SDK 0.7.2a1 · documentation review 2026-11-10.
Tested source: 8c6fda81b27c5c6c0c9539bbbc3daef7e285b1fa.
Exact-SHA EC2 CI: success. These results qualify the existing
wheel/source bytes, not arbitrary rebuilds or later documentation commits.
Environment and checks
Linux x86-64 CPU on the Oregon m6a.large test host (2 vCPU, 8 GiB). The full media container was bounded to 2 CPU/6 GiB, network disabled, non-root and read-only root with temporary storage. The installed wheel was tested.
| Check | Result | Interpretation |
|---|---|---|
| Full installed-media suite | 217 tests; 0 skips; 357.287 s | Suite runtime, not video throughput |
| Core Python 3.11 | 217 discovered; 79 expected optional skips; 11.172 s | Media-dependent paths run in the full suite |
| Core Python 3.12 | 217 discovered; 79 expected optional skips; 12.868 s | Same core scope |
| Core Python 3.13 | 217 discovered; 79 expected optional skips; 11.733 s | Same core scope |
| Core Python 3.14 | 217 discovered; 79 expected optional skips; 12.195 s | Same core scope |
| Package parity | 66 package files matched | Source, wheel, sdist and installed payload |
| Clean video/media wheel installs | Online setup and offline processing passed | Video-only excludes Torch/NumPy/Python TrustMark |
| Kit fresh exports | 5 verified | 2 images and all 3 video modes |
| Metadata-only recovery | 4 cases | PNG re-save / MP4 remux, no media re-encoding |
| External official c2patool 0.28.6 | 4 outputs | AI-source assertion, signature and asset hash; development issuer intentionally untrusted |
| Signed report negatives | Tampering and wrong-asset association rejected | Pinned report-key checks |
| Deployment checks | Restart persistence, registry read-only mount, key isolation, upload cleanup | Reference development deployment |
The 6 s, 512×320, 12 fps deterministic video contains 72 frames. Economy5s
selected windows [0,1] (4 frames), Economy2s [0,1,2] (6 frames), and
Resilient marked all 72. The checks explicitly distinguish cadence; this is
not a real-world durability corpus or performance comparison.
Artifact identity
| Artifact | SHA-256 | Bytes |
|---|---|---|
ariadne_mark-0.7.2a1-py3-none-any.whl |
05d2b28ae205dfc485d91cc06b7fdf09c53a4882a395e957a4bf15cd9a9931c2 |
140965 |
ariadne_mark-0.7.2a1.tar.gz |
db5945b325cbf48620bb842689dfcb323c0210266df536bc8821b8b027a6371f |
340500 |
Machine-readable summary and artifact metadata preserve the exact identities. The source README hash refers to the qualified archive, not a later edited README. The qualified artifacts are now published in Release v0.7.2a1. No PyPI publication or production-trusted hosted service is implied.
Limits of the result
Tests used development C2PA identities and pinned localhost TLS. Real CA/TSA arrangements, public DNS/TLS, load/capacity, backup restore and accessibility remain operator work. No new population error-rate, performance or remover campaign was run for this documentation/review fix. Existing spatial, temporal, remover and full-range 4:4:4 colour limitations remain. See the evidence index and test protocols.