Extend the whole-word rule to every identity lookup. Once the reader has
decoded a manifest, no identity is re-derived from its raw bytes: not the
signer, not a signer-platform token, not the AI-tool fallback for the
generator, and not the SynthID vendor inference, in identify and in
metadata.synthid_source alike. The byte scans read the manifest store only
when the reader could not open it.
Claim-generator tokens match as whole words through one shared
claim_generator_word_platform (suno no longer matches Sunoco), for images
and video alike.
The verdict now reads SynthID from the decoded C2PA info when the metadata
dict lacks it; three synthetic Google Photos tests had relied on the byte
fallback for that.
Verdicts over 1,379 C2PA files and 468 fixture and capture files are
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Registry tokens were raw substrings, so the Canva token matched the "Nova
Canvas" agent of an Amazon Bedrock image. Raw manifest bytes cannot take a
word-boundary rule: across the 1,379 C2PA files in the local corpus, 374 of
37,325 store matches had a letter or digit beside them, nearly all CBOR text
headers and DER tags. Decoded strings have no such framing.
- Match signer, claim-generator, agent and video-marker strings with
registry_word_matches.
- When the reader decoded the manifest, never re-derive the signer from raw
bytes: an unregistered signer stays unknown. Bytes decide only for
manifests the reader cannot open.
- Add fal's signer "Features & Labels"; its attribution had relied on the byte
fallback finding the "fal-ai" generator prefix.
Verdicts over 468 images and 365 videos are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Nova Canvas images published in aws-samples were reported as Canva: with no
Amazon row, the byte fallback matched the Canva token inside the "Nova
Canvas" software agent. Add an Amazon Web Services row ahead of Canva, which
also names Nova Reel video.
When the reader cannot open a manifest, scan for signers in the C2PA store
the container locates (JPEG APP11, PNG caBX) instead of the whole metadata
region. A vivo X300 camera capture was labeled "C2PA signer: Adobe" from its
XMP toolkit string. The SynthID provenance scan reads the same store.
Adds the Nova Canvas and Nova Reel samples as MIT-0 fixtures and records the
public-sample search for the other conformance-list signers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Collect provider samples from Higgsfield, Runway and Dreamina plus Google
Photos, Gemini, Grok and YouTube re-encodes into data/captures/2026-09 with
a hash-checked manifest, and use them to fix attribution and false positives:
- Attribute aggregator outputs: Higgsfield job metadata, Dreamina's
re-signed manifests, Runway's own C2PA signer, TC260 video producers.
- Veto any visible video mark when a valid C2PA claim names another vendor
(a Gemini Omni clip read as Sora on steam), and gate Kling and Hailuo
matches on edge, anchor and white-label evidence.
- Veto the Gemini sparkle when trusted provenance names another vendor.
- Report SynthID as present for Google Photos AI edits: Google's checker
found it on all four measured edits, though manifests record no action.
- Add Vidu video and Wan image marks, TC260 audio placements, and record
the AIGC key in metadata records.
- Document the research sweep, the Dreamina badge decision, and verified
Google statements on SynthID in Photos.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
scripts/sync_c2pa_soft_bindings.py --check hits api.github.com to resolve
the latest upstream commit SHA, unauthenticated. That endpoint's 60/hour
rate limit is per source IP, which GitHub Actions runners exhaust fast
since many workflows share the runner IP pool -- this was failing the
lint job with "403: rate limit exceeded" on every push regardless of
whether the change touched anything related.
Send an Authorization: Bearer header built from GITHUB_TOKEN/GH_TOKEN
when set, scoped to api.github.com only (raw.githubusercontent.com is a
separate, much higher-limit CDN and doesn't need it). Wire the workflow's
own GITHUB_TOKEN secret into that step's env so CI actually has one to
send, raising the limit to 5000/hour.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The OpenArt visible-mark engine (commit 9d29d2c) was registered without
a data/fixtures/visible/openart/example.png gallery canary and without
an _ALTERNATE_SIZE entry in test_visible_fill_quality.py, breaking
TestGallery::test_every_registered_mark_has_an_example,
TestGallery::test_engine_detects_its_own_example[openart], and
test_cv2_mask_quality_smoke[alternate-openart] on every CI platform.
Generate the fixture via render_visible_examples.py (self-detects at
conf=0.968) and add the missing size entry.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Every existing text-mark engine (Doubao, Jimeng, Qwen, Baidu, Kling,
Yuanbao) is tuned to one vendor's exact wordmark silhouette, so none of
them match a bare "AI生成" strip with no brand name in front of it.
Confirmed on raiw-app production: a free `visible` run left a bottom-right
"AI生成" watermark completely unremoved because no detector fired.
Research found the wording is not tied to one dominant generator: vivo's
own support documentation confirms its Gallery stamps a bare "AI生成"
label after AI消除/AI扩图/大片模式 editing, and independent reporting
documents the same watermark on Xiaomi Gallery's AI-edit output. Since
the label is a GB 45438-2025/TC260 compliance stamp rather than one
vendor's brand mark, this adds one generic fallback (`generic_ai_label`)
instead of another per-OEM tuned config.
The detection silhouette is font-rendered (no logo element to capture)
via the shared scripts/render_vendor_silhouettes.py pipeline already
used for Qwen/Baidu/Kling/Yuanbao's own text-only marks. Calibration is
synthetic-only (no real captured generator screenshot was available);
the module docstring and DETECT_NCC_THRESHOLD comment carry that caveat
explicitly and this should be recalibrated once real captures exist.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
OpenArt.ai stamps free-plan generations with a semi-transparent white
"OpenArt" wordmark (bowtie/infinity icon + brand name) placed near the
CENTER of the frame rather than in a corner. Confirmed from a customer-
reported raiw-app production case where the mark went completely
undetected and unremoved, with no C2PA/EXIF/TC260 signal for identify
to fall back on.
- Add a new "cc" (frame-center) corner mode to the shared
TextMarkConfig/TextMarkEngine base, alongside the existing
corner/edge anchors.
- Add openart_engine.py, following the existing text-mark engine
pattern. No real capture exists (OpenArt requires an account, out of
scope to create), so the bundled alpha asset is a procedural
reconstruction of the public logo and the NCC threshold is an
explicitly-labeled provisional guess, not a measured calibration.
- Register the mark in watermark_registry.py, which auto-derives
identify's platform attribution.
- Add tests/test_openart_engine.py (synthetic composites only) plus
registry/identify test coverage, and document the mark and its
calibration caveats in README.md, docs/supported-signals.md,
docs/module-internals.md, and the skill's commands.md.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A decorative sparkle/glint glyph in stylized typography (e.g. a book-cover
title) can clear the visible Gemini sparkle's false-positive gate at
confidence just under _SPARKLE_FP_CONF, since it shares the same bright,
near-white, sharp four-point signature the gate uses to recognize a genuine
sparkle. Confirmed against a raiw-app production incident (2026-08-27) where
this caused the removal step to permanently smudge unrelated cover art.
Investigated tightening _SPARKLE_FP_CONF and the margin/gradient/saturation
checks (plus a prototyped local edge-density signal); each was disproved by
a constructed counter-example showing a genuine weak Gemini sparkle over a
noisy/textured photo background lands in the same confidence band with the
same signature, so no safe targeted fix exists without a real labeled
corpus. No detection/removal behavior changed.
Adds DEBUG logging at the exact FP-gate bypass path for future calibration,
a regression test (TestDecorativeGlyphFalsePositive) pinning the confirmed
false positive and the blocking counter-example, and a docs record of the
finding in the Gemini sparkle section of module-internals.md.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
pillow-heif's opener rewrites the EXIF orientation to 1 and stashes the real
value in info["original_orientation"], while libheif turns the raster on decode
exactly when the container declares an effective irot/imir transform. A camera
HEIC (stored raster, EXIF-only rotation) and a re-encoded one (rotation baked
as irot) both decode with Image.size matching the raster and the tag blanked,
so _read_display_tags dropped the turn for both (issue #105): a camera HEIC
went through erase and came out sideways with nothing on it saying so.
For HEIF the caller's orientation_applied claim is now ignored the way it is
for TIFF, and the container decides: isobmff.heif_transform_applied scans
meta/iprp/ipco for an irot with a non-zero angle or any imir, the orientation
comes from original_orientation, and a transform-bearing file drops the tag.
AVIF keeps the generic path: this plugin build neither blanks its tag nor
applies the transform there, and a camera-style AVIF's orientation stays
unreachable from Pillow (recorded in known-limitations).
Tests build the camera fixture without exiftool: a pillow-heif-written file
with its irot angle neutralized to the identity behaves like the reporter's
exiftool-built camera file. Mutation-checked in both directions: forcing the
detector either way fails exactly the tests whose property broke.
pre-commit: 1) maintain.sh - exit 2 at the uv-secure gate, the three triaged
no-fix advisories in docs/development.md (rechecked 2026-09-09: accelerate
1.15.0 and lightning 2.6.5 still carry no fix); ruff check+format clean,
pyright src/ 0 errors, 2368 passed 3 skipped run separately; 2) /simplify -
single pass, no fan-out available, one duplicate test helper inlined, no
other findings; 3) docs sync - heif_transform_applied, original_orientation,
irot, imir live only in module-internals.md and known-limitations.md, both
updated, no other refs; 4) CLAUDE.md - audited against the structure rule,
no additions; revise-claude-md: no CLAUDE.md changes
Both readers in the imread stack apply the TIFF orientation tag (cv2's
libtiff path under IMREAD_UNCHANGED too, and Pillow's TIFF plugin,
identically), so the visible path's blanket orientation_applied=False
carried the tag onto an already-turned raster. _read_display_tags now
ignores the explicit claim for TIFF and decides on the IFD's stored size,
which also keeps a hypothetical non-turning build correct: a stored-shape
raster still carries the tag. Guarded at the api seam, not just imwrite.
pre-commit: 1) maintain.sh - exit 2 at the uv-secure gate, the three
triaged no-fix advisories in docs/development.md (rechecked 2026-09-08);
ruff check+format clean, pyright src/ 0 errors, 2363 passed 3 skipped
run separately; 2) /simplify - single pass, no fan-out available, nothing
to fix (api-state plumbing rejected as disproportionate); 3) docs sync -
identifiers live only in docs/module-internals.md, updated, no other
refs; 4) CLAUDE.md - audited against the structure rule, no additions,
revise-claude-md: no additions
Resolve image_io.py and test_image_io.py the same way the rebased
adaptation branch does, so the branch carries the author's original
commit 1cfdf19 on top of current main.
commands.md names liblib_pill and the auto pipeline routing OpenAI
provenance to qwen-zimage, both of which only exist since CLI 0.39.0,
while probe.py kept MIN_CLI_VERSION at 0.38.0: a 0.38.0 install would
pass the probe's currency check and then reject the very mark key and
routing the skill hands it. The references now carry the established
(since CLI X.Y.Z) annotations and the floor rises to 0.39.0, per
docs/agent-skill.md.
The floor was hand-synced against prose only a person could read, so it
drifted exactly the way its own comment warns about, twice before. A new
guard derives the floor's lower bound from every (since CLI X.Y.Z)
annotation in the skill tree and fails when the floor lags the newest
one; combined with the existing package-version ceiling, the floor is
now pinned from both sides. Skill version 1.0.6 in SKILL.md, plugin.json,
and marketplace.json.
pre-commit: 1) maintain.sh - exit 2 (uv-secure on the documented
lightning/accelerate blocks, unchanged lock; core checks separately:
C2PA sync current, Ruff 1 fix + 1 reformat folded into this commit,
Pyright src/ 0 errors, 2361 passed 3 skipped); 2) /simplify -
single-pass four angles, clean; 3) docs sync - grepped MIN_CLI_VERSION
and since CLI over docs/ and rules/: both mentions stay accurate;
4) CLAUDE.md - improver grade A earlier this session, revise: no
additions
download-artifact@v8 extracts the only pattern-matched artifact directly
into the target path, without the artifact-name directory the resolve
step assumed, so every workflow_run verification died at
"No distributed release identity found" before checking a single
surface. The directory-name attempt check now applies only when the
artifact directory exists; a flat identity is still pinned by the
run-id and attempt-eligibility asserts. The workflow_run path of
5c6588b had never executed: v0.38.0's run failed on later checks and
the surfaces were confirmed by a manual dispatch.
The new test drives the real workflow script against the flat layout
and reproduced the production failure before the fix.
pre-commit: 1) maintain.sh - exit 2 (uv-secure on the documented
lightning/accelerate blocks, unchanged lock; core checks separately:
C2PA sync current, Ruff clean with no rewrites, Pyright src/ 0 errors,
2360 passed 3 skipped); 2) /simplify - single-pass four angles, clean;
3) docs sync - grepped distributed-release over docs/: the one mention
stays accurate; 4) CLAUDE.md - improver grade A earlier this session,
revise: no additions
Round unsharp output in its existing buffer before uint8 conversion, preserving
constant images across OpenCV builds. Compare explicit stored WebP decodes.
Use host-absolute launcher fixtures and restrict Ubuntu Bash harnesses to POSIX.
Document the CI findings and the unresolved Dependabot Transformers/tokenizers
candidate conflict. Keep the existing model dependency constraints.
Repair image and metadata publication contracts, detection routing, research
artifact validation, release verification, and the published agent skill.
Refresh the dependency lock and reconcile documentation with verified behavior.
Add regression coverage and record the completed audit and corrected VideoSeal
measurements. Preserve the existing Accelerate and Lightning versions.
Validation: Ruff and Pyright pass; 2358 tests pass with 3 skipped. The maintenance
gate exits 2 on the documented Accelerate and Lightning security findings.
bytedance generator class (surgical six-head provider.pt), tc260 group
veto, stable-name receipt-gate.npz with the threshold inside the
artifact, snapshot-tolerant loader, weights pinned to revision
4c276376 (published via photo-classify-freeze-2026-09-07 release).
The ignore rule landed with a comment naming the private product, its storage
provider, the private rule file, and the size of the customer set. That text is
the disclosure the rule exists to prevent, so it says what a public reader needs
instead: these trees are local-only and not publication-cleared.
`data/research/` gets the same tracked rule as `data/spaces/`. It held 68 GB
behind `.git/info/exclude` alone, which no clone has, so a fresh checkout could
commit the whole corpus with nothing to stop it, and a test now pins both
entries.
A first draft of that test also carried a denylist of private identifiers,
which published six of them into a tracked file to keep two out, and only its
own self-exclusion hid that. The mechanized half is now the one that discloses
nothing: no cloud GPU harness under `scripts/`. Reading the ignore entries
rather than the file text matters too, because a comment naming
`data/research/README.md` satisfied the first assertion with the rule deleted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The Content Seal manifest named the production GPU worker in its provenance
column and in two row keys, several docstrings described their sources as user
uploads from the private corpus, and that corpus was called by the storage
product holding it. Each is now what a public reader can check: a pipeline and
strength, material that is not publication-cleared, and a private corpus. The
corpus test follows the renamed rows. Deployment and tier detail goes the same
way, since which backend each hosting tier can afford is a product decision,
not a library fact.
Twelve harnesses that need a cloud GPU account leave scripts/. The research
pages that cite their findings name them as harnesses kept outside this
repository and stop implying a reader can rerun them; the tracked inputs and
recorded verdicts remain the durable record, and the offline analysis step
stays here with its test. A development invariant states the rule so the next
harness is not added back.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The previous commit slipped past a red test: the pytest exit code was
swallowed by a pipe in the shell chain. This pins the test to
4c2763766dd1c8c212d64e01e1cc3f166243c47d.
Provider classes now name the GENERATOR that rendered the pixels, never
the brand. bytedance joins openai/google/muse-image as the shared
ByteDance lineage (Doubao and Jimeng render with one model family;
measured: separate heads cross-fire 32/22 at 3.5x mass, the surgical
union holds 83.9% on the frozen test cell while openai/google/meta/no_ai
stay byte-identical to run1). tc260 stops being a public value: its head
remains as the group veto for the rest of China's generators (without
it, 207/379 China test rows would be falsely named openai or
muse-image), and the loader tolerates provider snapshots that predate
bytedance. The receipt gate moves model-side: stable-name artifact
receipt-gate.npz carries its threshold, head and threshold always load
from one file, and the publish script stages the stable gate from the
weights source. Microsoft stays metadata-only: 144 of its 253 bank rows
name an OpenAI model as their C2PA softwareAgent and none names a
Microsoft renderer.
pre-commit: 1) maintain.sh - exit 1 (uv-secure: lightning 2.6.5 via
trustmark, PYSEC-2026-3624 no fix released; per docs/development.md the
core checks ran separately: ruff clean, pyright 0 errors, 2059 passed
3 skipped); 2) /simplify - single-pass four angles, clean, one noted
skip (gate-test npz staging duplication); 3) docs sync - grepped
bytedance/receipt-gate.npz/provider values over README, docs, skill,
rules: no stale references; 4) CLAUDE.md - improver: no changes,
revise: no additions (learnings live in canonical docs)
Extend the development-only benchmark family from images to audio and
video, and close four research directions from the watermark-landscape
map with local, revision-pinned evidence.
Benchmark schema: media_type audio and video rows, audioseal and
videoseal adapters with media-checked adapter binding, sample-domain
audio fidelity (SNR with unbounded_identical), frame-domain video
fidelity (mean PSNR, changed-frame fraction), and per-frame temporal
evidence in videoseal detection records. The report accepts all three
media types and carries SNR columns.
Audio: shared pinned AudioSeal oracle, synthetic and system-voice
speech cohorts with matched negatives, codec, noise, removed, and
forged arms, and the provenance experiment proving visual-only video
cleaning leaves the audio bitstream and watermark verdict invariant.
Video: TorchScript-only VideoSeal oracle (the pip package is unusable
on macOS arm64 and its loader resolves paths relative to cwd), video
cohort with crf/scale/fps attacks and four-state arms, temporal
evaluation measuring the crf decay curve, the frame-aggregation matrix
(computed locally after finding the TorchScript build ignores the
aggregation argument), and real provider clips as carriers.
Forgery: double-embedding study showing the last writer wins completely
under both oracles, scoped remover-trace probe, and the forged/
removed cohort arms encoding each scheme's correct expected verdict
(presence rule for audioseal, message rule for videoseal).
C2PA resolution: measured the five vendors' resolver contracts and the
C2PA 2.2 18.10.5 privacy gradient; product decision recorded that this
tool never resolves over the network. A registry guard pins the
published resolution-API set.
Gate: ruff and pyright src/ clean, 2114 passed / 3 skipped; maintain.sh
stops only on the known unfixed lightning 2.6.5 CVEs, pre-existing.
