Evidence status¶
This page is generated from versioned evidence policy and exact repository artifacts. Status is not inferred from whether a model metric looks favourable.
Status bundle fingerprint: 64a7be9cbe165a5079c93da85aaeb93a0e997b9c3a0b12443c51c3069eff2913
| Dataset / target | Status | Evidence scope | Sampling origin |
|---|---|---|---|
| Gaze-in-the-Wild | Reviewed empirical evidence | external participant-disjoint task-agnostic event validation | derived-60hz |
| Hollywood2EM | Frozen empirical evidence | external source-token-held-out event validation | resampled |
| Lund2013 | Frozen empirical evidence | external expert-labelled event validation | mixed-native-and-derived |
| Native 60 Hz / GP3-class validation | Empirical execution pending | native device validation | native-60hz-required |
| VISUS public derivative | Bounded empirical evidence | partial public derivative empirical analysis | native-tobii-60hz-partial |
Dataset boundaries¶
Gaze-in-the-Wild¶
Reviewed empirical evidence. Reviewed exact-distribution participant-disjoint event-model evidence on a derived 60 Hz task-agnostic analysis grid.
- Reference strength:
human-reference - Versioned source:
validation/evidence/gaze-in-wild/gaze-in-wild-exact-participant-disjoint-model-validation-evidence-v1.json - Bound evidence fingerprint:
b2fe85ec7e5d5cd425c0cd2593742bab835686c3f560d8a6c06e9c6d67dc547a - Open boundaries:
- Complete authoritative TrIdx-to-task mapping remains unresolved; task-stratified claims remain blocked.
- The 60 Hz grid is derived rather than native 60 Hz/Gazepoint GP3 validation.
- Cross-dataset validation and quarantine exit remain unauthorized.
Hollywood2EM¶
Frozen empirical evidence. Frozen aggregate derived-60-Hz performance on source-token-held-out folds; the split unit is not a verified participant identity.
- Reference strength:
derived-human-reference - Versioned source:
validation/evidence/hollywood2/hollywood2-source-token-60hz-frozen-summary-v1.json - Bound evidence fingerprint:
e1f1c030f843e118ebd65520dfab8e872efb4ea3e1d520299a993b0ca00ddabf - Open boundaries:
- Exact annotation-repository licence identifier/text and dataset-specific analysis terms remain unresolved.
- Source-token-held-out evidence is not participant-held-out evidence.
- Native 60 Hz or Gazepoint GP3 validity is not established.
Lund2013¶
Frozen empirical evidence. Frozen external expert-labelled event-validation suite; 60 Hz model evidence is derived from native 500 Hz recordings.
- Reference strength:
expert-human-reference - Versioned source:
validation/evidence/lund2013/lund2013-suite-manifest.json - Bound evidence fingerprint:
5dc6d6336b505b0a2283fe64d478a27b0394c9568a86fc4eb4d2771b8d600f93 - Open boundaries:
- Derived 60 Hz results do not establish native 60 Hz or Gazepoint GP3 validity.
Native 60 Hz / GP3-class validation¶
Empirical execution pending. Native 60 Hz/GP3-class expert-labelled event validation remains an explicit scientific requirement.
- Reference strength:
expert-human-reference-required - Open boundaries:
- No native 60 Hz/GP3-class expert-labelled event corpus has yet cleared the frozen empirical validation gate.
VISUS public derivative¶
Bounded empirical evidence. Verified partial empirical evidence from public VISUS-supervised derivative Tobii 60 Hz files; no full-dataset model-validation or Frozen Evidence claim.
- Reference strength:
bounded-empirical-observation - Versioned source:
validation/evidence/visus-public-partial/visus-public-partial-evidence-v1.json - Bound evidence fingerprint:
80e008228e39c2b17bae99a526e2a0157c2c850ebe803c5a370abe9167efde14 - Open boundaries:
- Only a partial public derivative is covered; the original 25-participant × 11-stimulus VISUS benchmark is not recovered.
- Original source licence remains unresolved.
- No model-validation, human-human-agreement, Frozen Evidence, or native Gazepoint GP3 claim is created.
Interpretation rule¶
A stronger status requires an explicit reviewed policy update plus integrity-valid evidence. Missing, malformed, byte-changed, fingerprint-mismatched, or scientifically incompatible evidence fails closed during site generation. Synthetic/demo outputs are never promoted into empirical evidence, derived 60 Hz remains distinct from native 60 Hz/GP3 validation, and source-token-held-out designs remain distinct from participant-held-out designs.