Evaluation and media provenance

Benchmarks, task datasets, attribution, metadata, interoperability and quality review

• articleWhen source material exists, how do 'Screen Story by', 'Screenplay by', 'Adaptation by' and 'Based on' credits work in the WGA manual?
  In the WGA manual, 'Screen Story by' goes, only through arbitration, to a writer who takes just a springboard, character or incident from story-type source material and builds a substantially new story; the source author then gets a form-of-acquisition credit. 'Screenplay by' fits when story-type source material exists. 'Adaptation by' is generally opposed and only awarded by arbitration in unusual cases. 'Based on Characters Created by' goes to writers holding separated rights on sequels.
• articleWhat counts as 'source material' under the WGA Screen Credits Manual and what credit wording can it get?
  Under the WGA Screen Credits Manual (notices after 31 December 2021), source material is assigned material that was previously published or exploited, or written outside the Guild's jurisdiction, that the story or screenplay is based on. Its author gets a credit naming the form acquired ('From a Novel by', 'Based upon a Story by', ...). Earlier participating writers' material and research are not source material.
• articleHow is word error rate defined, what does the published critique say about it, and how much does the choice of text normaliser (for example Whisper's) change what is counted?
  WER counts substitutions, deletions and insertions against the reference word count, so it can exceed 1 (upper bound max(N1,N2)/N1). Morris et al. (2004) argue that measures of information communicated suit most applications better. WER also depends on text normalisation. Whisper's English and basic normalisers delete bracketed and parenthesised text (so tags like [music] or (laughs) vanish) and standardise numbers and spellings before scoring. WhisperX reports insertion error rate and 5-gram duplicates to catch hallucinated repetition. WER says nothing about timing, speaker labels or punctuation.
• articleWhat do peer-reviewed critiques say about how TTS MOS tests are reported and how test design changes the scores?
  A bare TTS score like 'MOS 4.2' is underspecified. Kirkland et al. (SSW 2023) found most 2021-22 Interspeech/SSW papers omit the scale labels, increments and listener instructions. In their own test, asking about 'quality' rather than 'naturalness' raised MOS (3.52 vs 3.21). The VoiceMOS 2022 organisers state MOS values can't be compared across listening tests, because each test has its own listeners and systems. ITU-T P.800 also warns that results on the older 0-4 scale must be raised by one to compare. Report the question, scale, listener pool and stimuli with any MOS.
• articleWhat fields does the CISAC AVR+ cue-sheet schema (Global Cue Sheet Standard 2.0 implementation) define for productions, cues, works, recordings and shares?
  CISAC's AVR+ (announced 18 June 2026) is a JSON Schema (draft 2020-12, schema version 1.0) for audiovisual music cue sheets, implementing the Global Cue Sheet Standard 2.0. A sheet needs production, cue and cue-sheet sections. Each cue needs a title (or an 'unknown title' flag), duration, usage code and type code. A cue may also carry in/out times, a work block (ISWC, society work codes, per-territory shares by interested party, role and right type) and a recording block (ISRC, artists, label, product codes, master owner). Usage codes are B Background, U Bumper, C Closing Theme, F Featured/Visual, L Logo, O Opening Theme and S Segment Theme; type codes are I Instrumental and V Vocal. The documentation prose mentions 'Jingle' and 'Source', which are not enum values, so the schema enums govern.
• articleWhat have the ColorInterop and OpenEXR maintainers said publicly about how readers should handle an unresolved or unknown colorInteropID?
  As of 2026-10-22, the ColorInterop 'bypass' ID issue is still open, and the OpenEXR recommendation was last edited on 2026-07-16. A maintainer's comment on that issue reads the current text this way: an ID other than 'unknown' that matches no colour space is an error that should ideally be shown to the user, while an application that doesn't recognise a new ID may treat it like any missing colour space, possibly by applying file rules. OpenEXR 3.5.0 added header colour-metadata validation helpers but no rule for unresolved IDs.
• articleWhat objective methods (STOI/ESTOI, SRMR, speech-loudness measures, SI4B monitoring) are stated for judging dialogue intelligibility in a finished mix, and what limits do the sources give?
  STOI/ESTOI (pystoi) are intrusive: they compare a clean reference with processed speech, at 10 kHz over 15 third-octave bands from 150 Hz, in 30-frame segments, after dropping frames more than 40 dB below the loudest. Becerra et al. (2024) call intrusive metrics unsuitable for finished film dialogue because no reference exists, and advise caution with the non-intrusive SRMR. Torcoli et al. (2024) propose loudness-based checks instead: speech loudness deviation (critical below -10 LU) and speech-to-background loudness difference (critical below 0 LU), as a first proposal. On their data, the 0 LU threshold correlated best with self-reported listening effort.
• articleWhat do the Commission's Article 50 guidelines (20 July 2026) and the code's deployer section say about synthetic voice, audio deep fakes, music and audio-only labelling?
  For audio-only deep fakes, Section 2 of the EU code (for deployers) asks for a short audible disclaimer at the start in plain language (or an earcon with awareness measures), reminders at intervals for long-form or live audio and after interruptions such as ads, and a visual label when a screen is available. The Commission guidelines' examples count voice cloning of real podcast presenters as a deep fake. They do not count fictional-character voices (with no deception about the narrator) or loudness/noise processing that leaves the words unchanged. Speech in a specific real person's voice needs marking, assistive voices (AAC, customised neural voices) do not, music in an existing artist's style is a creative-work deep fake, and playlists are out of scope. The examples are illustrations, not determinations.
• articleHow do the C2PA security requirements differ for Edge, Backend and Distributed generator products, and what does the public Conforming Products List record about them?
  The security requirements apply to all classes unless noted. Edge binaries must not embed authentication secrets; Edge and Backend subsystems must authenticate each other; Distributed and Backend products have extra documentation and monitoring requirements. The public CPL lists 240 generator products, most at assurance level 1.
• articleWhat changed in the C2PA specification between 2.1 and 2.4 (soft bindings, ingredients, actions v2), and which version should an agent cite?
  Per the C2PA 2.4 version history: 2.1 added ingredients v3, BMFF v3 hashing, time-stamp manifests and a URN namespace; 2.2 added the Soft Binding Resolution API and ingredient fields for soft-binding manifest recovery; 2.3 deprecated older actions and ingredient versions and added soft-binding validation; 2.4 (April 2026) added crJSON, repository-receipt, environmental-sustainability and AI-disclosure assertions, plus HTML and structured-text embedding. The data-mining assertion was removed in 2.0, and that topic now lives in CAWG. The spec site root redirects to 2.4, so cite 2.4 with its version and date, and older versions only for history.
• articleWhere in a repository's public text can I look for the origin of a specific test fixture when no README states it, and what do those places say for c2pa-rs?
  Without a fixture README, the public trail for a fixture's origin is its earliest commit, any in-code labels and project docs. For earth_apollo17.jpg the trail ends at the 'Initial public release' commit of 2022-05-23, with no origin or licence, so the trail dates a file but does not establish rights.
• articleHave ACES maintainers published anything on why aces-docs says a constant lower-hull gamma 1.14 while aces-core CTL uses 1.14 + 0.07*log_peak?
  No ACES maintainer statement explains the mismatch as of 2026-10-22. Issue and PR searches in aces-docs, aces-core and aces-output return nothing for 'hull' or 'log_peak'. The docs page has had only housekeeping commits since the 1.14 sentence, and current aces-core main still scales the lower-hull gamma (and several other gamut parameters) with log_peak.
• articleWhat do the SAG-AFTRA 2023 and 2026 MOA texts each say about Employment-Based Digital Replica definition and Synthetic use, and where do the two wordings visibly differ?
  Comparing the 2023 MOA with 2026 Attachment D: the 2026 text keeps the Employment-Based Digital Replica criteria. A footnote treats an asset made without the performer's physical participation as such a replica for Section 64.A.(2)(b) purposes, with no creation payment. The Synthetic provision gains a stated principle strongly favouring human performance and a notice, meeting (10 business days) and failure-to-agree (15/20 business days) sequence, which sunsets on 2030-06-30 back to the 2023 wording. New text covers footage used to train GAI systems for Synthetics. Because the 2026 PDF is marked up, some final wording is inferred.
• articleWhere can an agent read the SAG-AFTRA digital-replica Section 64/64.1 text publicly, and is any of it a clean consolidated version?
  Publicly, the SAG-AFTRA AI terms are readable only as MOAs on the AMPTP contracts index: the 2023 MOA adds them as a bracketed Section [XX], and the 2026 MOA (fully executed 2026-05-27) restates Sections 64 and 64.1 in Attachment D as a marked-up text where struck and inserted words run together when extracted. No consolidated 2023 or 2026 agreement is listed, although the 2026 MOA refers to a 2026 Codified Basic Agreement.
• articleWhat does a SWHID guarantee and not guarantee about availability, resolution and trust, and what is its standardisation status?
  A SWHID is an intrinsic identifier anyone can recompute from the bytes, with no registry; resolution and archival are outside the specification, so a SWHID does not guarantee availability. Software Heritage offers a web and /api/1/resolve/ resolver; Identifiers.org (core ids only) and N2T also resolve. swhid.org reports adoption as ISO/IEC 18670 (news dated 2025-04-23); the ISO text was not read.
• articleWhich machine-readable TDM reservation mechanisms exist, at which versions, and what do their primary texts say about legal status?
  Four mechanisms are read in primary text. EU Directive 2019/790 Art. 4(3) conditions the general TDM exception on no express reservation 'such as machine-readable means' for online content (recital 18: including metadata and website terms). W3C TDMRep (Final CG Report 2024-05-10) describes itself as a technical answer to Art. 4: a tdm-reservation flag and policy URL via a well-known file, HTTP headers (which supersede the file), HTML meta and EPUB/PDF XMP. IPTC's plus:DataMining (2023.1) warns that laws may overrule it and that absence is not permission. CAWG training-mining 1.1 uses allowed/notAllowed/constrained per use, with constrained treated as notAllowed absent more information. No text maps one to another, and no legal conclusion is drawn.
• articleRelease by release from 2.0 to 2.3, what changed in the actions assertion rules (actors, mandatory first action, multiple assertions, deprecation, gathered)?
  From each version's history: 2.0 (January 2024) removed the actors field from the actions assertion; 2.1 (September 2024) made c2pa.created or c2pa.opened mandatory and allowed multiple actions assertions; 2.2 (May 2025) clarified action assertions; 2.3 (December 2025) deprecated old assertion versions, allowed gathered actions and recommends declaring allActionsIncluded. c2pa.actions.v2 occurs 9, 16, 15 and 20 times in the 2.0-2.3 pages.
• articleI see c2pa.actions or c2pa.actions.v2 in a manifest: what is the difference, which should a new tool write, and which spec version should I cite?
  Two labels exist: c2pa.actions (v1, deprecated in 2.3) and c2pa.actions.v2 (defined from 1.3). Validators in 2.1-2.3 handle both, so readers should accept both; new tools should write v2 and state the spec version they follow.
• articleIn which C2PA specification version did the c2pa.actions.v2 assertion first become a defined assertion, and what did earlier versions say about it?
  In the version-pinned C2PA pages, 1.0 and 1.2 mention c2pa.actions.v2 only as an example of label versioning; 1.3 (its own history dates it April 2023) is the first to define the v2 actions assertion, with a separate actions-map-v2 CDDL rule. Dates come from the specification's own history, not independent release records.
• articleWhich contrast and text-spacing thresholds are normative today under WCAG 2.2 (1.4.3, 1.4.6, 1.4.11, 1.4.12), how are ratios computed, and what level is each?
  WCAG 2.2 (REC, 12 Dec 2024) defines contrast as (L1+0.05)/(L2+0.05) on relative luminance and requires 4.5:1 for text and 3:1 for large text (18 pt or 14 pt bold) at AA (SC 1.4.3), 7:1 and 4.5:1 at AAA (SC 1.4.6), 3:1 for UI components and meaningful graphics against adjacent colours (SC 1.4.11), and no loss of content when users set line height 1.5, paragraph spacing 2, letter spacing 0.12 and word spacing 0.16 times the font size (SC 1.4.12). The text scopes these to web content; applying them to video, print or images is an extension.
• articleWhich public HRTF/SOFA-convertible measurement sets state terms an agent can check (MIT KEMAR, Viking), and where is the SOFA catalogue?
  MIT's KEMAR HRTF data (1994) are free to use without restriction provided the authors are cited (710 positions, 44.1 kHz); the Viking HRTF dataset v2 on Zenodo (2020) is licensed CC BY 4.0 in its record metadata and distributed as SOFA files plus 3D scans. Terms for third-party SOFA conversions of KEMAR were not readable (SOFA hosts refused connections to producer).
• articleAcross Zenodo, Software Heritage and the Wayback Machine, what exactly does each identifier pin (bytes, record, capture time), and how should an agent combine them to cite a specific version of a work?
  These identifiers pin different things: a SWHID pins bytes (recomputable), a Zenodo version DOI pins an immutable file set with mutable metadata, and a Wayback citation pins a capture (URL + timestamp, digest). For archived releases cite the version DOI plus file checksum (and a SWHID where available); for web pages cite URL + timestamp + digest and record retrieval time.
• articleHow do I cite one specific Wayback Machine capture of a web page and tell whether two captures have the same content?
  A Wayback capture is cited by original URL plus 14-digit timestamp; CDX rows also give mimetype, status, digest and length, and equal digests (e.g. two 2002 example.com captures) indicate equal stored payloads, though the README read does not say how the digest is computed and lists two sections it does not contain.
• articleWhen should I cite a Zenodo version DOI and when the concept DOI, and what do I see in the record to tell which is which?
  Zenodo issues a version DOI per version and a concept DOI for all versions; a published record's files and DOI cannot change while its metadata can; the concept id resolves to the latest version (705645 redirected to 800648 on 2026-10-22). Cite the version DOI plus a file checksum to pin bytes, and the concept DOI only to point at the work.
• articleWhat are the five SWHID object types and what exactly is hashed for each, so I can tell what a given swh:1:... string identifies?
  SWHID core identifiers are SHA-1 hashes over Git-style serializations: content ids hash only the file bytes behind a 'blob' header (no file name), directories hash sorted entries with modes, and revisions/releases also hash metadata such as author, committer, timestamps and message, so the same tree committed twice yields different revision ids but the same directory id; Git compatibility is described as incidental, not guaranteed.
• articleFrom footage of a test chart, what can an agent measure (sharpness, tone response, noise, dynamic range) and what prerequisites do open test documents imply?
  From chart footage an agent can estimate SFR and per-patch SNR only after linearising the tone response (gamma), avoiding clipped edges and handling edge angle; log encoding, sharpening, noise reduction and heavy compression break linear-model assumptions, so video-camera measurements need the non-linear treatment or must be labelled as approximate. No cinema/video-codec-specific method was found in these sources.
• articleWhat is ISO 12233, what does its public scope say it covers, and what does the slanted-edge test actually do?
  ISO 12233:2017 (now withdrawn, replaced by ISO 12233:2024 per the ISO page snapshot) covers resolution and SFR measurement for electronic still cameras; the edge-based e-SFR method (in the standard since 2000) takes an edge profile, differentiates, windows and Fourier-transforms it, normalised to unity at zero frequency; the 2017 chart's 4:1 contrast edges are less likely to clip than the old chart's.
• articleHow can an archive verify that an FFV1/Matroska transcode is lossless, and what do MediaConch policies and FADGI say?
  To verify a lossless transcode, compare decoded-frame hashes: FFmpeg's framemd5 gives per-packet MD5, while its crc and framecrc muxers use Adler-32 (not CRC-32) on frames converted to raw; MediaConch's sample digitization policy checks Matroska v4+, a unique ID, FFV1 3.4+, GOP size 1 and slice CRCs, plus SD geometry.
• articleWhat fixity and error-detection does FFV1 (RFC 9043) and Matroska (RFC 9559) build in, and how do documented tools use it?
  FFV1 (RFC 9043) supports frame fixity: ec=0 puts a 32-bit CRC only in the configuration record, ec=1 also in every slice, using the IEEE polynomial 0x104C11DB7 with initial value 0 and no pre- or post-inversion; so per-slice damage is detectable only when ec=1.
• articleWhat do the BBC Casualty accessible-audio trial, a Fraunhofer-linked thesis and Dolby's white paper report about how much viewers adjust dialogue-versus-background balance, and how variable it is?
  Older listeners in one small study (11 listeners aged 66-75, no hearing devices; 2020 bachelor thesis using the Fraunhofer IIS listener pool) chose speech-to-background differences far above the 8.4 LU default: about 15.3 LU with separately produced objects and 13.5 LU with dialogue-separated material, with wide interquartile ranges (11.8-20.8 and 9.8-17.1 LU). The BBC's 2019 Casualty trial let viewers slide between the broadcast mix and a simplified mix; Dolby reports widely differing preferences; SVT chose one fixed Enhanced Speech track. Together they support personal control where separated or object audio exists, with a pre-rendered alternative as the simpler delivery (cross-source synthesis, not a measured comparison).
• articleWhen an agent follows the repository ids, version names and sample counts printed in the 1X cards and README, which of them have drifted (renamed ids, v1.1 vs v2.0, 450 vs 250 test samples) and what must be pinned?
  At observation, the ids printed in the cards (1x-technologies/worldmodel and worldmodel_raw_data) answer with HTTP 307 to world_model_tokenized_data and world_model_raw_data, so a pin should use the sha, not the name. The GENIE cards and the README leaderboard refer to challenge v1.1 and data/val_v1.1, while the tokenised-data repository (sha 42e3e12f..., lastModified 2025-04-20) describes v2.0 as replacing v1.1 and still lists the v1.1 folders. The two dataset cards give different test-set sizes (450 vs 250 samples). These are observations of text and listings, not a history of changes.
• articleHow do WorldRoamBench's single-H20 throughput measurements for LingBot-World, HY-World 1.5, Matrix-Game 2.0/3.0 and Yume 1.5 compare with the authors' stated FPS, and what can and cannot be concluded?
  WorldRoamBench v5 (AMAP CV Lab, Alibaba Group; not the authors of the open models) timed 10 open models on one NVIDIA H20, 20 runs, as chunk frames over wall-clock chunk time. Measured: Matrix-Game 2.0 6.26 Hz (640x352, 12-frame chunks), Matrix-Game 3.0 5.34 Hz (1280x704, 40), Yume 1.5 6.49 Hz (1280x704, 32), LingBot-World 1.79 Hz (832x480, 12), LingBot-World 2.0 1.75, HY-World 1.5 1.30 Hz (832x480, 16). Against the authors' figures (LingBot 16 fps on one GPU node; Matrix-Game 2.0 25 FPS on one H100 per C3-B062-W13-05) these are lower, but the hardware, GPU count and settings differ, so none of the gaps shows an author figure to be wrong.
• articleWhat do contributors and users report when they time LingBot-World-Fast themselves on 8xH100 (chunk times, cold start, whole-clip time), and how does it compare with the authors' 16 fps / under-1-second statement?
  Non-author timing logs for LingBot-World-Fast (14B, 480x832, 81 frames, 8xH100) come from a contributor's profiling pull requests: steady chunks about 1.0 s, chunk 0 about 7.2 s, VAE encode+decode about 5.6 s (26% of generate()), whole generate() 22,146 ms unoptimised, 13,523 ms after three merged PRs and 11,668 ms with a fourth, unmerged PR. If a chunk is 12 frames (WorldRoamBench's chunk size; the PR sample uses chunk_size=3), steady-state is about 12 frames/s, below the authors' 16 fps; a separate user reproducing on 8xH100 with NVLink and FA3 saw about 2.5 s/chunk around chunk 7 (about 4.8 frames/s on the same assumption), twice the 1.21 s another user reported. The discrepancy was not resolved in the thread. No key-press-to-frame measurement was found.
• articleAt the archive handoff (access file or package to FFV1/Matroska, BagIt and PREMIS/OAIS), what does each primary text guarantee, which package-level information has no home in the archive container, and which records must be written explicitly?
  Integrity checks work at different layers: FFV1 with ec=1 stores a CRC in every slice (MediaConch's sample digitisation policy requires per-slice CRCs), ffmpeg's framemd5/framehash hash decoded frames so two files can be compared by content, and a BagIt 'valid' result only says every manifest checksum matched the stored files. PREMIS 3.0 leaves which events to record to the repository and provides eventOutcome/eventOutcomeDetail for results, so validation outcomes are not automatic archive metadata. Matroska (RFC 9559) can flag hearing-impaired, visually-impaired and text-description tracks, so access roles have a place if a transcode step sets them; the LoC FFV1/Matroska page says timecode support is still in development, so DCP/IMF structure and timecode need separate records (no inspected source maps CPL reels/edit rate/KDM into Matroska).
• articleWhat is the status and content of the IETF AI Preferences working group drafts (vocabulary and attachment), and what do they claim about the effect of expressing a preference?
  The IETF aipref drafts are working documents, not RFCs: vocab-08 defines usage categories (train-ai, ai-use, search) and a serialization and states it does not reflect working-group consensus; attach-05 defines a Content-Usage HTTP header and robots.txt rule and would update RFC 9309 if approved. Preferences are statements; enforcement is out of scope.
• articleWhat do RFC 9309 and the major AI crawler operators' documents say a robots.txt rule does for training and for AI features, and what do they not promise?
  RFC 9309 defines robots.txt as a request protocol (not access authorisation) with no AI-specific terms; AI operators publish separate tokens with different stated effects: OpenAI GPTBot (training) vs OAI-SearchBot (search) vs ChatGPT-User (user-initiated; robots.txt may not apply), Google-Extended as a control token with no separate user agent, and Anthropic's three robots. A robots.txt rule is evidence of a stated preference and the operator's stated policy, not of observed crawler behaviour.
• articleAt UniVRM v0.131.4, how does the VRM 0.x to 1.0 migration code map each meta field and what defaults does it set?
  UniVRM's VRM 0.x to 1.0 meta migration begins from the most restrictive 1.0 licence values and only loosens what 0.x fields state; the 0.x licenseName (e.g. CC0, CC BY) is not mapped to any 1.0 field (TODO), so 1.0 fields with no 0.x source (redistribution, modification, credit, political/religious use, etc.) stay restrictive after migration, and unrecognised values throw.
• articleHow does Europeana's own code match or categorise legacy www.europeana.eu/rights URIs, and how common were they?
  Europeana's corelib resolves rights URIs tolerantly: after lower-casing and https-to-http rewriting, even 'exact' mode is a substring test (to accept version or language suffixes), and loose mode matches short patterns and, by the code's own comment, malformed strings. A test fixture lists legacy europeana.eu/rights values in the millions (rr-f about 8.6 million) at an undated snapshot; this is a fixture, not a published statistic.
• articleWhat do Adobe Stock and Shutterstock say about model and property releases for commercial versus editorial content?
  Both platforms require a model release per recognisable person for commercial content and property releases for listed property; Shutterstock accepts editorial content without releases (credentials for some events), while Adobe Stock does not accept traditional editorial content and has a separate illustrative-editorial category.
• articleWhich official IPTC reference images and interoperability tests exist for checking embedded photo metadata, and is a reuse licence stated?
  IPTC's reference-images page describes images whose metadata fields spell their own names for interoperability testing but states no reuse terms; terms may sit in the image metadata or elsewhere (unchecked).
• articleIs there an official C2PA public test-file set for validators, and what licence does it state?
  The C2PA public test-file repository is CC BY-SA 4.0 as a whole (share-alike applies to derived test packs) and, at this pin, its 2.2 directories hold only READMEs and .gitkeep placeholders, so no C2PA 2.2 media test pack is present there.
• articleIs there an official public PNG test-image suite and what permission text does it carry?
  PngSuite's licence grants use, copying, modification and distribution of the images for any purpose without fee; it does not separately mention hosting, sale or model training.
• articleDoes Chatterbox mark its generated audio, how can I check whether a clip carries that mark, and what does the code actually do?
  Chatterbox's TTS and voice-conversion classes construct Perth's implicit watermarker and apply it before returning audio (the producer reports the same in the multilingual and turbo paths), and the README provides a detection snippet. Robustness claims are the project's own. There is no consent check in the code read. If Perth's implementation fails to import, the class is None, so Chatterbox construction would fail rather than silently skip watermarking (curator code inference, not executed); since the code is MIT, modified copies can omit the call.
• articleWhat must an attribution notice contain under CC BY 4.0 and Apache-2.0, how does Creative Commons advise assembling it, and where do licence texts and platform pages impose different or no attribution duties?
  CC BY 4.0 requires, where supplied, keeping creator identification, copyright notice, licence and disclaimer notices and a link to the material, plus stating modifications and the licence; Apache-2.0 requires prominent change notices in modified files, retained attribution notices and the NOTICE file contents. Capture these at intake; the CC TASL guide is advice narrower than the legal text (producer reading of the guide).
• articleAt the master-to-package handoff (IMF or DCP), what do the open validators actually check on structure, hashes, picture and sound, and what stays unchecked?
  'Verified' means different things across open DCP validators: libdcp checks asset hashes only when the option is on and the file is under a configurable size limit, and ClairMeta check profiles can bypass the PKL hash test and downgrade checks to warnings. A stored validation report is interpretable only with its options. No inspected validator checks creative decisions (grade, mix levels).
• articleWhich audiovisual-media and e-book products and services does Directive (EU) 2019/882 (the EAA) cover from 28 June 2025, and what are the main carve-outs and transition dates?
  The EAA covers consumer terminals used to access AV media services and e-readers (products), and services providing access to AV media services (incl. EPGs) and e-books/dedicated software (services) after 28 June 2025, not AV content itself. The 28 June 2030 transition lets service providers keep using products already in use; it is not a general deadline extension. Microenterprises providing services are exempt.
• articleWhat does the VGHF/SPN reissue-market study report about how many classic games remain commercially available, and what does it say about emulation and library access?
  The VGHF/SPN study (v1.1, 2023) estimates 13.27% of pre-2010 US-released games are 'in release' via rightsholders (n=1500, stated ±2.5% at 95%), with lower rates for older platforms. The stated margins are conservative maximum-variance or finite-population figures (curator check). The report states an advocacy purpose (exemption debates).
• articleWhich OpenUSD changes from v25.05 to v26.11 can break existing assets, tools or builds, and what migration step does each note give?
  OpenUSD 26.11 rejects binary crate files older than 0.4.0 by default (update with usdupdatecrate; 0.4.0-0.8.0 still read with a warning; a build constant can restore older reads), removes prim relocates, and treats physics:rigidBodyEnabled = false as if RigidBodyAPI were not applied (collider ownership and joints move to the nearest enabled ancestor), which can change existing physics scenes.
• articleHow does Core Specification v1.0.1 relate to OpenUSD releases, and which OpenUSD behaviours documented after the spec text diverge from or are absent from it?
  The spec says it derives from OpenUSD and treats OpenUSD behaviour as expected where the spec is silent (an unintentional omission). Spline inner loops are specified in v1.0.1 but deprecated in OpenUSD 26.11, so a spec-compliant implementation carries a feature the reference implementation now warns about; the removal of prim relocates in 26.11 is consistent with the spec listing only layer relocates.
• articleWhich exact rights URIs will Europeana accept in edm:rights, and does it still accept its old www.europeana.eu/rights/... URIs?
  Europeana's accepted-URI page (v36, 2026-08-18) lists six CC licences, CC0 and PDM, and six RightsStatements.org statements (CNE, InC, InC-EDU, InC-OW-EU, NoC-NC, NoC-OKLR), with no europeana.eu/rights URIs; older records may still carry the 2015-era europeana.eu/rights/rr-* URIs, whose current status the page does not state. The page also spells one CC 1.0 row 'by-nd-nc'.

More