Skip to content

Hollywood2 participant-ledger intake

GazeForge provides a metadata-only, fail-closed intake for an original Hollywood-2 eye-movement archive that the caller has legitimately obtained and is authorized to inspect. The intake exists to recover authoritative participant/task-group metadata without importing the restricted raw gaze corpus into the repository.

Why this gate exists

The public original-distribution metadata states that Hollywood-2 eye movements were collected from 16 volunteers, split into 12 active and 4 free-viewing subjects. Separate dissertation evidence reports that the distributed dataset lists unique subject IDs within task groups. The currently reviewed public download route is login-gated, however, so GazeForge has not recovered the authoritative archive README/subject ledger itself.

The canonical Hollywood2EM annotation repository independently exposes 16 stable three-digit filename tokens. That numerical match is not participant mapping evidence. Until an authoritative crosswalk is recovered and reviewed, those values remain opaque source tokens.

A separate publication-lineage audit also establishes that the author-posted 2013 arXiv v1 describes 16 subjects, whereas the final 2015 TPAMI article describes 19 subjects and a broader task scope including scene-context recognition. GazeForge treats this as unresolved version/cardinality drift rather than automatically reconciling the cohorts. See Hollywood2 participant cardinality audit.

Safety and rights contract

The intake never downloads the archive. The caller must provide a local file named gaze_hollywood2.zip and explicitly affirm that the local copy was obtained under terms that authorize inspection.

Before reading metadata, the intake rejects:

  • unsafe absolute, traversal, drive-letter, or backslash ZIP paths;
  • duplicate member names;
  • symbolic-link members;
  • encrypted members;
  • archives or metadata surfaces exceeding bounded size/count guardrails.

Only small text-like members whose filenames indicate README, participant, subject, observer, metadata, licence/license, documentation, or information content are opened. Archive members are never extracted to disk, and non-metadata/raw gaze members are not opened.

The output contains archive/member hashes, sizes, and conservative marker booleans/counts. It does not copy README text or subject IDs into the report.

Local use

python scripts/inspect_hollywood2_participant_ledger.py \
  gaze_hollywood2.zip \
  --confirm-authorized-local-copy \
  --output hollywood2-ledger-intake.json

Both the raw archive and the generated local intake report are git-ignored. Do not override those guards to commit source bytes or locally derived participant metadata.

What a positive metadata marker means

Detection of phrases such as “unique subject IDs”, active/free-viewing group language, or subject-ID-like lines is a discovery signal only. It does not automatically establish any of the following:

  • that recovered identifiers have been accepted as reviewed authoritative evidence;
  • that a complete subject→task-group ledger has been recovered;
  • that a GIN filename token maps to an original subject ID;
  • that a GIN token identifies the active or free-viewing condition;
  • that participant-disjoint Hollywood2 model validation is now valid.

Any such promotion requires manual review of authoritative metadata and an explicit, separately validated crosswalk.

Frozen governance contract

The intake contract is frozen at:

validation/governance/hollywood2-participant-ledger-intake-v1.json

Protocol fingerprint:

b47cc0acea2ded577d140d38ca97ad71ce414d82a41b2675e5cac6fc2dc9135b

The contract fixes the current state as ready for authorized local archive metadata inspection, not participant mapping resolved. It explicitly keeps analysis authorization, redistribution permission, participant mapping, participant-disjoint validation, cross-dataset validation, Frozen Evidence promotion, and native-60-Hz/GP3 validity false.

Success criterion

This gate can advance only when a legitimately obtained original archive yields authoritative README/ledger metadata that explicitly resolves the original subject identifiers and task-group membership. Even then, a separate authoritative link is still required before any original subject ID can be equated with a Hollywood2EM GIN token. The 16-versus-19 publication-lineage ambiguity must also be reconciled before participant-generalization claims are made.