open-standards/wai/research/media-codec-registry.md

Media-Codec Registry — immersive audio, cinema, volumetric, haptics, HDR, broadcast and streaming

Reference for the capabilities WAI dispatches to beyond the flat-media floor (PNG/FLAC/zstd) and modern menu (AVIF/JXL/Opus/AV1) and the neural bench. These cover object/scene audio, the theatrical mezzanine path, 3D-geometry compression, the standardized haptics codec, HDR dynamic metadata, and the broadcast and streaming video / audio bitstreams distribution chains carry.

Registration ≠ a bundled decoder. As everywhere in WAI, registering a capability fixes the capability string and says "payload = the named standard's canonical bytes; the decoder is sink/library-supplied." The point is one envelope that can name any of these and fall back cleanly — not re-implementing them. The "ingest, not compete" posture: dispatch to the field's open reference where one exists, flag the proprietary/SEP-pool ones honestly, and never make a non-floor capability a default.

Royalty key: RF = royalty-free · PROP = proprietary license · POOL = patent-pool / SEP royalties apply. Status checked against primary sources June 2026, and September 2026 for the broadcast and streaming rows. The footnotes give each row's basis and residual caveats; two broadcast rows are classifications rather than verified statuses — AVS3 (¹⁰: the fee schedule is not published) and EVC Baseline (⁹: royalty-free by design, with no formal guarantee).

License classification is enforceable, not just documented

Every registered capability carries an IP class in the registry — codecs::capability_license(cap) -> LicenseClass ∈ {RoyaltyFree, PatentPool, Proprietary, NonCommercial}, with is_royalty_free(cap) as the predicate. It is a property of the capability, looked up in the registry — deliberately not a per-file manifest claim, because a codec's licence can't be honestly asserted in an envelope's own header. A deployer audits a title or a whole catalogue in code: "encode only to royalty-free capabilities", or "flag every pool/proprietary capability in this library." That makes the per-title licensing choice a property of the format, checkable in code. wai capabilities prints the class per row (the registry reports the royalty-free count outright).

The classification is exhaustive, with no default class. Every registered capability sits under an explicit arm of capability_license; one registered without a class returns None, and licence_classification_is_exhaustive_with_no_default fails. That test also pins the full non-royalty-free set, so adding or reclassifying a pool / proprietary / non-commercial capability is a reviewed diff. (Until September 2026 the function ended in a royalty-free fall-through, so a newly registered pool codec would have been classified royalty-free by omission.) defaults_are_royalty_free_and_never_pool_or_proprietary holds the other half of the guarantee: the mandatory floor and the recommended set (codecs::mandatory_floor() / recommended_set(), equal to the native sink set) are all royalty-free, and no pool, proprietary or non-commercial capability is among them.

Replicate class: which reconstructions are byte-identical across decoders

codecs::capability_replicate_class(cap) answers a second registry question: what does every conforming sink reconstruct? For a standard codec the answer is what its named standard's decoder conformance fixes, not anything WAI adds; the per-standard reading is in Decoder conformance below. Two rules keep the answer honest:

• Exhaustive, with no default. Every registered capability sits under an explicit arm; one registered without a class returns None, and the registry tests reject it. Until September 2026 the function ended in a StandardDefined fall-through, which gave AAC, Opus, JPEG and the other tolerance-conformance codecs — and wai.mesh.glb and wai.image.jpegai — a same-bytes guarantee nothing had established.
• An exact class needs recorded evidence. StandardDefined promises the same bytes on every conforming sink, so a capability enters it only with a basis: a conformance clause that requires identical output, or the WAI extension whose *-equivalence pins the reconstruction. The registry tests pin one basis per member (STANDARD_DEFINED_EVIDENCE), and a StandardDefined capability without one fails standard_defined_rests_on_recorded_evidence. Without that evidence a standard codec is StandardUnclassified, the conservative class, which promises no byte identity.

is_cross_decoder_exact(cap) is true for the classes whose replicate_promise is ByteExact (StandardDefined, NeuralIntegerExact), except a capability whose output is computed from other envelopes' decoded output (composes_decoded_sources: wai.audio.scene, whose mixdown is exact given its sources' PCM). It is the one-call audit for "a hash over the decoded output verifies at another conforming sink", answered from the capability alone; for a scene, is_cross_decoder_exact_with_sources answers once its sources' capabilities are known, and answers false when it cannot tell (no source given, or a source outside wai.audio.*). It covers the output the standard or extension specifies — for a video codec the cropped decoded pictures, before colour conversion — not a conversion a sink applies afterwards.

Measured with tools/wai_replicate_probe.sh on an arm64 macOS laptop (18 logical cores, macOS 27.0) with ffmpeg 9.0.2, September 2026. Each format was encoded from a synthetic source: 20 frames of 320×240 testsrc2 video, a 1 s 440 Hz 48 kHz mono tone, and a 320×240 grayscale still, so that no colour conversion enters the comparison. Each was decoded by every independent decoder present, and the raw outputs were compared byte for byte. VideoToolbox decodes were forced onto hardware surfaces; a software-only format under the same flags fails rather than falling back.

capability	decoders	result
wai.video.avc (libx264, Annex B)	ffmpeg h264 · VideoToolbox hardware	identical
wai.video.hevc (libx265, Annex B)	ffmpeg hevc · VideoToolbox hardware	identical
wai.video.vp9 (libvpx, IVF)	ffmpeg vp9 · libvpx · VideoToolbox hardware	identical
wai.video.av1 (SVT-AV1, IVF)	dav1d · VideoToolbox hardware	identical
wai.audio.flac	ffmpeg · libFLAC · Core Audio, against the source PCM	identical
wai.audio.aac (ffmpeg aac, ADTS)	ffmpeg aac · ffmpeg aac_fixed · AudioToolbox	differ: 238 of 49 152 samples (max 1 LSB) float vs fixed; 8 629 (max 23 LSB) against AudioToolbox
wai.audio.opus (libopus, 64 kb/s)	ffmpeg opus · libopus	differ: 13 of 48 000 samples, max 1 LSB
wai.image.jpeg (cjpeg q75, grayscale)	libjpeg-turbo accurate-integer IDCT · its float IDCT · ffmpeg mjpeg	differ: 169 and 673 of 76 800 samples, max 1
wai.image.jp2, irreversible 9/7 (OpenJPEG, 20:1)	OpenJPEG · ffmpeg jpeg2000	differ: 67 of 76 800 samples, max 1
wai.image.jp2, reversible 5/3	OpenJPEG · ffmpeg jpeg2000, against the source	identical (the lossless path)

One stream per format on one host is a demonstration, not a survey, and it is not the basis of any class: the classes rest on the conformance clauses in Decoder conformance below. The measurement is a falsification check. An identical result on a StandardDefined row is what its standard requires; a difference there would contradict the class, and the probe exits non-zero (a perturbed hardware decode run through the probe does exactly that). A StandardTolerance or StandardUnclassified row may match on a given stream; the class permits it. AV1 was measured identical here, but it is StandardTolerance: its specification requires identical output only from streams that signal no film grain. AAC is StandardUnclassified: its decoders differ here, and its conformance document (ISO/IEC 14496-26) was not read, so the registry does not say whether that difference is within its criterion.

Immersive / object-based audio

capability	standard	royalty	reference	tier
wai.audio.iamf	IAMF (AOMedia Immersive Audio Model & Formats)	RF	libiamf (Apache-2.0)	Lossless¹
wai.audio.ac4	AC-4 (ETSI TS 103 190)	PROP	single licensor	Lossy
wai.audio.ivas	3GPP IVAS (Immersive Voice & Audio Services, EVS-based)	POOL²	3GPP	Lossy

The floor (wai.audio.{opus,flac}) + the deterministic object mix (wai.audio.scene) cover stereo and the bit-exact mixdown; these add the standard immersive carriers. IAMF is open, royalty-free, from the Alliance for Open Media, and codec-agnostic (it carries Opus / FLAC / LPCM for channel, scene/ambisonic, and binaural audio), which fits "name the capability" directly. AC-4 is registered as a licence-flagged slot so object-audio content already encoded in it can ride inside a WAI envelope. IVAS is the conversational/telepresence immersive codec.

¹ IAMF can carry lossless codecs (FLAC/LPCM) → lossless ceiling. Verified RF: published under the AOMedia Patent License 1.0 ("no-charge, royalty-free, irrevocable"); reference under BSD-3-Clause-Clear. (Deployed in streaming and consumer devices in 2025.) ² Verified POOL: IVAS is in a voice-codec patent pool (launched Dec 2024, alongside EVS) — not royalty-free.

Cinema mastering / intra-frame mezzanine

capability	standard	royalty	reference	tier
wai.image.jp2	JPEG 2000 still (ISO/IEC 15444-1; DCP via SMPTE ST 429 / MXF)	RF core³	OpenJPEG (BSD-2)	Lossless
wai.video.j2k	Intra-frame JPEG 2000 video (each frame an independent codestream)	RF core³	OpenJPEG	Lossless
wai.image.jxs	JPEG XS (ISO/IEC 21122), one codestream	POOL¹⁶	an open encoder and decoder (BSD-2-Clause-Patent); a second open decoder (MIT)	Lossy
wai.video.jxs	JPEG XS codestreams, one a frame, in wai.video.av1's framing (RTP: RFC 9134; SMPTE ST 2110-22)	POOL¹⁶	as above	Lossy

The theatrical interchange path: a DCP is JPEG 2000 in MXF. JPEG XS is the low-latency contribution path of live production over IP (registered 2026-10-22); its decoders conform at a strict point (bitwise identical output) or a relaxed one (PSNR bounds), ISO/IEC 21122-4:2022 §6.3 and Annex B.2, and WAI's registration requires the strict one (format-probes §3.6, where the cross-decoder measurement is). The proprietary production mezzanine formats are out of scope for a royalty-free floor (a licence-flagged slot could be added if a workflow needs one).

³ Verified RF (Part 1): JPEG 2000 Part 1 (the core coding system) is licensable fee-free under ITU-T patent policy 2.1 — 20+ organisations agreed to license essential IP for the standard without royalties. Residual caveats: the committee never gave a formal guarantee, some extensions may carry fees, and a 2004 note acknowledged possible undeclared "submarine" patents. Safe for the DCP/mezzanine Part-1 path; don't extend the RF claim to Part 2+.

Volumetric / 3D geometry compression

capability	standard	royalty	reference	tier
wai.mesh.draco	Draco (open-source library) mesh + point-cloud geometry	RF (Apache-2.0)	draco	Lossless⁴
wai.pc.gpcc	MPEG G-PCC (ISO/IEC 23090-9), sparse / LiDAR clouds	POOL¹⁴	mpeg-pcc-tmc13	Lossy
wai.pc.vpcc	MPEG V-PCC (ISO/IEC 23090-5), dense / dynamic clouds	POOL¹⁴	mpeg-pcc-tmc2	Lossy
wai.splat.sog	Gaussian-splat compression — a candidate: no payload format pinned (registered for the Self-Organizing Gaussians family, SOG / SOGS)	POOL, provisional: classified conservatively, since the licensing of a format not yet pinned is not established	none cited	Lossy

These compress the heavy asset clouds the existing capabilities already ship (wai.mesh.glb, wai.splat.spz, wai.splat.4d). wai.splat.sog is a candidate. The string was registered in June 2026 for the Self-Organizing Gaussians family of splat formats before any payload format was pinned, and none is pinned yet, so SPEC §5 (Candidate capabilities) makes it unemittable: an encoder MUST NOT emit it and a sink treats it as unsupported. It pins when a revision cites a published specification of the format and states its framing. It was registered royalty-free, but no licence of a format specification or of reference code was read, so that class was never established; it is now classified patent-pool, conservatively and provisionally, until a revision pins its payload format and the format's licensing is read. No MPEG page or document read for this registry names Self-Organizing Gaussians, SOG or SOGS (MPEG's Gaussian-splat and point-cloud work, below).

⁴ Draco mesh geometry has a lossless mode (exact-geometry ceiling, like wai.mesh.glb); its quantization knobs are the encoder's choice.

¹⁴ No licensing programme for G-PCC or V-PCC was read. Both are MPEG standards with patent declarations, and the class is a conservative classification, as footnote ⁵ records for MIHS: not asserted royalty-free.

MPEG's Gaussian-splat and point-cloud work (read 2026-09-30)

What MPEG's public pages and public output documents said on 2026-09-30, for the three volumetric registrations above. The last column says what was read. The meeting pages were read through a tool that returns an extraction of each page; the output documents are the public files the Exploration 45 page links, downloaded from mpeg.org and read in full. The draft call for proposals (ISO/IEC JTC 1/SC 29/WG 4 N 836) and the amendment texts (WG 7 N 1453 and N 1454) are not public (HTTP 404), and iso.org refused automated reads (HTTP 403), so no ISO catalogue stage or edition is confirmed here.

item	stage and what the source states	read
Exploration 45, Gaussian splat coding (edition 1, "Implicit Neural Visual Representation")	"ongoing"; documents run from MPEG 150 to MPEG 155; the page names no working group and no coding technology beyond "GSC"	https://www.mpeg.org/standards/Explorations/45/
Who works on it	"Gaussian Splat Coding standardization is coordinated across WG 4 (Video Coding), WG 5 / JVET, and WG 7 (3D Graphics and Haptics)"; documents for the joint exploration at the next meeting "are to be managed by WG 5"	ISO/IEC JTC 1/SC 29/WG 4 N 837, FAQ on Gaussian splat coding (MPEG 154), https://www.mpeg.org/wp-content/uploads/mpeg_meetings/154_SantaEularia/w26333.zip
The call for proposals	N 837: "MPEG approved the first draft Call for Proposals (CfP) for Gaussian Splat Coding". N 422: "MPEG WGs are currently preparing a Call for Proposals (CfP) covering coding technology for static and dynamic objects and scenes represented as I-3DGS, where I-3DGS denotes per-frame 3DGS content in the original INRIA format"; test material is wanted "before October 15th 2026", for MPEG 156	WG 4 N 837; WG 5 N 422, Call for content for Gaussian splats coding (MPEG 155), https://www.mpeg.org/wp-content/uploads/mpeg_meetings/155_Geneva/w26655.zip
I-3DGS and A-3DGS	I-3DGS "Refers to point clouds represented in the “INRIA” format"; A-3DGS "Encompasses alternative Gaussian splat representations"	WG 4 N 837; WG 2 N 531, Gaussian splat coding use cases (MPEG 155), https://www.mpeg.org/wp-content/uploads/mpeg_meetings/155_Geneva/w26708.zip
Anchors	a G-PCC anchor coded with the test model of "ISO/IEC 23090-9:2023"; a V-PCC anchor on "the V-PCC GS toolset and RecGS reconstruction profile of ISO/IEC 23090-5"; a video-based anchor that "uses packing of Gaussian splat positions and attributes to video planes that are encoded as separate videos using HEVC (HM)". An anchor "does not predetermine the architecture of a future GSC standard" (N 837)	WG 4 N 826 (MPEG 154), https://www.mpeg.org/wp-content/uploads/mpeg_meetings/154_SantaEularia/w26322.zip, and WG 4 N 751 (MPEG 152), https://www.mpeg.org/wp-content/uploads/mpeg_meetings/152_Geneva/w25782.zip, both titled Common test conditions for Gaussian splat coding; WG 4 N 837
V-PCC amendment (ISO/IEC 23090-5)	Committee Draft Amendment at MPEG 153, promoted by "MPEG 3D Graphics and Haptics Coding (WG 7)", defining "a V-PCC profile specialized for Gaussian Splats"; Draft Amendment at MPEG 154	https://www.mpeg.org/153rd-meeting-of-mpeg/; WG 4 N 837
G-PCC amendment (ISO/IEC 23090-9)	"the first amendment to Geometry-based Point Cloud Compression, ISO/IEC 23090-9", at Committee Draft Amendment at MPEG 154	https://www.mpeg.org/154th-meeting-of-mpeg/; WG 4 N 837
Both amendments	"The AMD1 amendments to both standards introduce explicit support for Gaussian Splat parameters in their respective syntaxes and reconstruction processes." The V-PCC amendment adds a reconstruction profile because "Without a clearly specified reconstruction profile, different implementations could decode the same bitstream into different Gaussian Splat parameters"	WG 4 N 837
NNC amendment (ISO/IEC 15938-17)	a Committee Draft of an amendment at MPEG 154, extending NNC to "parameters of 3D and 4D scene representations such as Gaussian splats"	https://www.mpeg.org/154th-meeting-of-mpeg/; 2030 landscape §6.4
Enhanced G-PCC, ISO/IEC 23090-38	FDIS at MPEG 155; "a new generation of geometry-based point cloud compression specifically designed for dynamic point cloud sequences"; "While the original G-PCC standard (MPEG-I Part 9) primarily compresses point clouds on a frame-by-frame basis, the Enhanced G-PCC introduces a temporal coding framework". Neither page says whether a 23090-9 decoder decodes it	https://www.mpeg.org/155th-meeting-of-mpeg/; https://www.mpeg.org/standards/MPEG-I/38/
MPEG-I Part 40, conformance and reference software for Enhanced G-PCC	Committee Draft at MPEG 152, for "the enhanced edition of the Geometry-based Point Cloud Compression (E-G-PCC) standard"	https://www.mpeg.org/152nd-meeting-of-mpeg/
ISO/IEC 23090-41, G-PCC for solid objects	Committee Draft at MPEG 154; it "complements the original G-PCC standard, ISO/IEC 23090-9, which was primarily designed for sparse point clouds"	https://www.mpeg.org/154th-meeting-of-mpeg/
ISO/IEC 23090-5, fourth edition	FDIS at MPEG 152, "allowing V3C to also serve as the structural foundation for V-DMC"; before it, "V3C could be instantiated for V-PCC (Video-based Point Cloud Compression) and MIV (MPEG Immersive Video)"	https://www.mpeg.org/152nd-meeting-of-mpeg/

What this settles. (a) No page or document above names Self-Organizing Gaussians, SOG or SOGS, and MPEG's call concerns content in a representation (I-3DGS), not a chosen coding technology: wai.splat.sog names no MPEG work item. Whatever MPEG standardises for Gaussian splats takes a capability string of its own when it is published; wai.splat.sog cannot be repurposed for it (SPEC §8). (c) Enhanced G-PCC is a separate part with its own conformance part, and no source read says a 23090-9 decoder decodes it, so it is not a wai.pc.gpcc payload (SPEC §5, Point clouds); a MINOR revision may register it under its own string once 23090-38 is published and its framing is pinned, and likewise 23090-41. What wai.pc.gpcc and wai.pc.vpcc carry is open item (i).

Standardized haptics

capability	standard	royalty	reference	tier
wai.haptic.mihs	MPEG-I Haptics (ISO/IEC 23090-31, Jan 2025)	flag⁵	MPEG RM	Lossy

Complements the native wai.haptic.signal (the keyframe-envelope, Fx-floor, sample-equivalence path). MIHS is the ISO standard codec: human-readable HJIF (JSON) + a binary MIHS stream (~8–16 kbps). ISO/IEC 23090-33 defines bit-exact decoder verification + valid/invalid conformance datasets — the conformance-vector philosophy already in the standard, so a *-equivalence conformance corpus is straightforward to mirror.

⁵ Unconfirmed → classified POOL conservatively. No public royalty-free declaration was found; it is an MPEG/ISO standard, so the ISO patent-declaration process applies and a pool is the likely outcome (as with the MPEG point clouds). Treat as pool until the ITTF patent database / a pool announcement confirms otherwise — do not assert royalty-free.

HDR dynamic metadata (descriptors, not media payloads)

capability	standard	royalty	tier
wai.meta.hdr10plus	HDR10+ dynamic metadata (SMPTE ST 2094-40)	RF (content)	Lossless
wai.meta.dovi	SMPTE ST 2094-10 dynamic metadata (RPU)	PROP	Lossless
wai.meta.st2094_50	SMPTE ST 2094-50:2026-08 dynamic metadata (Application #5)	POOL (not established)	Lossless
wai.meta.gainmap	ISO 21496-1:2025 gain map (metadata and gain map image)	POOL (not established)	Lossless

The two rows added 2026-10-22. ST 2094-50:2026-08 was read in its published text: its Foreword records that compliance may involve claimed IPR, that each holder assured SMPTE it will license on reasonable terms "demonstrably free of discrimination", and that no royalty-free commitment is stated; it is classified conservatively (pool), as NNC is. ISO 21496-1:2025 (first edition, ISO/TC 42) was not read: the text is sold, and its public preview could not be fetched; its patent declarations were not checked, so it too is classified conservatively. Its gain map is a companion of an image payload, the first companion bound to an image rather than a video; SPEC §7.1's rule 2 gained the image class for it. What exact-tonemap states of ISO 21496-1 (its binary metadata's fields, its application formula) is what ST 2094-50:2026-08's terms 3.1 to 3.11 say of it and what public, open-source readers of its binary metadata implement; each such statement is marked in the extension.

Architectural note. Unlike the codecs above, HDR metadata is not a standalone media payload — it is color / dynamic-range metadata that accompanies a video payload (carried in SEI / ST 2094). It stresses the §2 invariant ("payload = the named codec's canonical bytes"). The clean design: these are companion descriptors a manifest declares alongside a video capability; the video bytes stay the video capability's, and the descriptor is bound as a companion component (the multi-component / multi-rendition envelope's job — v1.1). Static HDR10 (SMPTE ST 2086) and HLG (ARIB STD-B67) are baseline characteristics already in-band in the video bitstream and need only a manifest color-descriptor field, not a separate capability. This is the one place to resolve in SPEC before an integration partner's engineers ask.

Projection descriptor (a companion descriptor, registered 2026-10-22)

capability	standard	royalty	tier
wai.meta.projection	WAI's WPJ1 record (projection): projection kinds of ISO/IEC 23090-2 as ITU-T H.274 §8.15's SEI messages state them, packing arrangements of ITU-T H.273 §8.4	RF (WAI-native)	Lossless

A companion descriptor as the HDR descriptors are: it says how a video rendition's pictures map onto what a viewer sees and how its two views are arranged, and is applied by a renderer after decoding. ISO/IEC 23090-2 is not openly readable and was not read.

Neural image coding standards

capability	standard	royalty	status
wai.image.ieee1857_11	IEEE 1857.11-2024, neural-network-based image coding	POOL, provisional¹⁷	candidate: no payload format pinned

IEEE 1857.11-2024 (published 2024-12-20) is not openly readable. Its open reference software (BSD-2-Clause, copyright of the working group) writes a picture header with no start code, from a working draft, and marks the code that would write one as not yet written; format-probes §10 reads that header. The open call for proposals asked for a reproducibility check by re-running a proponent's decoder, not for identical output across machines, and the software's quantised layers fall back to float arithmetic in places, so the registry class is NeuralFloat.

Neural audio codecs (registered 2026-11-09)

capability	model	licence of code and weights	status
wai.neural.encodec24	EnCodec 24 kHz (arXiv 2210.13438)	MIT: the paper gives the repository as where its code and models are; the repository, whose code fetches the pretrained models, is MIT and states no other licence; the model hub's card for this model states none	registered; payload pinned (SPEC §5, Payloads of the neural audio codecs)
wai.neural.encodec48	EnCodec 48 kHz	MIT, as above; the model hub's card for this model states MIT	registered; payload pinned
wai.neural.dac16	DAC 16 kHz (release 0.0.5)	MIT: the README releases the weights "as part of this repo under MIT license"	registered; payload pinned
wai.neural.dac24	DAC 24 kHz (release 0.0.4)	MIT, as above	registered; payload pinned
wai.neural.soundstream	SoundStream (arXiv 2107.03312)	no model of the publication released	candidate: no payload format or parameter set pinned; POOL, provisional, as every candidate

The operating points come from each model's release: EnCodec's README (24 kHz at 1.5 to 24 kbps with 2 to 32 codebooks; 48 kHz at 3 to 24 kbps with half as many, at twice the frame rate, in one-second chunks overlapping by 1%) and its model configuration; DAC's training configurations conf/final/16khz.yml and 24khz.yml (encoder rates 2·4·5·8, so a hop of 320; 12 and 32 codebooks of 1024; quantizer dropout 0.5). These licences answer open item (c) below for these four rows only; wai.neural.encodec32's checkpoint is still not re-verified. SoundStream's publication released no model; an open speech codec built on its architecture (Apache-2.0) ships its own models, not ONNX, at 16 kHz and 3.2 to 9.2 kbps and its own packet, which the ONNX route does not read and no revision registers.

Neural video, and tensor frames (registered 2026-11-09)

capability	source	royalty	status
wai.neural.dcvc	DCVC, NeurIPS 2021 (arXiv 2109.15047)	POOL, provisional¹⁹	candidate
wai.neural.dcvc_tcm	DCVC-TCM, IEEE TMM (arXiv 2111.13850)	POOL, provisional¹⁹	candidate
wai.neural.dcvc_hem	DCVC-HEM, ACM MM 2022 (arXiv 2207.05894)	POOL, provisional¹⁹	candidate
wai.neural.dcvc_dc	DCVC-DC, CVPR 2023 (arXiv 2302.14402)	POOL, provisional¹⁹	candidate
wai.neural.dcvc_uf	DCVC-UF, CVPR 2026 (arXiv 2606.04410)	POOL, provisional¹⁹	candidate
wai.video.nnvc	NNVC, the neural-network-based video coding exploration of the Joint Video Experts Team	POOL, provisional²⁰	candidate
wai.tensor.weft	WeftCodec format 1.0: WEFTCODEC-FORMAT v1.0.0, WEFT-AI-CODEC v1.3.0, as published with weftcodec 1.2.0	RF²¹	registered: RAW, TENSOR, FLOAT_TENSOR, decoded by codecs::weft

The DCVC family. The family's code is published in one repository under MIT, each model with its checkpoints: DCVC, DCVC-TCM, DCVC-HEM, DCVC-DC, DCVC-FM and DCVC-RT in folders of their own, DCVC-UF at its root (its code read at commit cbdae87a, 2026-08-23). What was read is each model's decode path, as far as the form of its decoder:

• DCVC decodes a motion latent and a contextual latent; the contextual latent's entropy parameters come from the hyperprior, a temporal prior from the motion-compensated context, and an autoregressive context over the latent decoded so far (a masked convolution).
• DCVC-TCM decodes a motion latent and a contextual latent; the latter's parameters come from the hyperprior and a temporal prior over contexts mined from the propagated feature.
• DCVC-HEM adds the previous frame's latent to the priors and decodes each latent in two spatial passes; DCVC-DC decodes in four.
• DCVC-RT decodes one latent, with no motion latent, its parameters from the hyperprior and a temporal context, in two spatial passes; DCVC-UF codes a chunk of frames as one latent.

None takes the form of neural/video_dcvc.rs's route (the latent's parameters from the hyperprior alone, then a temporal decoder over the latent and one reference feature). Each writes its bitstream with its own entropy coder through research code that fixes no format, so the five not already registered are candidates. wai.neural.dcvc_rt and wai.neural.dcvc_fm stay registered on the route (SPEC §5, The inter-frame route and the DCVC family), where a published checkpoint does not decode unchanged.

NNVC. The exploration's reference software is built on the VVC test model and adds neural-network tools, in-loop filtering and intra prediction among them; its anchor is the same software with every such tool off (VTM-11.0_nnvc). Its tagged releases run to NNVC-8.0, then release candidates of 9 (the software's tags, read 2026-11-09). No standard is published, and the bitstream follows the software's release, so no payload format can be pinned.

WeftCodec. A lossless codec for tensor weights, checkpoint deltas and inference state: a 96-byte header with CRC-32C, a dependency and a chunk table, the body and a CRC-32C trailer; every plaintext bound by BLAKE3 under a domain (WEFTCODEC-FORMAT v1.0.0 §1 to §13, read whole; WEFT-AI-CODEC v1.3.0 §0 to §5, read for its mode index and its TENSOR and FLOAT_TENSOR modes). The text is normative and versioned, its vectors frozen and append-only, and a new mode is a reserved value allocated, which an older decoder refuses as UnsupportedMode (§13). The text requires its two implementations (Rust, Go) to write the deterministic modes byte for byte alike. The registration admits the three modes whose decoding the text specifies entirely and needs no library: a frame of ZSTD or DICTIONARY decodes through a compression library's format, and REFERENCE, DELTA and the delta modes name a base the payload does not hold. The other tensor modes (token sequences, quantised and tiled tensors, key-value caches, activation streams) are refused by this registration as UnsupportedMode.

Decoder conformance — what each named standard fixes

SPEC §7 classes every standard codec — the floor and modern menu, the broadcast codecs, and the capabilities above and below — by what its named standard's decoder conformance fixes, because that is what intent = replicate over it promises: the same bytes on every conforming sink (StandardDefined), a criterion around a reference (StandardTolerance), or not yet classified (StandardUnclassified). Checked 2026-09-27. The last column says what was read; a row is only as strong as that. A row reaches StandardDefined only on a clause that requires identical output; a standard whose conformance text was not read stays StandardUnclassified.

capability	class	what the standard fixes	read
wai.image.png	StandardDefined	lossless: "All PNG filters are strictly lossless", and the Paeth predictor's calculations "must be performed exactly". Reducing to a lesser sample depth is the decoder's choice: §10.4 gives a rounding formula and a right shift	RFC 2083, full text
wai.audio.flac	StandardDefined	lossless: decompressing "returns exactly the original data" (§3); STREAMINFO carries an MD5 of the unencoded audio	RFC 9639, definitions and STREAMINFO table
wai.tensor.weft	StandardDefined	the plaintext exactly: the header carries H("weft/chunk/v1", plaintext), a decoder refuses a frame whose decode does not reproduce it, and every admitted mode is exact ("Decode(Encode(x)) == x byte-for-byte, or the decoder fails closed")	WEFTCODEC-FORMAT v1.0.0 §2.1, §12; WEFT-AI-CODEC v1.3.0 §1
wai.text.zstd, wai.text.xz	StandardDefined	lossless byte formats	not re-read
wai.video.avc	StandardDefined	§C.4: test bitstreams go to the HRD and the decoder under test; "for each picture output by the HRD, the values of all samples that are output by the DUT for the corresponding picture shall be equal to the values of the samples output by the HRD". Film grain characteristics are an SEI message, outside the decoding process	Rec. ITU-T H.264 (V16) (06/2026), §C.4
wai.video.hevc, wai.video.vvc	StandardDefined	§C.5.1: "for each such cropped decoded picture output by the DUT, the values of all samples that are output shall be equal to the values of the samples produced by the specified decoding process". Film grain characteristics (and, in H.266, other post-processing hints) are SEI messages, outside the decoding process	Rec. ITU-T H.265 (V11) (01/2026) and H.266 (V4) (01/2026), §C.5.1
wai.video.mvhevc	StandardDefined	§F.13.5.1: as §C.5.1, for every output layer: "for each such cropped decoded picture output by the DUT, the values of all samples that are output shall be equal to the values of the samples produced by the specified decoding process"	Rec. ITU-T H.265 (V11) (01/2026), §F.13.5.1
wai.image.jxs, wai.video.jxs	StandardDefined	two conformance points: "a strict conformance point that requires decoders to reconstruct codestreams into image data that is bitwise identical to the reference data, and a relaxed conformance point" (§6.3); the strict point when the PSNR difference is infinity for every codestream (Annex B.2). The registration requires the strict point of a sink, so its class rests on that clause and that requirement	ISO/IEC 21122-4:2022 public preview (foreword, §6.3, Annex B.2); Annex C's PSNR bounds are not in the preview
wai.video.vp9	StandardDefined	§8.1: "Decoders shall produce output frames that are identical in all respects and have the same output order as those produced by the decoding process specified herein"	VP9 Bitstream & Decoding Process Specification v0.6 (31 March 2016), §8.1
wai.audio.opus, wai.audio.opus.packet	StandardTolerance	§6: conformance is "defined through the reference implementation"; a decoder's output "MUST also be within the thresholds specified by the opus_compare.c tool" against the reference on the test vectors, with "the same final range decoder state". §6.1: quality 0 is the passing threshold, calibrated to additive white noise at 48 dB SNR; 100 means identical to the reference. §1.1: "Various operations in the codec require bit-exact fixed-point behavior"; §4.2.7.9: the rest of SILK frame reconstruction "does not need to be bit-exact"	RFC 6716 §1.1, §4.2.7.9, §6–§6.1
wai.audio.opus.dred	StandardTolerance	§4.1: "DRED acoustic feature decoding is strictly defined. The decoder implementation MUST use the same weights provided in this specification. The DNN weights can be further quantized and the exact implementation of the DNN arithmetic (including activation functions) can be approximated, provided that they comply with the following test"; the test compares decoded features with reference features, and its script passes at total ≤ 0.5, max ≤ 0.15 and pitch ≤ 0.02. §4.2: the vocoder is not normative. Measured (extensions/opus-dred.md §6): two builds of the draft's reference decoder on one host, and the build that wrote the test vectors, give three different feature sets from the same latents	draft-ietf-mlcodec-opus-dred-07 §2.2.2, §4.1, §4.2, §7.1 ([DRED-Compare-Script])
wai.audio.opus.dred.int	NeuralIntegerExact	the same weights (§4.1 there), through the integer profile dred-int/1: every feature an exact integer function of the payload and the pinned weights file, the same on every conforming sink; within the draft's thresholds of its reference features	extensions/opus-dred.md §4, §6
wai.image.jpeg	StandardTolerance	§7: a decoder shall "with appropriate accuracy, convert to reconstructed image data"; A.3.3 NOTE: the IDCT equations "contain terms which cannot be represented with perfect accuracy by any real implementation", and the accuracy of dequantisation + IDCT is specified in Part 2; §4.1: colour space "is outside the scope of this Specification". The lossless process (Annex H) is exact, but the capability's role is the DCT processes	ITU-T T.81 (1992), full text. Part 2 (ITU-T T.83 | ISO/IEC 10918-2) not read: the download failed
wai.image.jxl	StandardTolerance	§4.1.1 NOTE: JPEG XL specifies infinite-precision arithmetic, and "this document therefore specifies tolerances for the decoded samples"; §4.2: a decoded image is similar to the reference iff every sample is within a peak-error threshold and each channel's RMSE within an RMSE threshold; §4.3: thresholds per test (peak_error, rms_error)	ISO/IEC 18181-3:2022, public preview (clauses 1–4.3)
wai.image.jp2, wai.video.j2k	StandardTolerance	decoder compliance compares decoded components with reference components and their error metrics with specified values (Annex B.2.4, B.2.5; "Tolerances", G.4.3; MSE among the abbreviations; a "9-7I precision guarantee", A.1.7); the reversible 5-3 path is lossless (definitions 3.41, 3.58)	ISO/IEC 15444-4:2024 | ITU-T T.803, public preview: table of contents, terms and abbreviations only. The clauses that state the metrics and thresholds were not in the preview
wai.video.av1, wai.video.av1.lossless, wai.image.avif	StandardTolerance	§7.2: without film grain, decoders "shall produce output frames that are identical in all respects" to the decoding process's; with film grain, a decoder may instead produce identical frames before grain synthesis and output frames that "do not have perceptually significant differences" from the reference grain synthesis; "The definition of “perceptually significant differences” is beyond the scope of this specification and may be specified, for example, by a service provider as part of their accreditation program", so this criterion is not a stated bound. The film-grain syntax does not depend on lossless coding, so av1.lossless is in the same class. A text search of AVIF v1.2.0 found no film-grain restriction	AV1 Bitstream & Decoding Process Specification v1.0.0 with Errata 1, §7.2 and §5.9.30; AVIF v1.2.0, text search
wai.video.av2	StandardTolerance	§7.1: with film_grain_params_present equal to 0, output frames "identical in all respects" to the decoding process's; equal to 1, a conformant decoder satisfies one of two options — the exact reference film grain synthesis (§7.21.7), or identical intermediate frames (§7.21.2) and output frames that "do not have perceptually significant differences" from the reference grain synthesis, a term the specification leaves to others to define. Annex A.5 (decoder conformance) adds only level capability and the display schedule	AV2 Bitstream & Decoding Process Specification v1.0.0 (28 May 2026), §7.1 and Annex A.5
wai.audio.ivas	StandardTolerance	§7.1: bit-exact conformance, output matching the reference test sequences, for every implementation; §7.2.1: an implementation "based on floating–point code (TS 26.258)" whose output is not bit-exact is tested instead by a non-bit-exact procedure — TS 26.444's for mono operation (§7.2.1.1), and for stereo and immersive operation a per-frame maximum-loudness-difference corridor derived from reference builds that "are not bit-exact with each other nor with the reference implementation" (§7.2.1.2, Annex A)	3GPP TS 26.252 V19.2.0 (2026-06), §7 and Annex A. TS 26.444 not read
wai.audio.iamf	StandardTolerance	substreams are Opus, AAC-LC, FLAC or LPCM (codec_id); "specific renderers are out of scope" (with reference to §2.2)	IAMF working draft of 21 April 2025, text search
wai.video.vp8	StandardUnclassified	not checked. It shares VP9's framing (IVF), not its class: VP9's §8.1 says nothing about VP8	RFC 6386 read for its framing only (§4, §9.1, §9.5); its decoder-conformance text was not read for this registration (2026-10-18)
wai.video.evc, wai.video.evc.baseline, wai.video.avs3, wai.audio.mpegh, wai.audio.avs3	StandardUnclassified	not checked	nothing read for this purpose: ISO/IEC 23094-1 and 23008-3 and their conformance parts are not freely available, and T/AI 109.2 / 109.3 were not obtained
wai.video.lcevc	StandardUnclassified	not checked; its decoding process includes a signalled dither step (clause 8.11, "Decoding process for dither filter"), so identical output across decoders is not presumed	ISO/IEC 23094-2:2021 public preview sample (front matter to clause 5.8): table of contents and clause 3 only. The conformance part (ISO/IEC 23094-3) not read
wai.audio.ac3, wai.audio.eac3	StandardUnclassified	A/52 lets a decoder fill a mantissa coded with no bits (bap 0, under dithflag) with "any reasonably random sequence" (§7.3.4), and holds downmix coefficients to a relative accuracy of ±0.25 dB (§7.8), so two conforming decoders need not output the same samples	ATSC A/52:2018, its syntax (§5, Annex E §2) and §7.3.4, §7.8, §7.10 (2026-11-08)
wai.audio.aac	StandardUnclassified	not checked. Three decoders on one host gave three different PCM streams from one ADTS stream (measured above), so deployed decoders do not agree byte for byte; whether ISO/IEC 14496-26 admits that, or makes some of them non-conforming, was not established	nothing read for this purpose
wai.meta.st2094_50, wai.meta.gainmap	StandardUnclassified	ST 2094-50:2026-08 states its tone map in real numbers (§6.2.5, Formulas (2) to (5)) and gives no decoder conformance over rendered pixels, so identical renditions across sinks are not presumed; the integer derivations of exact-tonemap are WAI's, not a property of the capability	ST 2094-50:2026-08, every clause and annex; ISO 21496-1 not read (2026-10-22)
wai.audio.ac4, wai.mesh.draco, wai.pc.gpcc, wai.pc.vpcc, wai.splat.sog, wai.haptic.mihs, wai.meta.hdr10plus, wai.meta.dovi	StandardUnclassified	not checked	nothing read for this purpose. (wai.haptic.mihs's registry comment and SPEC §5 say ISO/IEC 23090-33 defines bit-exact decoder verification; that rests on a summary, not the standard text, and was not checked here); wai.splat.sog is a candidate with no pinned payload, so there is no conformance text for it to read
wai.mesh.glb	StandardUnclassified	the payload bytes pass through intact; what a sink decodes from them is not classified. A glTF binary carries more than raw buffers: glTF 2.0 images may be PNG or JPEG ("JPEG images MUST be compatible with JPEG File Interchange Format"), embedded through a bufferView, and geometry may be compressed through extensions (KHR_draco_mesh_compression; the vendor extension EXT_meshopt_compression). JPEG is StandardTolerance and Draco StandardUnclassified above, so a scene built from those can differ between conforming sinks; and glTF specifies an asset format, not a decoder conformance over rendered output (edge appearance, for one, "MAY be subject to implementation-specific techniques")	glTF 2.0.1 specification (Khronos registry HTML), text search; Khronos glTF extension registry README
wai.splat.spz	StandardUnclassified	the payload bytes pass through intact; SPZ stores quantised attributes, and its decode is not classified	not read

Two limits apply to every row. The standard fixes its decoded output in its own domain (sample depth, planes, colour representation); a conversion a sink applies afterwards — PNG depth reduction, YUV→RGB, a JPEG colour-space choice, rendering an IAMF mix — is not part of it, and the reference sink's RGB8 / f32 buffers include such conversions. And the class describes the standard, not the reference sink's libraries: nothing here tests those libraries against their standard's conformance suite.

JPEG AI (wai.image.jpegai) is not in a standard class: its registry class is NeuralEntropyExact (determinism-tiers §5), from SPEC §7's neural rule, not from what its standard's conformance fixes. What was read is an overview of the standard (arXiv 2510.13867, read 2026-09-27):

• the bit-exact conformance point is "only at the end of the entropy decoding pipeline"; "Outputs from subsequent processes may diverge slightly from the reference";
• one codestream may be decoded through "up to three different synthesis transforms", restricted by the profile (Part 2) and possibly by the encoder; bit-exact picture reconstruction was "deliberately excluded from the first version";
• Part 4 (ISO/IEC 6048-4, conformance) "specifies the requirements for generating a JPEG AI-compliant reconstruction" and "defines a testing suite for each profile and level pair"; it "considers the possibility that a decoder is conformant to the standard without having bit-exact reconstruction", which the overview goes on to call "standard compliance with a certain leeway". It says Part 4 was a draft at the time of writing.

Read together, these passages point to a criterion around a reference on the reconstruction — the kind StandardTolerance records — but the overview does not state one, nor its form or width. A measurement of one decoder does not bear on this class. WAI's exact JPEG AI decoder pins one reconstruction across machines under the conditions determinism-tiers states; that shows one implementation is reproducible, and says nothing about what another conforming decoder outputs, which is what the class records. Part 4 itself was not read. A web search on 2026-09-27 returned the ISO catalogue entry for ISO/IEC 6048-4 as under publication (stage 60.00); the catalogue page refused a direct fetch (HTTP 403), so that status is the search result's, not confirmed here. Keeping wai.image.jpegai at NeuralEntropyExact, where replicate is refused, is therefore a conservative choice pending Part 4, not a reading of the standard; if Part 4 states a criterion on the reconstruction, the capability is a candidate for StandardTolerance (SPEC §7, wai.image.jpegai and the standard classes). Only a Part 4 that required a bit-exact reconstruction could support a same-bytes class, and none of the text read suggests it does.

What would upgrade each conservative row: for a StandardUnclassified standard codec, its conformance clause, read, requiring identical output (then StandardDefined) or stating a criterion (then StandardTolerance). For wai.mesh.glb, either a conformance document for glTF decoding, or a narrower capability whose payload is restricted to uncompressed buffers with no embedded images, whose reconstruction could then be defined as the parsed buffers — a new capability string, since §8 fixes this one's payload.

Broadcast and streaming video / audio

The elementary bitstreams terrestrial and satellite broadcast, streaming and device playback carry. Registered so an envelope can name the bitstream a distribution chain already holds; all sink-supplied, none a default.

capability	standard	payload (canonical bytes)	royalty	reference / open decoder	tier
wai.video.av2	AV2 (AOMedia), spec v1.0.0, 28 May 2026	AV2 Annex B length-delimited bitstream (leb128 size before each OBU)	RF⁶	AVM v1.0.0	Lossy
wai.video.vvc	ITU-T H.266 | ISO/IEC 23090-3	H.266 Annex B byte stream	POOL⁷	VTM	Lossy
wai.video.hevc	ITU-T H.265 | ISO/IEC 23008-2	H.265 Annex B byte stream	POOL⁷	HM	Lossy
wai.video.mvhevc	ITU-T H.265 | ISO/IEC 23008-2, Annexes F and G (registered 2026-10-22)	H.265 Annex B byte stream of a multiview bitstream	POOL⁷	HTM (BSD-3-Clause)	Lossy
wai.video.avc	ITU-T H.264 | ISO/IEC 14496-10	H.264 Annex B byte stream	POOL⁷	JM	Lossy
wai.video.vp9	VP9 Bitstream & Decoding Process Specification	IVF frame container, FourCC VP90	RF⁸	libvpx	Lossy
wai.video.vp8	RFC 6386	IVF frame container, FourCC VP80	RF¹⁵	libvpx	Lossy
wai.video.evc	MPEG-5 Part 1, ISO/IEC 23094-1 (any profile)	Annex B raw bitstream file storage format	POOL⁹	XEVD	Lossy
wai.video.evc.baseline	ISO/IEC 23094-1, Baseline profile only	as above, Baseline-conformant	RF by design⁹	XEVD	Lossy
wai.video.avs3	AVS3-P2: T/AI 109.2-2021; IEEE 1857.10	AVS3-P2 elementary bitstream	POOL¹⁰	uavs3d (BSD-3-Clause; Baseline profile)	Lossy
wai.video.lcevc	MPEG-5 Part 2, ISO/IEC 23094-2	LCEVC enhancement as the ISO/IEC 23094-2 Annex B byte stream (start-code-delimited NAL units) — companion-only	POOL¹¹	LCEVCdec (BSD-3-Clause-Clear, patent licences excluded)	Lossy
wai.audio.mpegh	MPEG-H 3D Audio, ISO/IEC 23008-3	MHAS packet stream	POOL¹²	mpeghdec	Lossy
wai.audio.avs3	AVS3-P3: T/AI 109.3-2023; GY/T 363-2023 ("Audio Vivid"); GB/T 46271-2025	AVS3-P3 elementary bitstream	POOL¹⁰	—	Lossy
wai.audio.aac	ISO/IEC 14496-3: AAC-LC, HE-AAC, HE-AAC v2	ADTS stream	POOL¹³	—	Lossy
wai.audio.ac3	ATSC A/52:2018 §5; ETSI TS 102 366 (registered 2026-11-08)	AC-3 syncframes back to back, bsid 0 to 8	PROP¹⁸	—	Lossy
wai.audio.eac3	ATSC A/52:2018 Annex E; ETSI TS 102 366 Annex E (registered 2026-11-08)	an E-AC-3 bit stream: syncframes back to back, independent substream 0 first in each access unit	PROP¹⁸	—	Lossy
wai.audio.opus.packet	RFC 6716	one Opus packet (RFC 6716 §3), TOC byte first, no container	RF¹⁵	libopus	Lossy
wai.audio.opus.dred, wai.audio.opus.dred.int	draft-ietf-mlcodec-opus-dred-07	one DRED extension payload (draft §3) for the Opus packet it accompanies — companion-only	POOL¹⁶	the draft's reference implementation (Appendix B there); for .int, wai-rs dred_exact (features only)	Lossy

Where these bitstreams are carried (the reason for registering them, not a ranking): Brazil's TV 3.0 / DTV+ — VVC video, LCEVC enhancement, MPEG-H 3D Audio — launched by decree on 27 Aug 2025 and switched on in Rio de Janeiro, São Paulo and Brasília in June 2026. DVB's video/audio coding specification (A001, ETSI TS 101 154) added AVS3 video in 2022 and AVS3-P3 immersive audio in November 2024 (BlueBook A001r22). China's GB/T 46271-2025 (3D audio coding, distribution and presentation) was published 29 Aug 2025, in force 1 Mar 2026. Japan's 4K/8K satellite broadcasting (on air since 1 Dec 2018) uses HEVC. AVC with AAC, and VP9, are widely deployed in streaming and device playback. WAI registers LCEVC over AVC, HEVC, VVC, EVC, VP9 and AV1, the base codecs a published description of LCEVC gives as examples ("e.g., AVC, HEVC, VP9, AV1, EVC or VVC", https://en.wikipedia.org/wiki/LCEVC); with all six registered, every LCEVC base an envelope names is itself a registered capability. The list is WAI's registration, not a limit of the standard: ISO/IEC 23094-2:2021 defines the decoded base picture as the output of "a base decoder" (definition 3.17), says the base decoder's IDR picture "is not specified in this document" (3.30, Note 1), and its normative references (clause 2: ISO/IEC 11578, ITU-T H.273 | ISO/IEC 23091-2, ITU-T T.35) include no video codec; these clauses are in the public preview sample cited below. AV2, AVS3 video, JPEG 2000 and the WAI-native integer codecs are not registered bases; a later revision may add one, which SPEC §8 treats as a widening.

Payload framing is named per entry because §8 fixes a capability's payload format forever once registered, so each entry names one form. Where the standard defines a self-delimiting elementary form the entry names it (EVC's Annex B raw bitstream file storage format, MHAS for MPEG-H 3D Audio, ADTS for AAC). H.264, H.265, H.266 and LCEVC each define a bitstream as either a NAL unit stream or a byte stream (H.264 3.15, H.265 3.13, H.266 3.17, ISO/IEC 23094-2 3.2), and only the byte stream delimits its own NAL units, with the three bytes 0x000001 "as a prefix to each NAL unit" (ISO/IEC 23094-2 3.51; H.264 3.163, H.265 3.156 and H.266 3.138 say the same). So each of those pins is that standard's Annex B byte stream, and a bare or length-prefixed NAL unit stream is not the capability. For LCEVC this rests on ISO/IEC 23094-2:2021's public preview sample, which runs from the front matter to clause 5.8; the table of contents and clause 3 (definitions 3.1, 3.2, 3.6, 3.35, 3.36 and 3.51) were read (https://www.iso.org/standard/79143.html, sample at https://cdn.standards.iteh.ai/samples/79143/fcb43bd468bf4d29825bc07103a32573/ISO-IEC-23094-2-2021.pdf). The table of contents lists "Annex B (normative) Byte stream format"; the annex's body is not in the sample, so its syntax was not read. AV2 OBUs carry no size of their own and the AV2 specification states that framing is outside its scope except for the one method it defines, Annex B — so that is the pin. VP9 frames are not self-delimiting and the VP9 specification defines no stream framing, so the pin is IVF, the minimal frame container the VP9 reference library (libvpx) reads and writes. VP8 (registered 2026-10-18) takes the same pin with FourCC VP80: a VP8 frame does not delimit itself either, since its last partition is "the remainder of the data not used by any of the previous partitions" (RFC 6386 §9.5) and the decoder "is simply presented with a sequence of compressed frames" (§4). RFC 6386 was read for this framing question only (§4, §9), not for decoder conformance. The Opus packet form (registered the same day) is one packet of RFC 6716 §3, its TOC byte first, with no container; the payload's length is the packet's length (the undelimited form of §3, not the self-delimiting framing of its Appendix B). It is the form a live track carries, one packet per object, and it carries none of Ogg Opus's header fields (pre-skip and output gain, RFC 7845 §5.1), so it is a string of its own beside wai.audio.opus, the Ogg Opus stream. The two AVS3 rows name the elementary bitstream of T/AI 109.2 (video) and T/AI 109.3 (audio). Neither text was obtained, so whether either standard defines a second form of its bitstream was not checked; if one does, the row must name the form before §8 fixes the string (open item 5(h)).

AV2 placement — registered, not recommended. The recommended set is what every conforming sink SHOULD implement, and WAI's value there is availability. AV2 is royalty-free (⁶) and has a reference decoder, but as of September 2026 AVM is a reference model, a production software decoder is still in development, and hardware decode in shipping devices is not confirmed. Asking every sink to carry it now would ask for a decoder most cannot deploy. So it is registered (a WAI file can carry AV2 today, royalty-free) and SPEC §5 tells encoders to ship it in a WAI2 envelope with a wai.video.av1 rendition of the same content. A v1.0 fallback of wai.video.av1 would not do: dispatch step 5 hands the same AV2 payload to the fallback's decoder, which cannot read it. Promotion is a MINOR revision once deployable decode exists; the capability string and framing stay fixed. defaults_are_royalty_free_and_never_pool_or_proprietary asserts it is RF and not (yet) a default.

LCEVC and the companion mechanism — the fit, and where it stops. LCEVC is an enhancement layer: residuals added to the upscaled output of one specific base encode. SPEC §7.1's companion components fit most of it:

• Applied with, never instead of. A companion is never returned by pick / primary, so a sink with no LCEVC decoder presents the base — the backward-compatible behaviour LCEVC is designed for (TV 3.0 uses it to carry a UHD service on top of a usable HD one).
• Intrinsic binding. Base and enhancement ride in one envelope, not a sidecar.
• One base. ISO/IEC 14496-15's dual-track storage has the enhancement track refer to exactly one base track; companion_of is the analogue.

Two gaps, one closed here and one not:

1. Binding granularity (closed by a rule in both directions). companion_of names a capability, not a component. HDR metadata describes content and can legitimately apply to several renditions of one codec, and several descriptors can apply to one base; LCEVC residuals belong to one encode and are added once. So an enhancement layer's companion_of MUST name one of its base codecs — the capabilities that name the AVC, HEVC, VVC, EVC, VP9 and AV1 bitstreams, EVC Baseline and lossless AV1 included — and resolve to exactly one selectable rendition of it, and that base takes at most one enhancement layer — a sink applies every companion of the rendition it picks, so a second one would stack a second set of residuals. WaiMulti::check_component_roles rejects a base outside that list (JPEG 2000, AV2, AVS3, the WAI-native integer codecs, anything not video), zero or several renditions of the named base, a second enhancement on one base, and a companion-only capability mislabelled as a rendition (which pick / primary also never select). A component-id binding would lift the one-base-rendition limit; it is not part of this change. The v1.0 side has its own check: Wai::check_component_roles rejects a companion-only capability as a WAI1 capability or fallback, and the wai wrap tool refuses to write either (with --companion it writes the conforming WAI2 form instead).
2. Per-picture timing (open). A timed container aligns base and enhancement samples by timestamp; a WAI2 payload is an untimed elementary stream. The payload's framing is pinned (the Annex B byte stream, above), and ISO/IEC 23094-2 defines an access unit as the NAL units of one coded picture "associated with a particular output time" (3.1), but this revision does not pin how a separately carried enhancement's access units align picture by picture with its base, so the embedded carriage — enhancement inside the base bitstream as SEI / registered user data, carried under the base capability, no separate component — is the unambiguous form until it does.

Not registered this round, and why.

• Extended HE-AAC (USAC). ADTS cannot carry it; it needs LOAS/LATM or ISOBMFF framing, i.e. a separate capability with its own pinned payload.
• HDR Vivid. The HDR dynamic-metadata companion published alongside Audio Vivid in the same national-standard set; licence not verified.

⁶ Verified RF. The AVM v1.0.0 reference software's PATENTS file is the Alliance for Open Media Patent License 1.0 ("no-charge, royalty-free, irrevocable … patent license"); AOMedia's 9 June 2026 release states AV2 was "developed under AOMedia's royalty-free patent policy". Residual caveat, the same as for AV1 and VP9: a third-party licensing company announced (Dec 2025) an intent to form a pool asserting AV2 patents, and on 25 Aug 2026 a second licensing administrator launched an exploratory phase for a pool covering devices and software implementing AV2 and AV1, saying it expects AV2 to be added to its existing video-distribution pool licence (which covers VP9 and AV1) in 2027. WAI classifies a format by its own licence and records such assertions here rather than reclassifying on them. Sources: https://av2.aomedia.org/, https://github.com/AOMediaCodec/avm/blob/v1.0.0/PATENTS, https://accessadvance.com/2026/08/25/access-advance-launches-exploration-phase-for-an-av1-av2-device-and-software-patent-pool-invites-participation-of-potential-licensors/. ⁷ Verified POOL. VVC, HEVC and AVC essential patents are licensed for royalties through patent pools (several for VVC and HEVC, the long-running H.264 pool for AVC) plus bilateral licensors. None is royalty-free. ⁸ RF. VP9 is published with BSD-licensed source and a royalty-free patent grant from its developer. Residual caveat as ⁶: a third-party pool asserts VP9 patents; recorded, not adopted as the class. ⁹ EVC split by profile. ISO/IEC 23094-1's Baseline profile uses only tools more than 20 years old or offered royalty-free, and its proponents describe it as royalty-free; the Main profile adds tools whose holders license them on FRAND terms. So wai.video.evc (any profile — a decoder must handle Main) is POOL and wai.video.evc.baseline is RF by design. Residual caveat, like JPEG 2000 Part 1 (³): no formal royalty-free guarantee exists for Baseline. ¹⁰ Classified POOL. The AVS patent pool issued calls for essential patents for AVS3-P2 video and AVS3 audio on 11 Jan 2024, licensing on RAND terms with "competitive licensing fees"; AVS licensing has historically charged devices only. The fee schedule for AVS3 was not found published, but the pool is royalty-bearing, not royalty-free. AVS3-P3 and Audio Vivid are treated as one bitstream family (DVB and the open-source tooling name them together); the bitstream identity of T/AI 109.3, GY/T 363 and GB/T 46271 was not checked against the standard texts, which were not available. ¹¹ Verified POOL (single SEP licensor). LCEVC's essential patents are licensed under programmes announced Aug 2025: royalties on consumer devices and on video-distribution services, a free integration licence, and free-to-air broadcast royalties waived. The open decoder library is BSD-3-Clause-Clear, which expressly excludes patent licences. ¹² Verified POOL. MPEG-H 3D Audio essential patents are licensed through a patent-pool programme. ¹³ Verified POOL. The AAC pool programme covers AAC-LC, HE-AAC, HE-AAC v2 and Extended HE-AAC; a product shipping an AAC encoder or decoder pays a per-unit royalty. ¹⁵ RF, by the codec each one carries (registered 2026-10-18). wai.video.vp8: VP8's reference source is published under a BSD licence with a royalty-free patent grant from its developer, the basis ⁸ records for VP9; whether a third-party pool asserts VP8 patents, as one does VP9's (⁸), was not checked. wai.audio.opus.packet: Opus, classified royalty-free as wai.audio.opus is; the packet form adds no coding tool, only leaves out the Ogg container. ¹⁶ POOL, conservatively (registered 2026-10-22). One IPR disclosure bears on DRED: the IETF's disclosure 6063, on the individual draft that became draft-ietf-mlcodec-opus-dred, states a pending application and an intent to license royalty-free "if the technology covered by this disclosure is included in a standard adopted by the IETF". The draft is not yet a standard, so the royalty status is not established, and the registry's rule classes such a capability patent-pool, never royalty-free (REGISTRY.md, Licence classes). It moves when the condition is met.

¹⁶ POOL, conservatively (registered 2026-10-22). A licensing pool for JPEG XS essential patents, on RAND terms, is reported by a secondary summary; its terms were not read. The open encoder's licence grants a patent licence only for its contributors' own claims. Classified patent-pool, never royalty-free, until a primary source says otherwise.

¹⁷ Provisional (registered 2026-10-22). The reference software's licence notes that patent rights may apply; no licensing statement for the standard was read. A candidate's class is provisional in any case (SPEC §5).

¹⁸ PROP, conservatively (registered 2026-11-08). One licensor runs the licensing programme for AC-3 and E-AC-3 decoders and encoders, and still does. Expiry of their essential patents is reported, by press and patent databases rather than the licensor: AC-3's in March 2017, and for E-AC-3 a last US patent whose adjusted term ended on 2026-01-30. Neither was confirmed with the licensor, so neither is a basis for royalty-free; the class is AC-4's, a single licensor's. WAI carries the syncframes unread and bundles no decoder, so registration licenses nothing: a sink that decodes them brings its own decoder and its licence. The texts are ATSC A/52:2018 (§5, Annex E, read for the syntax and §7.3.4, §7.8, §7.10) and ETSI TS 102 366, its European publication.

¹⁹ Provisional (registered 2026-11-09). The DCVC family's code and checkpoints are published under MIT, which grants no patent licence; the patent status of its techniques was not examined. A candidate's class is provisional in any case (SPEC §5).

²⁰ Provisional (registered 2026-11-09). An exploration of a joint ITU-T and ISO/IEC team toward a codec in the VVC line, whose royalty terms no text states; classified as the VVC line is, until a published standard says otherwise.

²¹ RF (registered 2026-11-09). The format's text is published under CC BY 4.0, and its Rust reference implementation under Apache-2.0, whose grant covers each contributor's patent claims in the work (the Go implementation's licence was not read). No pool or licensing programme names the format.

Implementation & demonstration status

WAI's job is carriage + dispatch; decoders are sink-supplied (even the floor codecs are wrapped reference libraries, not re-implemented). So "implemented and demonstrated" means, per capability:

• Carriage proven for every registered class — wai wrap <cap> <payload> builds an envelope around the canonical bytes (over a class that does not back an unpinned replicate, only with --intent, SPEC §7); wai unwrap returns them byte-identically; the registry reports the tier + license. A companion-only capability is carried as a WAI2 companion instead (wai wrap <base-cap> <base-payload> --companion <cap> <payload>, then wai unwrap --component 1). The tests/carriage.rs conformance asserts this for the entire registry (0 B … 64 KiB payloads), each capability in the form SPEC §5 allows it, and runs in the no-system-deps CI matrix.
• Real end-to-end decode demonstrated for the open formats with an installed/built reference tool — the full carry → dispatch → decode path on a real file: JPEG 2000 (wai.image.jp2, OpenJPEG — cinema), IAMF (wai.audio.iamf, ffmpeg — immersive audio), Draco (wai.mesh.draco, draco_decoder — 3D geometry), G-PCC (wai.pc.gpcc, MPEG tmc3 built from source — point clouds; patent-pool, so demonstrated-but-not-royalty-free), and IVAS (wai.audio.ivas, 3GPP TS 26.258 IVAS_cod/IVAS_dec built from source — immersive voice/audio: real stereo PCM → IVAS bitstream → full PCM reconstruction, and the carried bitstream decodes bit-identically to a direct decode; patent-pool, so demonstrated-but-not-royalty-free). See tools/wai_carriage_demo.sh.
• Real encode + structural decode demonstrated — V-PCC (wai.pc.vpcc, MPEG tmc2 built from source with its bundled HM video codec): a real point cloud is encoded end-to-end through the full V-PCC pipeline (patch projection → occupancy/geometry video → HM intra encode) into a conformant V3C bitstream, which PccAppParser then structurally decodes into its V3C units (V3C_VPS / V3C_AD / V3C_OVD / V3C_GVD), and which WAI carries byte-exact. Full pixel reconstruction needs HM's standalone decoder, which the V-PCC-patched HM source builds only as a namespaced library (the PccAppVideoDecoder wrapper's job), not a standalone TAppDecoder — so the reconstruct step stays sink-supplied while the encode + V3C-conformance + carriage are real. Patent-pool, like G-PCC.
• Decoder sink-supplied (open reference exists, not bundled / not installed here): wai.video.j2k, wai.haptic.mihs, wai.meta.hdr10plus.
• Candidate, refused (SPEC §5, Candidate capabilities): wai.splat.sog has no pinned payload format, so the demo checks that wai wrap refuses it and writes nothing.
• Proprietary (no open decoder, by design — carried, sink-supplied): wai.audio.ac4, wai.meta.dovi, wai.audio.ac3, wai.audio.eac3 (registered 2026-11-08; carriage only, the demo script has no row for them).
• Companion-only (wai.meta.hdr10plus, wai.meta.dovi, wai.video.lcevc): carried in the demo the one conforming way, as a WAI2 companion bound to a base rendition (a real AVC Annex B stream when libx264 is installed). Each row requires the companion back byte-exact, the role check to pass, wai wrap to refuse the WAI1 form, and the base to decode to all 10 frames; the LCEVC row also requires a second enhancement on the same base to be refused. The companion payloads are representative bytes.
• Broadcast and streaming codecs. Every row of that table is covered by tests/carriage.rs (byte-exact wrap / unwrap, tier + licence present) and by the registry tests in codecs/mod.rs; the LCEVC binding rules by the container.rs unit tests. The crate gains no decoder and no C dependency: every one is sink-supplied. Real encode + decode demonstrated for wai.video.avc, wai.video.hevc, wai.video.vp9 and wai.audio.aac in tools/wai_carriage_demo.sh when ffmpeg has the encoders: the payload is built in the framing SPEC §5 pins (Annex B byte stream, IVF, ADTS), carried byte-exact, and decoded by ffmpeg with the matching raw demuxer forced. A video row passes only if all 10 encoded frames come back; the AAC row only if the decoded frame count equals the ADTS frame count of the encoded stream, counted by walking its frame headers rather than by the decoder (a payload cut to 600 B, or by its last byte, fails; random bytes under the forced demuxers are rejected). Carriage only for wai.video.vvc (no VVC encoder installed on the demo host), wai.video.av2, wai.video.evc, wai.video.evc.baseline, wai.video.avs3, wai.audio.mpegh, wai.audio.avs3, and wai.video.vp8 and wai.audio.opus.packet (registered 2026-10-18; the demo script has no row for either). wai.video.lcevc is companion-only and is demonstrated as a WAI2 companion (above).

The reference sink decoders for the volumetric + immersive-audio rows (MPEG tmc13 for G-PCC, tmc2 for V-PCC, 3GPP IVAS_cod/IVAS_dec) are fetched and built at pinned versions by tools/build_reference_decoders.sh, which prints the PATH / TMC2_CFG to export before tools/wai_carriage_demo.sh — so every real+decode / real+parse row reproduces from a clean checkout.

wai wrap / unwrap is the universal carriage operation: the payload is opaque to WAI, so the same envelope carries a JPEG 2000 codestream, an IAMF stream, a Draco buffer, or a proprietary AC-4 frame identically — that opacity is the point, and the reason WAI does not need to re-implement any of them.

What stays the manifest's job (not new capabilities)

• ADM / S-ADM (ITU-R BS.2076) — the object-audio authoring metadata model that feeds IAMF; agnostic, not a codec. A manifest/authoring concern.
• Static HDR10 / HLG color descriptors — a manifest color field (primaries, transfer, mastering-display), not a payload.

Spec follow-ups flagged by this round

1. Companion-component binding — DONE. The WAI2 multi-component envelope gains a component role: "companion" (with companion_of) marks a descriptor applied with the chosen rendition. HDR dynamic metadata (wai.meta.hdr10plus / wai.meta.dovi) binds to a video payload in the same envelope — pick/primary skip companions, companions_of(cap) returns the descriptors for the chosen component. Back-compat: role absent ⇒ rendition. SPEC §7.1 + a conformance test (companion_descriptors_bind_to_a_primary). Immersive-audio metadata was named here too, and that test first carried an IAMF stream as a companion; no capability for immersive-audio mix metadata is registered, and IAMF carries its own audio substreams, so it is standalone media. Since September 2026 SPEC §7.1 states three rules a companion must meet — its capability is companion-only (or unregistered), companion_of names a base it applies to (a video-class capability for HDR metadata — wai.video.*, the neural video codecs, the films — and a base codec for LCEVC), and that base is present — and check_component_roles enforces each, with a negative test per rule.
2. License classification — DONE. codecs::capability_license() / is_royalty_free() classify every capability; wai capabilities reports them; a coherence test enforces that no capability is unclassified. Built as a registry property, not a manifest field (a per-file licence claim would be spoofable; the codec's licence is intrinsic to the capability).
3. Royalty footnotes — VERIFIED (June 2026). IAMF (AOMedia PL 1.0), JPEG 2000 Part 1 (ITU-T 2.1), Draco (Apache-2.0), HDR10+ (content+device RF) → confirmed royalty-free; IVAS (a voice-codec pool) → confirmed pool; G-PCC/V-PCC → pool, a conservative classification (¹⁴); AC-4 / ST 2094-10 dynamic metadata → proprietary. Residual: JPEG 2000 extensions/submarine-patent caveat (³), MPEG-I Haptics pool status unconfirmed (⁵).
4. Broadcast and streaming codecs — REGISTERED (September 2026). AV2, VVC, HEVC, AVC, VP9, EVC (any profile / Baseline), AVS3 video, LCEVC, MPEG-H 3D Audio, AVS3 audio / Audio Vivid and AAC, each naming one payload framing (LCEVC's is the ISO/IEC 23094-2 Annex B byte stream; the AVS3 rows' are unconfirmed against their standards, (h)) and a licence class with its basis (⁶–¹³). LCEVC is companion-only, binds to exactly one rendition of one of its base codecs, and a base takes at most one enhancement; companion-only capabilities are refused as a v1.0 capability, and as a fallback wherever the fallback would be used (§4 step 5). Licence classification made exhaustive (no default class), and so was the replicate classification, with an exact class admitted only on recorded evidence. AVC, HEVC, VVC and VP9 are StandardDefined on their conformance clauses; AV2 is StandardTolerance, as AV1; the other broadcast rows are StandardUnclassified (see Decoder conformance above). Residual: EVC Baseline's RF-by-design status (⁹), unpublished AVS3 fee schedule and unverified AVS3-P3 / Audio Vivid bitstream identity (¹⁰), third-party pool assertions against AV2 / VP9 (⁶, ⁸). Added 2026-10-18 for live ingest: wai.video.vp8 (RFC 6386, IVF with FourCC VP80, RF, StandardUnclassified) and wai.audio.opus.packet (one Opus packet, RF, StandardTolerance as wai.audio.opus), both sink-supplied and neither a default (¹⁵).
5. Open. (a) Per-picture alignment of a separately carried LCEVC enhancement with its base. Its framing is pinned (the Annex B byte stream), but how its access units map to the base's pictures outside a timed container is not; until it is, the embedded carriage is the unambiguous form. (b) A component-id companion_of binding, which would let one envelope carry enhancements for several renditions of one base codec. (c) The neural capabilities' model-weight licences were carried over, not re-verified, when the classification became explicit — the EnCodec 32 kHz checkpoint is distributed with a model collection whose weights are licensed CC-BY-NC 4.0, which would make wai.neural.encodec32 non-commercial if it applies. (d) Extended HE-AAC and HDR Vivid (see above); E-AC-3 and AC-3 were registered on 2026-11-08 (¹⁸). (e) The StandardUnclassified rows (VP8, EVC, AVS3 video and audio, LCEVC, MPEG-H 3D Audio, AAC, AC-4, G-PCC, V-PCC, Draco, SOG (which first needs a pinned payload), MIHS, the HDR descriptors, glb, SPZ) are waiting on each standard's conformance text; any whose clause requires identical decoded output moves to StandardDefined, with that clause recorded, and any with a stated criterion to StandardTolerance. (f) Closed: the spatial-audio extension now says a mixdown is only as portable as its sources' decoded PCM, and is_cross_decoder_exact no longer reports wai.audio.scene exact from the capability alone (is_cross_decoder_exact_with_sources checks the sources). (g) Closed: wai.image.jpegai is NeuralEntropyExact and does not back replicate (SPEC §7); it is not cross-decoder-exact. A cross-platform measurement of WAI's own decoder cannot change that, as Decoder conformance explains. (h) AVS3 framing. wai.video.avs3 and wai.audio.avs3 name the elementary bitstream of T/AI 109.2 and T/AI 109.3, which were not obtained. Before §8 fixes the strings, confirm from those texts that each defines one form of its bitstream, or name the form the capability carries. (i) What wai.pc.gpcc and wai.pc.vpcc carry. Under SPEC §8 (Editions), a registration that names no edition carries the edition of its standard current when it was registered, here 2026-06-17, and no later amendment, so the Gaussian-splat amendments (V-PCC at Draft Amendment, G-PCC at Committee Draft Amendment), a later edition, and V-DMC (a V3C application from the fourth edition of ISO/IEC 23090-5) need strings of their own. Still open: which edition of each part was current on 2026-06-17 (iso.org refused automated reads; the common test conditions name a 2023 edition of 23090-9), and whether wai.pc.vpcc carries the V3C bitstream of MIV. (j) The video-based anchor of both common-test-conditions documents installs a package named PLAS, which neither document describes. Whether it relates to the family of formats wai.splat.sog was registered for was not established.