Datasheet for the COVER-Fish Reference-Update Benchmark
Purpose
This is a versioned environment for measuring how changes to a reference corpus alter a frozen fine-grained recognition system. It separates source-package membership, aggregation rules and query domains so that gains and regressions can be reported together and source layers can be rolled back.
Contents
- Five source manifests (S0–S4) with provenance, rights and content hashes.
- Eight canonical gallery states from FishBase R0 through +Commons.
- FishBase 25.04 taxonomy and final row/prototype bindings.
- QT26-QC and Q-INT benchmark manifests.
- Transition ledgers and compact result evidence.
- Fixed dependency lock, replay examples and verification tools.
The final clean gallery state has 107,722 rows and 19,144 species prototypes. Frozen tensor and image archives are dependencies, not duplicated payloads.
Tracks
Gallery-state replay changes state membership while freezing the scoring stack. Aggregation replay changes the declared aggregation rule while freezing rows and embeddings. Neither track treats the public evaluation labels as an acquisition optimization target.
Rights and access
The control files are fully downloadable. Third-party source images retain their row-level terms and are available through the fixed Full Payload Archive or source pointers. Licence-filtered TSV files identify rows but are not physical image subsets. Derived embeddings do not reproduce source pixels; underlying licence and attribution remain bound through row IDs and manifests.
Known limits
The benchmark is instantiated on fish recognition and one frozen BioCLIP 2.5 embedding surface. Source packages differ in identity assurance and domain. Some Level B source rows are pointer-only or permission-required. Cross-domain museum analysis requires the separately distributed Fish-Vista benchmark.