VALÉORINE Comics
Benchmark protocol
Qualification is incomplete. Catalogue coverage, successful software tests and a match to a known cover do not prove identification accuracy on an unknown physical copy.
Independent photographic validation
No independently adjudicated photographic benchmark is published with this release. Software tests and reference-image matches do not establish these accuracies.
The pilot requires at least 100 documented editions and 1,000 authorised real photographic episodes, split by physical copy and adjudicated independently. These are acceptance targets, not measured results.
| Measure | Target | Result |
|---|---|---|
| Work top-1 accuracy | ≥ 99% | Not measured |
| Candidate top-5 recall | ≥ 99% | Not measured |
| Auto-confirmation precision | ≥ 99.5% | Not measured |
| False confirmation rate | < 0.5% | Not measured |
| Unknown-object rejection | ≥ 98% | Not measured |
| Critical OCR fields F1 | ≥ 97% | Not measured |
| Ambiguities resolved within two further views | ≥ 85% | Not measured |
No dataset version or frozen benchmark manifest is published for this release. No worldwide superiority or economic valuation is asserted.
Remaining acceptance work
Local photo-text reading assists transcription and requires user review. Its accuracy across languages and editions, and fine-grained visual recognition, still require independent qualification. Valuation requires documented completed sales and a chronological backtest. Private collection, storage, transfer and restoration require authenticated acceptance tests.
Build identity and machine-readable evidence
- Source commit
- Unavailable
- Built at
- Unavailable
- Deployment timestamp
- Not independently recorded