VISUS public partial 60 Hz evidence

GazeForge includes a narrow, real-data VISUS-derived evidence record recovered from a public University of Stuttgart VISUS-supervised software repository. This tranche is intentionally separate from the canonical VISUS source-audit and Frozen Evidence workflows.

What was recovered

The public repository Maurice189/eye-slitscan is pinned at commit a8ea2402936122f9e5c98152460bd16a4ba97740. Its README identifies the project as a bachelor thesis at VISUS, University of Stuttgart, supervised by Kuno Kurzhals.

At that exact commit, GazeForge recovered and hash-verified four files without vendoring their bytes:

• 01-car pursuit.xml, a 625-frame dynamic-AOI annotation for 01-car pursuit.avi;
• P1A-01-car pursuit.tsv;
• P2B-01-car pursuit.tsv;
• P9B-01-car pursuit.tsv.

The naming convention matches the original VISUS benchmark convention of participant/group/stimulus exports. The AOI XML contains a 1920×1080, 25 fps stimulus with two dynamic AOIs: Red Car across frames 1–625 and White Car across frames 553–590.

Every source file is bound to both its upstream Git blob SHA-1 and a SHA-256 content digest. The raw source bytes are not committed to GazeForge.

Observed sampling contract

All three Tobii recordings are internally coherent 60 Hz exports:

Participant	Samples	Median interval	Inferred rate	Both-eye valid
P1A	1,499	16,625 µs	60.1504 Hz	93.73%
P2B	1,499	16,625 µs	60.1504 Hz	96.40%
P9B	1,500	16,625 µs	60.1504 Hz	96.60%
Total	4,498	—	60.1504 Hz	95.58%

The probe follows the pinned upstream Tobii parser and AOI-mapping semantics: mapped fixation points use media coordinates, fixation events are deduplicated by fixation index, and AOI membership uses the active ViPER bounding box with half-open pixel bounds.

Dynamic-AOI results

Across the three recordings:

• 185 fixation events were recovered;
• 166 fixation events intersected at least one dynamic AOI;
• fixation-event AOI-hit fraction: 89.73%;
• total fixation duration: 75,039 ms;
• fixation duration associated with at least one dynamic AOI: 70,179 ms;
• duration AOI-hit fraction: 93.52%;
• 3,679 / 4,498 samples intersected at least one dynamic AOI (81.79%).

Participant-level fixation results are:

Participant	Fixations	AOI-hit fixations	Event hit fraction	AOI-hit fixation duration
P1A	77	67	87.01%	22,280 ms
P2B	51	48	94.12%	23,903 ms
P9B	57	51	89.47%	23,996 ms

The Red Car dominates AOI membership, as expected for the car pursuit stimulus. White Car is active only late in the stimulus and can spatially overlap Red Car, so AOI-specific counts are not mutually exclusive.

Reproducibility

The isolated workflow .github/workflows/visus-public-partial-probe.yml downloads the four exact commit-addressed upstream files and recomputes each Git blob identity before analysis. The first successful probe was workflow run 33924318147, head 6d6a80aa5ea9b0ae6ad1878786ecf5346ad7196b, with deterministic probe fingerprint b1a301151ffae7efefdfccce647f509ec2b7ffe911b88b4979834ca526d1d4b1.

The committed evidence record is:

validation/evidence/visus-public-partial/visus-public-partial-evidence-v1.json

It is validated by gazeforge.visus_public_partial. The v1 fingerprint is immutable; a corrected or expanded corpus must use a new evidence version.

Claim boundary

This evidence is deliberately narrow.

• It covers exactly 3 public Tobii recordings × 1 VISUS stimulus.
• It does not resolve or replace the full original 25 participants × 11 stimuli VISUS benchmark.
• The original dataset license remains unresolved; GazeForge records an analysis-use basis but does not assert unrestricted redistribution.
• GazeForge does not redistribute the source TSV/XML bytes.
• It does not establish human-human agreement.
• It does not constitute model validation.
• It does not pass or weaken the canonical VISUS Frozen Evidence gate.
• It is Tobii 60 Hz evidence, not native Gazepoint GP3 evidence.

The existing full VISUS source audit therefore remains fail-closed and unchanged.