and pixel packing into allocator-backed row buffers or the destination frame. Use PixelOperations<TPixel> and the existing packed pixel conversion paths instead of creating an intermediate Image<Rgb24>.

Do not reuse JPEG or WebP color constants merely because those codecs already contain vectorized YUV conversion. Reuse their vector dispatch, lane handling, row processing, and pixel packing patterns only when the signaled HEVC/AV1 range and matrix semantics are preserved.

Existing ImageSharp code to reuse

Need	Reuse target	Constraint
Stream parsing	BufferedReadStream, existing endian readers, bounded decoder-core patterns	HEIF box extents remain the source of truth.
Memory ownership	MemoryAllocator, IMemoryOwner<T>, Buffer2D<T>, allocator-backed row buffers	No unbounded file-sized arrays or per-block allocations.
Pixel output	PixelOperations<TPixel>, Rgba32, Rgba64, existing packed conversion methods	Preserve alpha and high-bit-depth precision.
SIMD utilities	SimdUtils, Vector128_, Vector256_, Vector512_, Numerics helpers	Add codec-specific math only where semantics differ.
SIMD structure	JPEG color-converter factories and WebP YUV row converters	Reuse dispatch/tail patterns, not incompatible coefficients.
Decode lifecycle	JPEG/WebP decoder cores, ImageSharp cancellation and dimension-limit handling	Identify must not reconstruct pixels.
Public format API	PNG, JPEG, and WebP format/decoder/encoder/options/metadata patterns	Keep HEIC, AVIF, and generic HEIF dispatch behavior explicit.
Generated save APIs	_Formats.ttinclude and existing format templates	Generate SaveAsHeic and SaveAsAvif; require an explicit codec for generic HEIF output.
Metadata	Existing ICC, Exif, XMP, and CICP-related metadata structures where available	Preserve payloads and apply only specified transforms.
Test infrastructure	ReferenceCodec, image comparers, feature-test helpers, codec test base classes	Include independent artifacts and cross-codec tests.
Performance tests	Existing ImageSharp benchmark project and hardware-intrinsics test controls	Benchmark real decode stages and representative images.

Implementation phases

Every phase exit gate is an internal prerequisite for the next phase. Only the Phase 9 exit gate together with a fully passing verification matrix marks the PR complete.

Phase 0: establish a reproducible baseline

Tasks:

• Build the merged solution in Release and record compile errors and warnings attributable to the WIP.
• Run only the existing HEIF/HEVC/AV1 tests first, then record disabled tests and unexecuted asset coverage.
• Record the pinned official libaom, local libavif, HM, and Android libhevc revisions, then pin the independent HEIC container interoperability oracle and external fixture toolchain.
• Create a provenance map from each WIP codec file to its specification section and exact upstream source. Preserve the current SVT-AV1 origins where applicable and identify which missing paths will use libaom.
• Convert the completion boundary above into a feature matrix with unsupported, parses, decodes, encodes, and verified independently states.
• Audit every source file under src/ImageSharp/Formats/Heif. Document every type and shared contract, and add technical comments wherever the code depends on non-obvious specification syntax, fixed-point arithmetic, transform staging, entropy state, buffer layout, or SIMD behavior. Keep public XML documentation limited to observable API behavior.

Exit gate:

• The post-merge branch has a recorded Release baseline, every existing failure is classified, all upstream code origins are known, and the complete HEIF source tree passes the documentation audit before additional porting begins.

Phase 1: correct the format contract

Tasks:

• Define distinct HEIC and AVIF public format types over the shared internal HEIF container and register the correct brands, MIME types, and extensions.
• Add generated SaveAsHeic and SaveAsAvif APIs and format metadata integration through the same mechanisms as established codecs. Define generic SaveAsHeif only if its options require an explicit supported payload codec.
• Define decoder options using existing DecoderOptions behavior, including target pixel type, metadata handling, cancellation, and image-size limits.
  • Generic decode selects the caller's pixel type and the non-generic entry point defaults to Rgba32. TargetSize and Sampler are applied once after HEIF presentation composition. MaxFrames bounds retained sequence samples. SkipMetadata, SegmentIntegrityHandling, ColorProfileHandling, Configuration, and cancellation flow through the container, item, sequence, and nested-codec boundaries.
  • Complete adversarial dimension and allocation-limit coverage for still items, grids, auxiliary images, and sequence tracks before closing this contract item. Large payload and image buffers already use the configured allocator, but the complete cross-product has not been verified.
• Define codec-specific encoder options with observable semantics for quality, speed/effort, lossless mode, chroma subsampling, bit depth, alpha quality, and metadata handling. Avoid exposing internal HEVC or AV1 tuning knobs without a stable user-facing meaning.
• Make Rgba32 the default 8-bit decode output so alpha is not silently lost.
• Remove unsupported JPEG 2000, JPEG-XR, JPEG-XS, and AVC capability claims unless those payload codecs are added to the completion matrix. Retain legacy JPEG as an explicit supported HEIF image-item codec.

Exit gate:

• API review confirms that names and documented behavior match existing ImageSharp patterns and promise only the completed HEVC, AV1, and legacy JPEG HEIF payload paths. Capabilities must not be registered before their implementation reaches the final PR gate.

Phase 2: rebuild the bounded HEIF still-image reader around validated items

Tasks:

• Implement a bounded box reader supporting 32-bit, 64-bit, and to-end box sizes where allowed, with overflow-safe arithmetic and correct parent bounds.
• Accept the applicable HEIF, HEIC, and AVIF still-image brands through the major or compatible brand rules. Recognize and reject sequence/movie brands without implementing their track surface.
• Skip unknown non-essential boxes and properties. Reject unknown essential properties attached to a decoded item.
• Parse meta children independently of physical order and build the item/property/reference model described above.
• Correct item ID lookup, one-based property indices, association flag masks, full-box versions, and large IDs/offsets.
• Resolve idat and mdat item data, multiple extents, construction methods, and 64-bit offsets through bounded item streams.
• Parse and associate HEVC and AV1 configuration, dimensions, plane information, ICC/CICP color, alpha auxiliary type, grids, transforms, Exif, XMP, and thumbnails.
• Make Identify return dimensions, bit depth, color type, alpha presence, profiles, and metadata from the parsed model without decoding AV1 tiles.
• Add malformed-container tests for every external length, offset, count, ID, association, extent, and relationship boundary.

Exit gate:

• The parser resolves each current HEIC, HIF, and AVIF asset into a stable logical model, malformed inputs fail without escaping bounds or allocating attacker-controlled sizes, and Identify has reference-verified metadata and payload classification.

Phase 3: complete SIMD-first AV1 still-image reconstruction

Implement and verify in dependency order:

• OBU framing, sequence headers, frame headers, tile groups, byte alignment, and trailing bits.
  • Isolate every declared OBU payload with a zero-copy bounded span reader, advance ignored metadata and reserved units without parsing their bytes as headers, validate padding and trailing bytes, enforce primary/redundant/combined frame-header order, accept the final unsized low-overhead OBU by consuming the bounded image-item remainder, and reject lengths that cross the containing boundary. Focused malformed-input tests and independent 8/10/12-bit AVIF reconstruction pass in Release.
  • Re-audit the complete bounded parser against pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f, including ULEB128 and Annex B lengths, operating-point filtering, sequence/frame ordering, redundant-header identity, temporal delimiters, padding, byte alignment, tile-group bounds, tile sizes, and complete ordered tile coverage. AV1-ISOBMFF's final unsized low-overhead OBU remains the deliberate item-specific extension and is covered by a real fixture. Correct the stale direct tile tests to address item-relative tile bytes and make header round-trip diagnostics skip span-valued properties that reflection cannot box. Both Release source builds remain zero-warning and zero-error, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and all 119 focused framing, lifecycle, bit-reader, tiling, and official-sequence cases pass without failures or skips. Commit e6f0ef84fc8273a7d50fdfc1489f54f058a2a424 records the checkpoint.
• One coherent bounded decoder lifecycle retains parsed tile state until successful frame completion, commits frame-owned reference samples and metadata only after reconstruction, resets retained state at a new sequence boundary, and disposes tile, reference, entropy, and presentation ownership deterministically. The real two-layer lifecycle fixture proves the header/reference ownership boundary through a fake tile reader; dependent inter reconstruction remains covered by the unchecked items below.
• Tile partitioning, mode information, segmentation, delta quantization, transform-size selection, coefficient token decode, inverse quantization, and inverse transforms.
  • Match libaom's depth-first traversal and frame-edge behavior for all ten AV1 partition types. Independent 8/10/12-bit streams collectively select every terminal partition shape and contain nested block geometry that requires recursive Split traversal; their complete native planes remain byte-exact under normal hardware dispatch and the scalar fallback.
  • Verify the complete 8/10/12-bit monochrome, 4:2:0, 4:2:2, and 4:4:4 base profile matrix against pinned libaom native planes and pinned libavif presentation output. All 12 fixtures match exactly with normal dispatch, AVX-512 disabled, AVX disabled, and all hardware intrinsics disabled through FeatureTestRunner.
  • Correct monochrome plane classification, maximum-superblock-row loop-filter traversal, and identity/one-dimensional inverse-quantization-matrix selection exposed by the profile matrix. The quantization rule matches libaom's tx_type < IDTX boundary and is verified by the exact native-plane oracle rather than a tolerance.
  • Re-audit partition and transform traversal, mode and segment ordering, superblock delta-Q and delta-LF updates, coefficient EOB, reverse magnitude, sign, and Golomb parsing, inverse quantization, and inverse-transform dispatch against pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f. Correct the stale coefficient round-trip oracle to compare retained entropy-scan order and dispose its encoded owner. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and all 947 focused cases pass without failures or skips. Commit fc9320262cc08dd5daf41f231998403b804b98da records the checkpoint.
• Intra prediction, including every directional, smooth, Paeth, CFL, filter-intra, and palette case permitted by AV1.
  • Implement SIMD-first chroma-from-luma storage, 4:4:4/4:2:2/4:2:0 subsampling, rounded mean subtraction, and 8/10/12-bit prediction with exact scalar fallback and FeatureTestRunner parity.
  • Implement allocation-free SIMD-first palette reconstruction for palette sizes 2-8, transform widths 4-64, and 8/10/12-bit samples with exact scalar fallback and FeatureTestRunner parity.
  • Verify palette mode syntax and presented reconstruction with an independently encoded palette AVIF fixture.
  • Re-audit conventional intra syntax, palette and filter-intra eligibility and ordering, chroma-from-luma eligibility and alpha mapping, top-right and bottom-left availability, edge extension, filtering, upsampling, and predictor dispatch against pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f. Every directional angle, DC, Paeth, smooth, filter-intra, chroma-from-luma, and palette operator remains family-owned under the JPEG-style static operator contract. All 93 focused cases pass without failures or skips, including the exact 39-frame official all-intra native sequence, exact palette native and presentation fixtures, and the 12-profile exact matrix. The existing production implementation required no correction. Commit 42384454eb45b6107608ed33379c7a69b101d533 records the checkpoint.
• Intra-block copy for still-image intra frames.
  • Decode tile-adaptive integer displacement vectors, derive and validate spatial references, apply the inter transform sets, and reconstruct luma and subsampled chroma through allocation-free Vector512/Vector256/Vector128 operators with exact-width stores and scalar fallback. FeatureTestRunner verifies every transform width, bit-depth storage path, chroma phase, intrinsic tier, scalar continuation, and destination-padding boundary.
  • Verify displacement-vector entropy, spatial candidate ordering, wavefront legality, native 8/10/12-bit reconstruction, and presented output against independently encoded opaque intra-block-copy AVIF fixtures from the pinned generic libaom/libavif reference. The fixtures require actual intra-block-copy block selection, compare every native plane sample from retained scalar-decoder Y4M output, and compare every presented RGBA byte exactly under normal hardware dispatch and each narrower fallback configuration without a tolerance.
  • Correct libaom's 64-sample spatial-scan step threshold and ranked-reference clamping, consume intra-block-copy transform-size contexts, and compare both complete official extreme-displacement frames exactly with the pinned-libaom native YUV420 reference under normal and scalar dispatch.
• Inter-frame prediction for layered still-image items.
  • Retain completed ungrained reference planes, frame headers, frame information, and published CDF snapshots in one eight-slot owner; apply refresh flags only after successful completion; resolve full and short reference signaling against occupancy and frame-ID validity; and implement primary-reference selection and frame_size_with_refs.
  • Inherit primary-reference CDFs, segmentation features and unchanged maps, loop-filter deltas, and same-role global-motion parameters. Initialize the frame-owned per-8x8 temporal motion field and project eligible retained motion vectors in normative reference order.
  • Implement allocation-free SIMD-first translational single-reference interpolation for regular, smooth, sharp, and bilinear filters; reduced four-sample kernels; horizontal, vertical, and separable two-dimensional convolution; exact AV1 rounding; 8/10/12-bit clipping; padded reference origins; and guarded destination strides. The operator contract and concrete operator files mirror JPEG color conversion, and FeatureTestRunner verifies normal, AVX-512-disabled, AVX-disabled, and scalar execution against an independent fixed-point oracle.
  • Decode temporal segmentation prediction and consume the retained segmentation map when segmentation_update_map == 1 and temporal update is enabled.
  • Finish the simple single-reference path.
    • Own and decode the adaptive single-reference, inter-mode, DRL, normal-motion-vector, interpolation-filter, inter-intra, OBMC, and motion-mode distributions through the frame CDF lifecycle.
    • Derive the spatial and temporal reference-MV stack, select NEAREST/NEAR/NEW/GLOBAL and DRL candidates, decode NEW motion-vector differences, validate the final vector, and preserve libaom's exact duplicate and weighting behavior.
    • Verify the disputed source predicates against pinned libaom: identity GLOBALMV blocks of sufficient size omit interpolation symbols through is_nontrans_global_motion(), and the single-reference spatial extension loops stop at MAX_MV_REF_CANDIDATES (two), not the full eight-entry stack capacity.
    • Consume the exact inter-intra and binary or ternary motion-mode syntax that precedes interpolation filters, then reconstruct simple translational blocks through the retained padded reference and existing SIMD-first predictor before residual traversal.
    • Store frame-sized retained and temporal motion fields in ImageSharp allocator-owned memory with deterministic counted disposal. Allocation tracking verifies aliases, success ownership, failure unwinding, presentation and decoder-result ownership, and exactly-once final disposal.
    • Request the established contiguous allocation contract for every padded frame plane and verify complete-plane reconstruction with a constrained ImageSharp allocator without copying.
    • Complete source review of the new motion-mode and single-reference tests.
    • Pass the exact single-reference checkpoint's Release verification. The net10.0 and net11.0 source builds complete with zero warnings and errors, while the net10.0 test-project analyzer build completes with zero errors and pre-existing repository warnings. All 293 focused net10.0 syntax, CDF lifecycle, candidate, vector, interpolation, lifecycle, reconstruction, ownership, and FeatureTestRunner cases pass with zero failures or skips, including exact dependent-frame native-plane and presentation comparisons, constrained allocation, and motion-field lifetime coverage. git diff --check is clean.
    • Correct the logical progressive color payload and compare the final dependent frame's native planes and libavif presentation exactly. Pinned libaom decodes both layers from the primary item's two iloc extents, and pinned libavif produces the retained two-frame YUV444-alpha reference and final PNG. The exact production-path comparisons pass through FeatureTestRunner.
  • Decode all compound reference trees and inter modes, derive and retain paired reference-MV candidates, reconstruct both unscaled translational references, and combine them with equal-weight averaging through Vector512, Vector256, Vector128, and scalar paths without per-block allocation.
  • Verify equal-average compound reconstruction with the pinned 19-frame libavif sequence, exact pinned-libaom final native planes, exact pinned-libavif final presentation, direct 8/10/12-bit production-block tests, constrained allocation, and exactly-once motion-field disposal. Both Release source targets pass with zero warnings and errors; the test-project analyzer build passes with zero errors and pre-existing repository warnings; 119 direct focused cases and three FeatureTestRunner cases pass without failures or skips; and git diff --check is clean.
  • Decode and reconstruct selected inter-intra prediction and selectable distance-weighted, wedge, and difference-weighted compound blending through reconstructed reference planes. Direct syntax and 8/10/12-bit production tests, SIMD/scalar mask and blend tests, exact four-sequence native/presentation comparisons, constrained contiguous allocation, and exactly-once tracked returns pass. Both source targets build with zero warnings and errors; the test project builds with zero errors and pre-existing repository warnings; 25 direct focused cases and six FeatureTestRunner cases pass without failures or skips; and git diff --check is clean. Commit 864b11ad9791d3254a38691246270c569001b5a1 records the checkpoint.
  • Decode and reconstruct OBMC through above and left neighboring prediction, exact libaom masks, allocator-owned workspaces, and the existing SIMD-first masked blender. Commit 59a4fbad5ed3bc7c34331c4bdc65acc7acde3f04 records exact pinned native/presentation evidence, direct 8/10/12-bit and subsampled reconstruction, constrained allocation, both Release source builds, and normal/scalar dispatch.
  • Decode scaled references through exact Q14 scale factors, signed Q4-to-Q10 coordinate rounding, variable per-output phases and steps, regular/smooth/sharp/bilinear reduced and full filters, bit-depth-dependent two-pass rounding, scaled OBMC neighbors, and the normative 288-sample retained border without per-block allocation. The independently generated 40x40-to-80x80 layered fixture proves the base reference before retention and in the production retained session, then matches the final pinned-libaom Y/U/V planes and pinned-libavif RGBA presentation exactly. Active inter deblocking additionally verifies per-block reference/mode deltas and skipped inter/intra boundary decisions. Normal and scalar FeatureTestRunner runs, constrained contiguous planes, unique allocation identities, 8/10/12-bit production-block coverage, zero-warning net10.0 and net11.0 source builds, the zero-error test-project build with 1,013 pre-existing warnings, 87 focused cases, documentation review, and git diff --check pass.
  • Decode local warped motion through pinned integer projection and allocation-free 8x8-tiled affine filtering across byte and high-bit-depth planes. Commit 0b2db86809e26757460e719a6b59c0fe58bb0586 records exact traced native/presentation evidence, constrained allocation, and normal/scalar dispatch.
  • Decode non-translational global motion through reconstructed reference planes. Commit c5637ea0187df35b385bf43e2fe85cd955f01099 records pinned single-reference rotation/zoom syntax and reconstruction, both compound reference workspaces, 8/10/12-bit storage, exact native and presentation output, constrained allocation, balanced ownership, and normal/scalar dispatch.
  • Decode all ordinary inter modes, motion modes, and regular/smooth/sharp dual-filter pairs from the official four-frame 352x288 YUV420 motion-vector sequence. The production path includes sub-8x8 chroma assembly and allocation-free no-round compound intermediates. Exact pinned-libaom native planes, normal/scalar dispatch, constrained tracked allocation, direct compound predictor parity, zero-warning source builds, the zero-error analyzer build, Roslyn compiler diagnostics, 3,985 focused cases, documentation review, and git diff --check pass. Commit 1c58d855f70b024170ced9eb0a7005f0f9c955ad records the checkpoint.
  • Verify every connected inter mode and filter with exact independent content. Complete official coverage masks prove all ordinary inter modes, all motion modes, and every switchable regular/smooth/sharp filter pair, while the committed bounded AVIF fixtures independently prove every compound, inter-intra, OBMC, scaled, local-warp, and global-warp reconstruction branch.
  • Return an explicitly selected lower spatial layer at the item's presentation extent without rescaling retained references. The essential-lsel derivative selects the exact pinned-libaom 40x40 YUV444 base, scales native planes to 80x80 before color conversion, and matches pinned-libavif output. The allocator-owned presentation buffer uses group-safe rows, portable Vector128 two-times filtering with scalar fallback, exact 8/12-bit libyuv-shaped vectors, constrained-allocation lifetime tracking, and leaves committed final-layer and display-only film-grain results unchanged.
  • Decode the official two-spatial-layer L2T1 sequence at its default operating point. The 159,037-byte IVF has manifest-matching SHA-1 E14825F50FF845B8A6932C64CB254007A0B5E3A1 and SHA-256 D65824E8E7C19BC958DB810D728B2132B513B5A93CB5F99CA3D7995E19A226E1; its pinned-libaom Y4M has SHA-256 7A427631ECBF144F435AA4612F1201415FB1A9BCF9A67BA010AEF830B0C3AB81. All eight 1280x720 YUV420 frames match exactly under normal and scalar dispatch, and an 8 KiB constrained tracked allocator verifies balanced exactly-once returns. The existing decoder required no production correction. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and all five focused compact-sequence cases pass without failures or skips.
  • Decode the official one-spatial-layer, two-temporal-layer L1T2 sequence at its default operating point. The 48,883-byte IVF has manifest-matching SHA-1 E94687EB0E90179B3800B6D5E11EB7E9BFB34EEC and SHA-256 EADE696AB60415D476E1D2A489EC2CCCD24D5EAAADFA4559C6488B7932274C5E; its pinned-libaom Y4M has SHA-256 4012DE2D4AFD095E7BB68EAE18B50B0674781BB4971CECABC0E5471E63373ED3. All eight 640x360 YUV420 frames match exactly under normal/scalar dispatch, and an 8 KiB constrained tracked allocator verifies balanced exactly-once returns. The existing decoder required no production correction. The scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and both focused cases pass without failures or skips.
  • Decode the official two-spatial-layer, two-temporal-layer L2T2 sequence at its default operating point. The 119,907-byte IVF has manifest-matching SHA-1 32EF2F14EE9CB11A24A22934F4C065E926E5D236 and SHA-256 14257D5FD3901581DBEB88133378CD79D50EA32506ADB729909AB56B4A52AF3C; its pinned-libaom Y4M has SHA-256 1ABB981CFF76BA9557DA437B258D8A95FCA755DED8E3949D857E8388AB1D6AE3. All eight 1280x720 YUV420 frames match exactly under normal/scalar dispatch, and an 8 KiB constrained tracked allocator verifies balanced exactly-once returns. The existing decoder required no production correction. The scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and both focused cases pass without failures or skips.
  • Decode the official ten-frame active-film-grain sequence. The 28,634-byte IVF has manifest-matching SHA-1 11BB40026103182C23A88133EDAFCA369E5575E2 and SHA-256 BA3EDD82A58414F009E1C821B947EC8DE4E0420F33803CA7BFEA39AF6AEEA155; its pinned-libaom grained Y4M has SHA-256 C71958DE496A949D12E819190F240BD56010E5AD57D8A0D55903BF380A9E0C26, while pinned libaom's skip-grain output differs with SHA-256 9E0762A82E582E854AFF2DC61EBA1E48C0F1D41BBC21E5C896573DA3BB2F1200. All ten displayed 352x288 YUV420 frames match exactly under normal/scalar dispatch; active synthesis, dependent retained-reference reconstruction, and 2 KiB constrained tracked allocation with exactly-once returns pass. The fixture corrected the AV1 default loop-filter reference deltas from a swapped Golden/Backward pair to the normative [1, 0, 0, 0, -1, 0, -1, -1] order. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, 28 focused cases pass without failures or skips, documentation review is complete, and git diff --check is clean. Commit 15613b01c0bc098fd9926b886216558e0d26eff9 records the checkpoint.
  • Decode the official eight-bit monochrome sequence. The 10,000-byte IVF has manifest-matching SHA-1 A17584012187CD886B64F8CB0F35BFD8D762F9DC and SHA-256 3DCC2A3C181F74788BD9575D94F0862D9E914415AC358A4DA4ED7D904F0535B7; its ten-frame 576,000-byte pinned-libaom luma reference has SHA-256 6A4F8A68366E801D60A14C00BE6EA88A861B1B631D86DEF8A8DB65EFB6B06FA8. All ten displayed 320x180 native YUV400 frames match exactly under normal/scalar dispatch, the shared compact-sequence oracle keeps every prior YUV420 fixture exact, and a 2 KiB constrained tracked allocator verifies balanced exactly-once returns. The existing decoder required no production correction. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, all 17 official-sequence cases pass without failures or skips, documentation review is complete, and git diff --check is clean. Commit da9a46b3d72fd682ea9293af3f88fde26defb222 records the checkpoint.
  • Reconcile every completed inter-frame child with the current production path and pinned libaom 03087864cf4bea6abb0d28f95cf7843511413d8f. The connected syntax, reference-state, motion-derivation, translational, compound, inter-intra, OBMC, scaled, local-warp, global-warp, layered-presentation, film-grain, and monochrome paths contain no remaining explicit unsupported branch. Both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and all 342 focused inter-frame cases pass without failures or skips. The existing production implementation required no correction. Commit 6b3a5a5e1849de7881b929f46a5744fa3efe3fd6 records the checkpoint.
• Lossless and high-bit-depth reconstruction with correct clipping and intermediate precision.
  • Route lossless 4x4 blocks through allocation-free reversible inverse Walsh-Hadamard reconstruction for 8/10/12-bit samples, including the DC-only specialization, Vector128 production traversal, scalar fallback, exact clipping, and FeatureTestRunner parity.
  • Verify lossless syntax, inverse quantization, prediction, and presented reconstruction with independently encoded 8/10/12-bit AVIF fixtures. The tests require coded residuals with palette and intra-block copy disabled, compare every native YUV sample with the pinned generic libaom-backed decoder, and compare every presented RGBA byte with pinned generic libavif exactly under normal hardware dispatch and the scalar fallback.
• Deblocking loop filter.
  • Implement allocation-free SIMD-first 4-, 6-, 8-, and 14-tap filtering for vertical and horizontal edges in 8/10/12-bit storage through closed edge operators, with exact scalar fallback and FeatureTestRunner parity against an independent definition.
  • Verify deblocking syntax, per-block intra/inter reference and mode delta derivation, skipped-transform boundary decisions on both sides of each edge, and boundary traversal with independently encoded 8/10/12-bit AV1 samples and exact scalar-libaom planes; verify presented reconstruction and public precision with genuine AVIF containers at every supported bit depth.
• CDEF.
  • Replace the scalar helper bucket with an allocation-free SIMD-first Av1CdefFilter: vector direction analysis, 4x4/4x8/8x4/8x8 constrained filtering, byte/16-bit output operators, and one exact scalar fallback verified through FeatureTestRunner.
  • Move the immutable deblocked-plane snapshot and frame-wide direction/variance maps to allocator-owned decoder memory, preserving luma-first paired direction derivation, fixed-storage skipped-block selection, asymmetric chroma direction conversion, frame-edge sentinels, SIMD source widening, and direct frame-plane output.
  • Verify CDEF syntax, selected nonzero strengths, unit/block traversal, subsampling, frame edges, and exact native reconstruction with independently encoded active-CDEF AV1 streams across 8/10/12-bit samples and pinned scalar-libaom output.
  • Verify exact presented reconstruction with independently encoded active-CDEF AVIF containers across 8/10/12-bit samples and pinned scalar-libavif output. FeatureTestRunner covers normal dispatch, AVX-512 disabled, AVX disabled, and all hardware intrinsics disabled without a comparer tolerance.
• Super-resolution scaling.
  • Replace the per-output dot product with the libaom-shaped four-output SIMD filter, exact scalar remainder, compile-time coefficient span, direct decoder-padding source rows, and one allocator-owned overlap row shared by every plane.
  • Verify exact 8/10/12-bit native reconstruction against independently encoded active-super-resolution AV1 streams under normal SIMD dispatch and with hardware intrinsics disabled.
  • Verify exact presented reconstruction with independently packaged AVIF containers across 8/10/12-bit samples.
• Loop restoration.
  • Replace scalar self-guided restoration with the libaom-shaped summed-area, coefficient, alternating-row radius-two, radius-one, and projection pipeline using AVX2, cross-platform 128-bit vectors, and an exact scalar fallback through one semantic Av1SelfGuidedFilter contract.
  • Verify active Wiener and self-guided unit selection plus exact 8/10/12-bit native reconstruction against independently encoded pinned-libaom streams under AVX2, 128-bit, and scalar dispatch through FeatureTestRunner.
  • Verify exact presented reconstruction with independently packaged active-restoration AVIF containers across 8/10/12-bit samples.
  • Verify all sixteen self-guided parameter sets at 8, 10, and 12 bits against an independent direct-window definition, including narrow, odd, AVX2-tail, 128-bit-tail, and padded-stride processing units through FeatureTestRunner.
  • Extend independent integration coverage across restoration-unit boundary shapes, chroma subsampling layouts, and super-resolution-adjusted unit geometry.
• Frame padding and film-grain synthesis when signaled.
  • Preserve the normative Gaussian sequence, LFSR, autoregressive templates, scaling interpolation, block selection, overlap, clipping, and odd-dimension extension with allocator-owned scratch.
  • Apply noise and horizontal overlap through measured SIMD-first semantic operators, with exact scalar fallbacks and scalar serial dependencies where the algorithm prevents useful vectorization.
  • Verify exact independent pinned-libaom fixtures across 8/10/12-bit monochrome, 4:2:0, 4:2:2, and 4:4:4 layouts, full and restricted ranges, identity signaling, overlap, and odd dimensions through FeatureTestRunner.
  • Benchmark full-HD-equivalent 8/12-bit 4:2:0 application across the supported dispatch tiers and verify zero allocation.

For each SIMD-suitable item, design the data layout, operator contract, scratch ownership, traversal, and descending-width dispatch around the production SIMD path first. Implement the behaviorally equivalent scalar fallback through the same contract, then verify every supported vector width and the scalar path against normative or independent results before enabling the stage. Serial adaptive work such as entropy decoding remains scalar where data dependencies prevent meaningful vectorization. Remove constant feature-disable flags and unsupported branches only when their replacement is verified. Unsupported syntax must produce a codec-specific invalid-image error; it must never silently skip a normative reconstruction stage.

Exit gate:

• Independently encoded opaque AVIF still images, including reference-dependent layered items, reconstruct correctly across all exposed AV1 profiles, bit depths, subsampling modes, and normative image compression tools. Native planes are compared exactly with pinned libaom output, and presented pixels are compared after applying the same signaled color conversion in the reference path. The exact 12-profile native and presentation matrix passes all four normal/fallback cases; five real progressive layered native, presentation, constrained-allocation, and ownership cases pass; and all 25 official-sequence cases pass without failures or skips. Both Release source targets remain zero-warning and zero-error, the scoped Release test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and git diff --check is clean. Commit 2219edbc91edb6a2d905f1221a45207eb2719abc closes Phase 3.

Phase 4: complete HEVC still-image reconstruction

Implement and verify in dependency order:

• HEIF item-local length-delimited NAL units, hvcC, VPS, SPS, PPS, and the parameter-set and slice-header syntax needed to decode the one independently coded picture carried by a supported still-image item. The production path remains item-local and exposes no Annex B byte-stream API or access-unit/timeline abstraction. Ten official Sony GENERAL Range Extensions first pictures, five real HEIC item payloads, and five public HEIC presentation cases pass exactly in Release: 20 cases with no failures or skips.
• One bounded decoder lifecycle owns only the parameter sets, current-picture state, slice/tile entropy state, and reconstructed planes required by that image item. Every allocator-backed child constructor and the owning picture decoder now unwind unpublished rents in reverse order. Focused tests exercise every constructor allocation-failure position and a complete real HEIC reconstruction through 2 KiB allocator groups with balanced exactly-once returns; all 22 official native-plane, real-item, ownership, constrained-allocation, and public presentation cases pass without failures or skips. Both Release source targets build with zero warnings and errors, the scoped Release test-project analyzer build completes with zero errors and 1,011 repository warnings, Roslynk reports zero compiler errors, and git diff --check is clean. Commit b6ba2dbf2f90c7b0a5d9da4e9f45d045bbce1a65 records the syntax reconciliation and lifecycle correction.
• CABAC arithmetic decoding and every required context transition. Commit 35e65e9546dd78847d4de7cae082e26229b3a744 records exact pinned-HM table and algorithm comparison, 9 focused arithmetic/context cases, 17 complete production-slice and allocator-lifecycle cases, zero Roslynk compiler errors, clean scoped StyleCop and whitespace verification, and clean git diff --check.
• Coding-tree, coding-unit, prediction-unit, and transform-unit traversal across all permitted sizes and partition modes. Commit 66d1f3c41b0baeaa98726b57d5feaeba50874f16 records exact first-picture native-plane comparison for official ITU RQT_A through RQT_E, STRUCT_A and STRUCT_B, and TUSIZE_A; exact pinned-HM complete-sequence MD5 and RQT decoded-picture-hash verification; analyser-proven partition, coding, prediction, and transform-size coverage; 150 focused Release cases; a zero-warning, zero-error net11.0 source build; zero Roslynk compiler errors; clean scoped formatting and StyleCop verification; and clean git diff --check.
• Intra prediction for every luma and chroma mode, including strong intra smoothing and constrained prediction rules. Commit 9bc5b9842 records exact published native output for the official 245-combination IPRED_B_Nokia_3 picture, exact pinned-HM plane hashes for the independently coded first CIP_A_Panasonic_3 picture, all four required FeatureTestRunner configurations, 149 focused Release cases, both zero-warning source builds, zero Roslynk compiler errors, clean scoped StyleCop and whitespace verification, and clean git diff --check.
  • Implement and verify the allocation-free SIMD-first predictor primitive for all 35 modes, 4/8/16/32 blocks, and 8/10/12-bit samples.
  • Implement reconstructed-plane reference collection, substitution, asymmetric chroma-unit handling, and reference-filter selection from caller-owned availability and scratch spans.
  • Build each availability flag from reconstructed-plane and coding state, and connect the predictor to transform-unit traversal for the independently verified 8-bit 4:2:0 and 4:4:4 item paths.
• Scaling lists, inverse quantization, transform skip, every required inverse transform for independently coded pictures, range-extension precision, and lossless reconstruction. Commit 97538abd2 records exact pinned-HM native-plane verification for the five official fixtures described above.
  • Implement allocation-free SIMD-first inverse DCT for every 4/8/16/32 width and height combination, the four-by-four intra-luma inverse DST, normative intermediate and residual clipping, and saturated prediction addition for 8/10/12-bit samples. Verify normal and forced-scalar execution against fixed results and a dense scalar oracle.
  • Implement the SIMD-first inverse-quantization primitive for flat and scaling-list paths across every transform size, component matrix, prediction mode, transform-skip scaling rule, signed normalization direction, and range-extension precision path.
  • Derive effective luma and chroma quantization parameters, including bit-depth offsets and chroma-format mapping, then connect scaling-list selection and inverse quantization to transform-unit traversal.
    • Implement the immutable effective-QP value used by reconstruction, including independent luma/chroma bit-depth offsets, the normative 4:2:0 mapping plateaus, the 4:2:2/4:4:4 saturation rule, and combined picture/slice/coding-unit chroma offsets.
    • Select each transform unit's coding-unit luma QP and chroma-adjustment-list entry, then pass the derived component QP into inverse quantization.
  • Implement transform skip, coefficient rotation, implicit residual DPCM, transquant bypass, and lossless reconstruction for the independently coded Phase 4 scope.
    • Implement allocation-free SIMD-first transform-skip normalization, complete-block coefficient rotation, transquant-bypass copying, implicit intra-direction selection, and horizontal/vertical inverse residual DPCM with a scalar fallback and signed residual clipping.
    • Decode explicit inter residual-DPCM modes when Phase 5 supplies the owning dependent-picture HEVC inter-prediction path; this syntax cannot occur in the independently coded Phase 4 pictures.
  • Connect coefficient decoding, inverse quantization, transform selection, reusable scratch, and add/clip to transform-unit traversal.
    • Implement grouped diagonal/horizontal/vertical scans, last-significant and significance context derivation, coefficient levels and signs, sign-data hiding, Range Extensions bypass alignment, four persistent Rice states, and one pooled scratch owner with no per-transform allocation or stackalloc.
    • Invoke coefficient decoding and the existing reconstruction kernels from transform-unit traversal, then verify complete transform blocks from official independently decoded HEVC fixtures.
• Deblocking and sample-adaptive offset for every signaled luma/chroma and bit-depth path.
  • Verify the production still-picture path against the official DBLK_A_SONY_3, DBLK_A_MAIN10_VIXS_4, SAO_A_MediaTek_4, and SAO_A_RExt_MediaTek_1 streams. Pinned HIGH_BITDEPTH HM at commit 9c1f298659ab0cee9dc13d23d0304221575410b9 reproduces every archive's published complete-output MD5, and ImageSharp matches the retained first picture's native planes exactly across 8-bit 4:2:0, 10-bit 4:2:0, and 12-bit 4:4:4.
  • Explicitly verify active deblocking and sample-adaptive-offset signaling in the existing ten-picture official GENERAL matrix while retaining its exact published native-plane comparisons across 8/10/12-bit monochrome, 4:2:0, 4:2:2, and 4:4:4.
  • Verify the same four pictures through FeatureTestRunner with normal dispatch, AVX-512 disabled, AVX disabled, and all hardware intrinsics disabled. The official SAO picture also matches all three native planes through 2 KiB allocator groups with balanced exactly-once disposal.
  • Keep the SAO traversal on the JPEG color-converter architecture: one nested classifier contract in HevcSampleAdaptiveOffsetFilter.Operator.cs, separate nested semantic readonly struct band and edge operators, descending Vector512/Vector256/Vector128 dispatch, and one scalar tail. Commits 702739916b2a440ad2dc9bae8b680ca7fcc636c4 and 1c7292759 record the implementation, official fixtures, exact provenance, dedicated filter tests, and explicit full-profile signaling assertions. Both Release source targets build with zero warnings and errors, Roslynk reports zero compiler errors, all 175 focused HEVC cases pass without failures or skips, documentation review is complete, and git diff --check is clean.
• Implement and verify tiles, wavefront entry points, dependent slices, entry-point offset conversion across emulation-prevention bytes, and the other parallelization syntax permitted by the exposed still-image profiles. The retained first pictures from the official DSLICE_A, DSLICE_B, DSLICE_C, TILES_A, TILES_B, WPP_A, WPP_B, WPP_C, WPP_D, WPP_E, WPP_F, ENTP_A, and ENTP_C bitstreams match exact native Y/Cb/Cr planes. Pinned HIGH_BITDEPTH HM at commit 9c1f298659ab0cee9dc13d23d0304221575410b9 reproduces every archive's published complete-output MD5. The production correction follows HM's exact tile-origin and wavefront-row dependent-slice context rules and retains wavefront context across dependent segment boundaries. Direct boundary arithmetic verifies entry-point conversion across emulation-prevention bytes, while combined dependent-slice wavefront and tile cases pass through 4 KiB constrained allocator groups with balanced exactly-once disposal. All 191 focused HEVC cases pass without failures or skips, both Release source targets build with zero warnings and errors, the scoped test-project analyzer build completes with zero errors and 1,010 pre-existing repository warnings, Roslynk reports zero compiler errors, the retained fixtures use the existing LFS attributes without changing .gitattributes, and git diff --check is clean. Commit b337bce1a6931885d61ef2cda792df0a29869b8b records the checkpoint.
• Supplemental enhancement information that changes ImageSharp still-image presentation or metadata exposed by ImageSharp. Commit fcd16c862c437e3249c7e693012c717512985163 records pinned-HM syntax for display orientation, no-display, mastering-display color volume, content light level, alternative transfer characteristics, ambient viewing environment, and content color volume; complete production ordering across scaling, auxiliary alpha, codec orientation, and container transforms; exact independent RGBA comparison; malformed and conflicting metadata rejection; constrained exactly-once allocation; normal/scalar FeatureTestRunner coverage; all 195 focused HEVC cases; both zero-warning Release source builds; zero Roslynk compiler errors; clean scoped StyleCop and whitespace verification; an unchanged .gitattributes; and clean git diff --check.
  • Establish exact HM 18 native-plane conformance for four real 8-bit item payloads covering 4:2:0, 4:4:4, cropping, multiple grid positions, and wavefront entry points.
  • Establish exact independent presentation verification for single images, grids, auxiliary alpha, ICC, 4:2:0, and 4:4:4 output without weakening the exact native-plane codec oracle.
  • Establish exact native-plane conformance for 8/10/12-bit monochrome, 4:2:0, 4:2:2, and 4:4:4 across every exposed HEVC profile and required Range Extensions tool. Commit 9d7348efd6496b54cef7ec27e42352641439c400 records 25 additional official H.265.1 independently coded pictures, exact profile/tool assertions and native-plane hashes, the PCM precision and chroma-deblocking corrections, all 220 focused HEVC cases, both zero-warning Release source builds, zero Roslynk compiler errors, verified fixture hashes and existing LFS pointers without an attributes change, and clean git diff --check.

Each subsystem keeps a specification-shaped scalar oracle derived from the HEVC specification and the pinned implementation reference, while production hot paths are designed SIMD-first with scalar fallback. No valid syntax in the exposed HEIC profiles may terminate in an unsupported branch or silently omit a normative reconstruction stage.

Exit gate:

• Independently encoded HEIC files reconstruct correctly across every exposed HEVC profile, chroma format, bit depth, range-extension tool, and normative still-image compression path.

Phase 5: shared color conversion, alpha, grids, and presentation transforms

Tasks:

• Implement monochrome, 4:2:0, 4:2:2, and 4:4:4 plane access with every valid signaled chroma sample position.
  • Implement SIMD AV1 4:2:0 and 4:2:2 chroma reconstruction and downsampling, including centered/colocated reconstruction and odd image edges.
  • Implement pooled HEVC plane conversion for monochrome, 4:2:0, 4:2:2, and 4:4:4, including exact encode-coordinate and decode-interpolation coverage for all six progressive 4:2:0 sample locations.
  • Connect reconstructed HEVC pictures and complete the independently verified AV1/HEVC chroma-position matrix before closing this item.
• Implement full- and limited-range expansion for 8, 10, and 12-bit samples across every supported plane layout.
  • Implement AV1 range expansion and quantization with SIMD-first byte/ushort loaders and storers and pooled planar rows.
  • Implement the equivalent HEVC plane paths with independent luma/chroma precision, native 16-bit high-depth staging, and exact limited-range endpoint coverage.
  • Independently verify the complete HEVC range/precision/plane-layout cross-product after reconstruction is connected.
• Implement every non-reserved HEVC/AV1 color-primary, transfer-characteristic, and matrix-coefficient signaling path, including identity conversion, with correct fixed-point rounding and clipping.
  • Implement one shared H.273 matrix and transfer-function operator set in both directions for AV1 and HEVC, including identity, derived coefficients, constant luminance, SMPTE ST 2085, PQ/HLG ICtCp, IPT-C2, YCgCo-Re, and YCgCo-Ro.
  • Add independent conformance vectors for signaling combinations that libavif deliberately rejects, then verify the operator set through completed HEVC reconstruction.
• Make SIMD the default decode path for YUV range expansion, chroma reconstruction, H.273 matrix and transfer conversion, clamping, and 8/10/12-bit RGB packing. Use the established static operator pattern with Vector512, Vector256, and Vector128 dispatch followed by one scalar tail that remains the behavioral oracle.
  • AV1 uses the JPEG-style static operator contract and descending SIMD-width traversal for decode and encode, with scalar execution limited to the remaining lanes or hardware without vector support.
  • HEVC plane conversion uses the same shared static operators, descending SIMD-width traversal, JPEG byte-plane pack/unpack, pooled high-depth staging, and scalar fallback; full-HD SIMD/scalar timings are recorded above.
  • Connect the HEVC reconstruction path and record complete HEIC item and sequence measurements before closing this item.
• Apply embedded ICC profiles after codec YUV/CICP conversion and image composition through ImageSharp's existing color-profile converter. Cover preserve, convert, compact, and ignore behavior for primary items, grids, alpha-composited images, and every presented sequence frame; never substitute CICP or an assumed sRGB profile for an embedded ICC profile.
  • Primary AVIF verification passes four independently scoped cases: exact profile preservation, non-sRGB conversion with changed pixels and paired-source comparison, canonical-sRGB compaction without pixel changes, and metadata skipping. Grid, auxiliary-alpha, sequence, and HEVC cases remain open.
• Decode alpha auxiliary items as monochrome planes, validate dimensions and bit depth, and compose them without losing precision. Define premultiplication behavior from AVIF signaling and ImageSharp's pixel contract.
• Validate grid tile type, dimensions, properties, order, and canvas coverage. Compose directly into the destination frame, cropping only the permitted right and bottom tile overlap.
• Apply clean aperture, rotation, mirroring, and pixel aspect ratio according to the container property order and ImageSharp metadata/processing conventions.
• Preserve ICC, CICP, Exif, and XMP metadata using existing ImageSharp profile types. ICC preservation and ICC color conversion are separate contracts and must both be independently verified.

Exit gate:

• The complete cross-product of valid HEVC/AV1 bit depths, subsampling modes, ranges, color signaling, alpha, grids, and transforms is covered by focused vectors and representative HEIC/AVIF integration files and matches independently decoded references within a documented conversion tolerance. Tests must prove ICC conversion changes pixel values according to an independent profile transform instead of checking profile retention alone. Integer identity/lossless cases must match exactly.

Phase 6: implement a real AV1 still-image encoder

Delete the JPEG payload path from the production encoder. Keep the synchronous ImageSharp encoder contract synchronous unless the established base contract provides an async implementation; never use async void.

Implement in vertical slices that always produce a decodable AV1 bitstream:

• Convert RGB/RGBA to AV1 planes for every 8, 10, and 12-bit output, range, matrix, and monochrome/4:2:0/4:2:2/4:4:4 combination permitted by the selected AV1 profile.
  • Implement the pooled sequential RGB-to-planar pipeline, 8/10/12-bit quantization, every implemented H.273 operator, and SIMD 4:2:0/4:2:2 downsampling with odd-edge handling.
  • Complete the exhaustive format/range/matrix/transfer/primary cross-product and validate encoded planes against an independent implementation before closing this item.
• Write sequence, frame, tile-group, and metadata OBUs for a reduced still picture.
• Implement a temporary smallest-valid intra-only vertical slice using existing partition, prediction, transform, quantization, coefficient, and entropy structures.
• Complete block geometry, neighbor/context updates, transform selection and forward transforms, quantization, coefficient tokenization, and range coding.
• Add mode decision and rate/distortion selection in increasing effort levels. Reuse computed prediction and transform results instead of duplicating analysis across stages.
• Add lossless mode and validate the exact lossless constraints rather than treating quality 100 as lossless.
• Add in-loop filter decisions and signaling. A legal choice to disable an encoder feature is distinct from a decoder skipping a signaled feature.
• Add alpha as an auxiliary AV1 item with independently controllable quality where the public option warrants it.
• Write the AVIF item graph, extents, av1C, pixel information, color properties, metadata, and alpha relationships with correct brands.
• Write embedded ICC and CICP color properties independently, preserve the source profile selected by the encoder options, and verify both metadata round-trip and independently converted output pixels.
• Make output deterministic for identical pixels, metadata, options, and configuration.
• Complete the partition, prediction, transform, quantization, entropy, filter, and rate/distortion choices needed for quality and effort settings to provide a genuine compression tradeoff rather than selecting from a fixed coding subset.
• Exercise every encoder bit-depth, plane-layout, range, color, alpha, lossless/lossy, quality, and effort combination through independent decode.

Every vertical slice must be decoded by libavif before additional compression features are added. Self-round-trip tests are supplementary because matching encoder and decoder bugs can otherwise hide invalid bitstreams.

Exit gate:

• libavif and another independent AV1 decoder accept ImageSharp output across the complete encoder matrix, and ImageSharp reconstructs the same files. Lossless output is pixel-exact; lossy output demonstrates effective quality/effort tradeoffs and meets recorded quality and size expectations without malformed or non-AV1 payloads.

Phase 7: implement a real HEVC still-image encoder

Implement in vertical slices that always produce a HEVC bitstream accepted by an independent HEIC decoder:

• Convert ImageSharp pixels into every exposed HEVC bit depth, range, and monochrome/4:2:0/4:2:2/4:4:4 combination.
• Write VPS, SPS, PPS, slice headers, parameter arrays in hvcC, and a smallest-valid intra picture as a temporary vertical slice.
• Complete coding-tree partitioning, intra prediction selection, forward transforms, scaling/quantization, coefficient scanning, and CABAC encoding.
• Add real rate/distortion mode decision, quality and effort controls, lossless mode, deblocking decisions, and sample-adaptive-offset decisions.
• Add alpha auxiliary images, grids, metadata, color properties, image relationships, and correct HEIC brands.
• Write embedded ICC and CICP color properties independently, preserve the source profile selected by the encoder options, and verify both metadata round-trip and independently converted output pixels.
• Exercise every encoder profile, bit depth, chroma format, range, color, alpha, lossless/lossy, quality, and effort combination through independent decode.

Exit gate:

• Independent HEIC decoders accept ImageSharp output across the complete encoder matrix. Lossless output is pixel-exact; lossy output demonstrates effective quality/effort tradeoffs and meets recorded quality and size expectations.

Phase 8: SIMD and allocation optimization

Production hot paths are designed and implemented SIMD-first. The same architecture must provide a behaviorally equivalent scalar fallback for unsupported hardware, remainder work, and parity verification; scalar code is not a temporary production architecture that SIMD is added to later. Every supported hardware path and the scalar fallback are verified against independent specification-shaped results before the subsystem is complete.

Tasks:

• Remove known avoidable allocations first: per-transform arrays, the intermediate RGB image, repeated block scratch arrays, and file-sized buffering.
  • Complete the active AV1 forward and inverse transform-family checklist above.
• Benchmark codec-specific costs for CABAC/range decode, inverse transforms, still-image prediction, deblocking, SAO, CDEF, restoration, chroma upsampling, color conversion, alpha packing, and grid copies.
  • Add a permanent frame-wide HEVC intra-prediction benchmark. On .NET 10, SIMD planar, vertical-angular, and horizontal-angular prediction measured 155.2, 150.7, and 255.0 microseconds per padded 1920x1088 frame, compared with forced-scalar timings of 1.890, 1.081, and 0.979 milliseconds: 12.2, 7.2, and 3.8 times faster with zero managed allocations.
  • Add a permanent frame-wide HEVC reference-preparation benchmark. On .NET 10, complete and partially substituted borders measured 465.3 and 566.7 microseconds per 2,040-block padded 1920x1088 frame, compared with forced-scalar timings of 477.5 and 608.0 microseconds, with zero managed allocations.
  • Add a permanent frame-wide HEVC inverse-transform benchmark. On .NET 10, dense-coefficient 32x32 twelve-bit inverse DCT, transposition, and add/clip measured 3.643 milliseconds per padded 1920x1088 frame, compared with 45.23 milliseconds with hardware intrinsics disabled: 12.4 times faster with zero managed allocations.
  • Add a permanent frame-wide HEVC inverse-quantization benchmark. On .NET 10, dense 32x32 twelve-bit flat and scaling-list paths measured 69.98 and 238.4 microseconds per padded 1920x1088 frame, compared with forced-scalar timings of 1.469 and 1.609 milliseconds: 21.0 and 6.7 times faster with zero managed allocations. Pre-expanding the scaling matrices once reduced the SIMD scaling-list path from 4.974 milliseconds to 238.4 microseconds.
  • Add a permanent frame-wide HEVC residual-reconstruction benchmark. On .NET 10, dense 32x32 twelve-bit transform skip, horizontal RDPCM, and vertical RDPCM measured 79.64, 275.61, and 110.81 microseconds per padded 1920x1088 frame, compared with forced-scalar timings of 571.5 microseconds, 1.512 milliseconds, and 1.257 milliseconds: 7.2, 5.5, and 11.3 times faster with zero managed allocations.
  • Add a permanent frame-wide HEVC grouped coefficient-scan benchmark. On .NET 10, generating the complete diagonal 32x32 scan for all 2,040 transform blocks in a padded 1920x1088 frame measured 1.432 milliseconds with zero managed allocations. CABAC decisions remain serial and require an independently encoded complete-slice workload after transform-unit traversal is connected.
  • Add a permanent frame-wide AV1 chroma-from-luma prediction benchmark. The lane-selection run rejected the explicit Vector512 experiment because AVX2 was 2.6% faster at 8-bit and 4.8% faster at 12-bit on the test system. With that path removed, the selected hardware and forced-AVX2 paths are equivalent; the final .NET 10 run measured 88.76 and 74.89 microseconds per padded 1920x1088 frame at 8 and 12 bits, compared with forced-scalar timings of 1.594 and 1.604 milliseconds: 18.0 and 21.4 times faster with zero managed allocations.
  • Add a permanent frame-wide AV1 palette-reconstruction benchmark. On .NET 10, runtime-selected hardware measured 20.40 and 62.99 microseconds per padded 1920x1088 frame at 8 and 12 bits, compared with forced-scalar timings of 488.96 and 509.58 microseconds: 24.0 and 8.1 times faster with zero managed allocations. Forced Vector512 measured 20.81 and 62.70 microseconds, so production continues to honor the runtime's preferred vector width instead of overriding this machine's Vector256 selection.
  • Add a permanent frame-wide AV1 lossless inverse-transform benchmark. Dense four-by-four reconstruction across a padded 1920x1088 frame measured 595.6 microseconds at 8 bits and 512.4 microseconds at 12 bits, compared with forced-scalar timings of 2.074 and 2.270 milliseconds: 3.5 and 4.4 times faster with zero managed allocations.
  • Add a permanent frame-wide AV1 deblocking benchmark. Complete vertical and horizontal fourteen-tap passes across a padded 1920x1088 frame measured 4.040 milliseconds at 8 bits and 4.184 milliseconds at 12 bits, compared with forced-scalar timings of 9.416 and 9.410 milliseconds: 2.33 and 2.25 times faster with zero managed allocations.
  • Add a permanent frame-wide AV1 super-resolution benchmark. Upscaling a 1280x1080 coded luma plane to 1920x1080 measured 2.052 milliseconds at 8 bits and 2.045 milliseconds at 12 bits, compared with forced-scalar timings of 10.627 and 10.634 milliseconds: 5.18 and 5.20 times faster with zero managed allocations. BenchmarkDotNet could not change the Windows power plan, so retain these figures as an uncontrolled local baseline rather than a release comparison.
  • Add a permanent full-HD-equivalent AV1 self-guided restoration benchmark. AVX2 measured 6.674 milliseconds at 8 bits and 7.076 milliseconds at 12 bits, compared with 11.425 and 11.119 milliseconds for the 128-bit path and 40.386 and 40.522 milliseconds for the scalar fallback: AVX2 is 6.05 and 5.73 times faster than scalar with zero managed allocations. BenchmarkDotNet could not change the Windows power plan or query the processor name, so retain these figures as an uncontrolled local baseline rather than a release comparison.
• Implement vector paths only for confirmed hot loops, using existing Vector128, Vector256, and Vector512 helper and dispatch patterns where supported.
• Prioritize shared color conversion and pixel packing, chroma upsampling, inverse-transform add-and-clip, intra predictors, HEVC deblock/SAO, AV1 loop filter/CDEF/restoration, and contiguous grid copies.
• Benchmark the complete decode color pipeline on representative 8/10/12-bit AVIF and HEIC images with and without embedded ICC profiles. Report absolute end-to-end timings and allocations in addition to the isolated YUV/CICP and ICC stage costs.
• Design each hot-path operator and data layout from the applicable upstream SIMD algorithm, mapping lane width, signedness, intermediate precision, rounding, saturation, edge extension, and high-bit-depth behavior before implementation.
• Implement the scalar fallback through the same operator contract, tables, constants, scratch ownership, and traversal as the SIMD paths so it remains a specification-shaped oracle without becoming a separate architecture.
• Test scalar and each available hardware path with intrinsics explicitly enabled and disabled, including widths shorter than a vector, exact-vector widths, non-multiples, edges, maximum sample values, and high-bit-depth overflow cases.
• Remove dead or commented SIMD experiments once a verified production path replaces them.

Exit gate:

• Benchmarks show a material improvement on representative AVIF files, allocation measurements meet an agreed budget, and every vector path is behaviorally identical to the scalar path for integer reconstruction or within the documented color-conversion tolerance.

Phase 9: hardening, documentation, and release readiness

Tasks:

• Fuzz the box parser, AV1 OBU parser, HEVC NAL/parser, entropy decoders, and dimension/allocation boundaries using the same safety expectations as established ImageSharp codecs.
• Test seekable and non-seekable streams, short reads, cancellation, truncated data, unknown optional boxes, unknown essential properties, oversized dimensions, malicious counts, and offset arithmetic overflow.
• Run the focused HEIF/HEIC/AVIF suite after every final codec edit, then the full ImageSharp suite in Release.
• Build all supported target frameworks and run packaging/API compatibility checks used by the repository.
• Update public documentation, format tables, MIME/extension lists, samples, and THIRD-PARTY-NOTICES.TXT.
• Remove placeholder images, legacy JPEG-in-HEIF production paths, stale TODO-only code, disabled tests that now have coverage, and unsupported capability claims.

Exit gate:

• The full Release build and test matrix passes, every valid HEVC/AV1 color/compression/bit-depth entry is implemented, independent HEIC and AVIF interoperability is recorded for the complete feature matrix, the public API has been reviewed, provenance and patent/license obligations are complete, and no advertised feature depends on a placeholder, narrow temporary subset, silent fallback, disabled normative stage, or unsupported valid syntax branch.

Verification matrix

Every valid combination in the HEVC and AV1 profiles exposed by the final public contract needs focused coverage and representative integration coverage. Each axis needs independently produced HEIC/AVIF inputs and ImageSharp-produced outputs; pairwise and targeted cross-product cases must cover interactions where exhaustive media fixtures would be redundant.

Status	Area	Required coverage
- [ ]	Container/payload	HEIC with HEVC, AVIF with AV1, and generic HEIF/HIF with supported HEVC, AV1, or legacy JPEG items; detection by brands/items.
- [ ]	Bit depth	Every bit depth permitted by the exposed HEVC and AV1 profiles, including 8, 10, and 12-bit decode and encode.
- [ ]	Planes	monochrome, 4:2:0, 4:2:2, and 4:4:4 decode and encode in every valid HEVC/AV1 profile and depth combination.
- [ ]	Range	full and limited decode and encode.
- [ ]	Color	every valid non-reserved color-primary, transfer-characteristic, matrix-coefficient, and chroma-position signaling path; identity RGB; ICC; CICP defaults and overrides.
- [ ]	Alpha	opaque, binary, gradient, different alpha quality, high bit depth, malformed relationship.
- [ ]	Structure	single item, multiple extents, idat, mdat, grids with cropped edge tiles, metadata items.
- [ ]	Transform	pasp, clap, irot, imir, and valid combinations.
- [ ]	AV1 decode tools	every normative transform type/size, still-image predictor, partition, palette, segmentation, quantization, entropy/context, lossless, deblock, CDEF, super-resolution, restoration, and film-grain path valid in independently decodable AVIF still-image items.
- [ ]	AV1 encode compression	real mode decision and rate/distortion selection across partitions, predictions, transforms, quantization, entropy coding, filters, lossless/lossy quality, and effort settings; no permanent fixed coding subset.
- [ ]	HEVC decode tools	every normative NAL/parameter/slice, CABAC, coding-tree, intra prediction, transform, quantization, range-extension, lossless, tile/wavefront, deblock, and SAO path valid in independently decodable still-image items for the exposed HEIC profiles.
- [ ]	HEVC encode compression	real coding-tree, prediction, transform, quantization, CABAC, filter, and rate/distortion decisions across lossless/lossy quality and effort settings; no permanent fixed coding subset.
- [ ]	Streams	file, memory, non-seekable, short-read wrapper, cancellation.
- [ ]	Failure	truncation at every box/OBU layer, invalid sizes/offsets/counts, unknown essential properties, unsupported profile.
- [ ]	Interop	libavif and libaom for AVIF/AV1; a pinned independent HEIC container and HEVC codec implementation for HEIC/HEVC.
- [ ]	SIMD	scalar, 128, 256, and 512-bit paths where supported; tails and edge blocks.

Reference outputs must be versioned artifacts or generated by a pinned reference command whose exact tool version and arguments are recorded. Do not use ImageSharp's own decoder to establish the expected pixels for its encoder, and do not replace final-image assertions with internal buffer or non-zero checks.

No valid HEVC or AV1 color, compression, or bit-depth row may remain unsupported, partially implemented, disabled, or deferred when the PR is marked complete.

Working rules for implementation

• Keep changes vertical and reviewable. A slice should add one behavior, its focused tests, independent evidence, and any required notice update.
• Keep every AV1 prediction family on the established JPEG color-converter operator architecture. Each distinct traversal contract owns a family-named predictor type; its .Operator.cs defines the static interface, and its family-named files own the closed generic widest-to-narrowest SIMD traversal. Semantic readonly struct operators implement scalar, Vector128, Vector256, and Vector512 arithmetic through that contract. Only modes which share the same traversal and contract may share a predictor family; do not nest a separate predictor beneath a broad intra/inter family or create hardware-width-specific class hierarchies.
• Keep AV1 one-dimensional transforms on the JPEG color-converter owner pattern. Av1ForwardTransformer.Operator.cs defines the nested forward contract, and each Av1ForwardTransformer.<Semantic>Operator.cs file defines one nested semantic readonly struct with concrete scalar and SIMD bodies. Av1Inverse2dTransformer owns the inverse contract and semantic operators in the same layout, while Av1InverseTransformer owns its byte and high-bit-depth reconstruction-output operators. Shared lane arithmetic belongs in explicitly named Operations types, never in forwarding semantic operators. Do not create top-level transform operators, Forward1d or Inverse1d wrapper types, generic semantic forwarding cores, CRTP contracts, or hardware-width-named operator types.
• Design SIMD-suitable codec work SIMD-first. Establish vector-friendly storage, operator boundaries, scratch ownership, traversal, every applicable lane width, and benchmark-gated dispatch before implementing the equivalent scalar fallback; never build a scalar production architecture and bolt SIMD onto it later.
• Inspect every owning method and upstream invariant before adding guards. Validate external file data at the parser/model boundary and rely on those established invariants internally.
• Do not extract one-use helpers merely to label code. Extract shared primitives only when they have genuine reuse or remove substantial complexity.
• Put comments at the points where HEVC/AV1 rounding, edge extension, context propagation, or SIMD lane behavior is not evident from the code. Comments should explain why the algorithm has that shape and identify the normative rule and pinned implementation reference.
• Use observable behavior only in public API documentation.
• Do not use reflection, built-assembly probing, native runtime fallbacks, fabricated images, or self-round-trip-only evidence.
• Build and test in Release configuration.

Recommended implementation order

The dashboard and immediate execution queue define the remaining critical path. In phase terms, work proceeds as follows:

• Finish the Phase 8 checkpoint for the implemented AV1 Vector512 transforms, including Release, feature-isolation, and benchmark evidence.
• Finish the complete focused Release matrix, final source/documentation review, and checkpoint implementation for the simple single-reference inter slice. Commit 8b9aa4b2cfe59ff0f8c348d9bb06678a5ff12f2a records the corrected real dependent-frame AVIF, exact pinned-libaom native-plane and pinned-libavif presentation comparisons, and 293 focused cases passing with zero failures or skips.
• Implement compound reference selection and equal averaging with paired reference-MV derivation, allocation-free SIMD/scalar parity, and exact independent native/presentation evidence through a reference-dependent bounded AVIF sequence. Commit e1a764b8f2e8a8db98a003fb3b913213f40084db records the checkpoint.
• Implement and independently verify selected inter-intra prediction and selectable distance-weighted, wedge, and difference-weighted compound blending through exact pinned native-plane and presentation comparisons, direct 8/10/12-bit production coverage, constrained allocation, and SIMD/scalar parity.
• Implement and independently verify OBMC through exact traced native-plane and presentation evidence, direct 8/10/12-bit and subsampled production coverage, constrained allocation, and normal/scalar dispatch. Commit 59a4fbad5ed3bc7c34331c4bdc65acc7acde3f04 records the checkpoint.
• Implement and independently verify scaled references through pinned fixed-point oracles, direct 8/10/12-bit production reconstruction, a genuine 40x40-to-80x80 bounded AVIF, exact retained-base/final-native/final-presentation comparisons, active inter deblocking, constrained allocation, and normal/scalar dispatch. Commit 8f57c29b250b4291f034b31c1ea3de3fa3345467 records the checkpoint.
• Implement and independently verify local warped motion through pinned projection, direct 8/10/12-bit predictor coverage, a genuine traced two-frame bounded AVIF, exact final native and presentation comparisons, constrained allocation, and normal/scalar dispatch. Commit 0b2db86809e26757460e719a6b59c0fe58bb0586 records the checkpoint.
• Implement and independently verify non-translational global prediction with a genuine traced bounded sequence, exact native and presentation comparisons, constrained allocation, normal/scalar dispatch, and direct 8/10/12-bit compound production coverage. Commit c5637ea0187df35b385bf43e2fe85cd955f01099 records the checkpoint.
• Complete progressive color and auxiliary-alpha verification through exact native and presentation comparisons, every required FeatureTestRunner fallback, and constrained tracked allocation.
• Correct the audited 12-bit inverse ADST4, Identity4, and Identity16 arithmetic through exact pinned boundary vectors and SIMD/scalar feature isolation.
• Remove every other unsupported valid AV1 still-image syntax path and prove the complete AVIF decode matrix with independent inputs and scalar/SIMD parity. Commit 2219edbc91edb6a2d905f1221a45207eb2719abc closes Phase 3.
• Close Phase 4 with genuine HEIC presentation evidence across every exposed profile and required Range Extensions tool while retaining the exact native-plane matrix.
• Current: close Phase 5 and the decode portion of the bounded sequence ledger: color, ICC, alpha, grids, presentation transforms, reference-dependent samples, and complete animated AVIF/HEIC decode.
• Close the still-image portions of Phases 0, 1, and 2 that remain as release gates: documentation, provenance, public format boundaries, API review, parser hardening, and malformed-input coverage.
• Complete Phase 6 with a real AV1 encoder and independent AVIF cross-decode across every public option and metadata combination.
• Complete Phase 7 with a real HEVC encoder and independent HEIC cross-decode across every public option and metadata combination.
• Complete animated AVIF and HEIC writing only after both still payload encoders pass their independent exit gates.
• Close Phase 8 end-to-end performance and allocation budgets for every completed subsystem, then close Phase 9 and the full verification matrix.

The phase numbers describe technical workstreams; they are not permission to work out of order. SIMD, allocation control, documentation, and independent evidence are completion requirements inside each active workstream. The PR remains incomplete until every dashboard stage and verification-matrix row is complete.