Gaze-in-the-Wild explicit task-mapping intake

GazeForge currently does not have an authoritative complete mapping from the numeric TrIdx values in the distributed Gaze-in-the-Wild ProcessData files to the four publication task labels. The reviewed public-source search remains frozen as unresolved by gaze-in-wild-authoritative-task-mapping-exhaustion-evidence-v1 (fingerprint c0e8e0da36d329d6445dace1a56646b89c9df5f4f406761311d588eddaf36aa7).

This intake exists so that a future explicit author, first-party repository, original-distribution, or publication-supplement record can be reviewed without weakening that boundary.

What this mechanism does

The workflow has four local phases:

1. Candidate — hash the exact source and a local transcription manifest, validate the syntax, canonical TrIdx domain, publication-task vocabulary, one-to-one structure, and current exhaustion-evidence binding.
2. Review template — create a pending record bound to the exact candidate fingerprint.
3. Sealed review — after a human reviewer fills the evidence fields, recompute the review fingerprint.
4. Certificate — replay the exact source and manifest and emit a non-empirical certificate only when every review gate passes.

The source and transcription manifest remain local. The candidate and certificate contain hashes, counts, canonical ledgers, and boundary flags, but not the raw TrIdx → task pairing.

Admissible source classes

A manifest may declare only one of these syntactic authority classes:

• author_statement
• first_party_repository_record
• original_distribution_metadata
• publication_supplement

Passing this syntactic allow-list is not authority verification. The reviewer must separately establish why the exact source is authoritative and record that evidence in the review.

Secondary repackaging, inferred directory order, publication task order, and a count match are not admissible substitutes for an explicit lookup.

Required mapping structure

The canonical trial-index domain is exactly 1, 2, 3, 4. The canonical publication-task vocabulary is exactly:

• Indoor_Walk
• Ball_Catch
• Visual_Search
• Tea_Making

A candidate may be partial so that newly found evidence can be inspected safely. A partial candidate cannot be approved. Promotion requires exactly four unique TrIdx values and exactly four unique canonical task labels.

The local manifest must use record type gaze-in-wild-explicit-task-mapping-source-v1, bind the exact source-file SHA-256, affirm that the source was obtained through an authorized channel, and contain mapping_entries with trial_index and task_label fields copied from the source.

The TrIdx 4 anti-inference gate

The existing evidence base contains secondary-only corroboration for part of the numeric mapping but does not authoritatively resolve the complete lookup. In particular, TrIdx 4 → Tea_Making must not be filled by elimination merely because three other task labels appear elsewhere.

An approved review therefore requires both:

• tridx4_explicit_in_source_verified=true, with resolved evidence showing that the authoritative source explicitly states the TrIdx 4 pairing; and
• no_elimination_or_order_inference_used_verified=true, with evidence that neither publication order nor directory order nor elimination supplied any pairing.

A complete-looking transcription still fails closed if either assertion is absent.

CLI

Use the local helper only after you possess a legitimate candidate source:

python scripts/inspect_gaze_in_wild_explicit_task_mapping.py candidate SOURCE MANIFEST \
  --output gaze-in-wild-task-mapping-candidate.json

python scripts/inspect_gaze_in_wild_explicit_task_mapping.py review-template \
  gaze-in-wild-task-mapping-candidate.json \
  --output gaze-in-wild-task-mapping-review.json

Edit the review manually. An approval must include a timezone-aware review timestamp, reviewer identity, resolved evidence text for every gate, and explicit true values for authority, mapping explicitness, transcription, publication-task semantics, source-version scope, direct TrIdx 4 evidence, and the no-inference rule. Rights, empirical-validation, and quarantine-exit promotion flags must remain false.

Then seal and validate it:

python scripts/inspect_gaze_in_wild_explicit_task_mapping.py seal-review \
  gaze-in-wild-task-mapping-review.json \
  --output gaze-in-wild-task-mapping-review-sealed.json

python scripts/inspect_gaze_in_wild_explicit_task_mapping.py certificate \
  SOURCE MANIFEST \
  gaze-in-wild-task-mapping-candidate.json \
  gaze-in-wild-task-mapping-review-sealed.json \
  --output gaze-in-wild-task-mapping-certificate.json

What a certificate does not establish

Even a genuinely reviewed mapping certificate creates no empirical result by itself. It does not claim task-stratified performance, participant-disjoint performance, cross-dataset validity, native 60 Hz validity, Gazepoint GP3 validity, acquisition-hardware cadence, quarantine exit, dataset rights, or raw-data retention rights.

Once an authoritative mapping is actually recovered and reviewed, a separate empirical tranche must consume the reviewed mapping and rerun the relevant Gaze-in-the-Wild validation before any task-stratified scientific claim is promoted.

Until such source evidence exists, Issue #1's authoritative TrIdx → publication-task mapping gate remains open.