GazeAudit authoritative validation matrix

This document is the publication-facing index of GazeAudit’s canonical scientific-MVP validation outcomes. It does not replace the protocol, source-lock, execution, or artifact records for any case study.

All three real-data cases were governed by the same integrity rule: scientific choices were fixed before outcome inspection, the controlled execution produced a checksummed artifact, the artifact was archived before reveal, and post-outcome changes are limited to verification, documentation, archival, reproducibility, and non-scientific infrastructure maintenance.

Case	Validation purpose	Canonical classification	What it demonstrates
GazeBase	multi-detector inferential robustness with a predeclared completeness gate	`incomplete`	GazeAudit does not drop failed detector specifications or manufacture a robustness claim when the declared specification space is incomplete
Korthals et al.	propagation of AOI measurement uncertainty into a paired scientific endpoint	`robust_negative`	a real-data scientific direction can remain stable under the frozen measurement-error model
Pedrotti/de Chambrier	sensitivity of a gaze-path-rate contrast to sampling and added missingness	`materially_fragile`	a scientific endpoint can retain direction yet fail the predeclared magnitude-recovery requirement

1. GazeBase multi-detector audit

Frozen protocol fingerprint: 3f64122f62cbc9762b0bd0e0b95c7fef6ff40c4700215ee4b90f005af77003b1.
Exact scientific commit: 609203dfc37f2ef4984de82b888660e7cc30675c.
Final artifact ID: 10284017265.
Artifact name: gazebase-partitioned-final-609203df.
ZIP SHA-256: 2fae4e99243e6738047a34b8dc24a183e8fb98606e4873723d43b8baa4ae937b.
Execution fingerprint: 35975d89dea728bd3aa146739dc58ea2bcb9322e357a44856b424fffb4183632.
Artifact-manifest fingerprint: 965ba4b83b06bd3da549b9ca9e36e79443e33b2bbde76d81f87ed7767f94f71d.
Fixed cohort: n=322.
IVT, IVVT, IDT, IDVT, and Engbert: 322/322 finite participant estimates.
NH and REMoDNaV: 0/322 under the already-frozen NaN-preserving/no-interpolation policy.
Canonical classification: incomplete because the 95% completeness gate failed.

The five computable detector estimates are descriptive archived outputs only. The canonical analysis does not silently remove NH or REMoDNaV.

See docs/case_studies/GAZEBASE_MULTIDETECTOR_PROTOCOL.md and docs/case_studies/gazebase_real_data_execution.md.

2. Korthals target-tracking AOI audit

Scientific workflow: korthals-scientific-execution #1.
GitHub run: 34721629856.
Execution commit: 3312423cb43b1d9da3877914455dd471aba81174.
Protocol-v2 fingerprint: 1cd4150c9341db145b8f6aee544f9634c67c46116e61e7e675ae3c50282d2d55.
Source-lock fingerprint: 8bcd9e1dfc8d6ca07c6222a20f563bf8b3bd67aed0de0d3b0e97628799de3a08.
Artifact ID: 10306308440.
ZIP SHA-256: 2eaab8171d9d5f70e138eb929554648cb6bed10b21646ab0acd7cfae4d607d11.
Execution fingerprint: 2db2b92fea136ba516f38a87a3c218cf90c9a2a69c55e8ea37b63cad8febb51b.
Audit scientific fingerprint: aa4c04af9c011fe997ef107c8250b20d6a20824b180abdeb16871b5b5a2fc768.
Artifact-manifest fingerprint: c55dbeb476197196a36d1c1af857bd6424e2db6fb1c7c5e5bdc94169bd380de0.
Hard effect: -0.09864087564504398.
Measurement-error expected effect: -0.07400756743578296.
95% measurement-model-induced interval: [-0.07652491273093086, -0.07146159624178107].
All 2,000 prespecified draws were below zero.
Canonical classification: robust_negative.

The interval is induced by the frozen measurement model. It is not a population confidence interval or Bayesian posterior interval.

See docs/results/korthals2026_target_tracking_aoi_v2.md.

3. Pedrotti/de Chambrier sampling + missingness audit

Scientific workflow: pedrotti-scientific-execution #1.
GitHub run: 34793103977.
Execution commit: 270d444c599363ebd25f25e8a95f5f3c78e55b02.
Protocol fingerprint: efb194ad2c492768962320198d281c98574eb2dc22291279bfd176a15606a0d5.
Source-lock fingerprint: d71949f035ecdf14b9e225b8ca441d291625767dc4a07fa4854882ff2dcae898.
Source-manifest fingerprint: ca676426a01c1d57dda93b62709d4e9040369e4193c0b44353be78d644a33b1f.
Artifact ID: 10328414078.
Artifact name: pedrotti-scientific-execution-v1-270d444c599363ebd25f25e8a95f5f3c78e55b02.
ZIP SHA-256: b411fe6a5f4ad7035ac606d60a952306c0b818d956f7b4930827b8cbe10ae8d5.
Results fingerprint: 5bafe0d1d1d0d054a681cb7136fc4e259c0f75e07d51653acde472c7c6e4dce9.
Execution fingerprint: 35dd53da10646400aca657c388f8b77a6a974c555de7efbd6f83faeb970b1eb6.
Artifact-manifest fingerprint: 05a986abe8a7be7a7ccbf2bf24b7467a3c623381d05d2dab8f4ae81b0618d661.
Reference long-minus-short gaze-path-rate effect: -75.37814047218419 pixels/second.
Sampling-family recovery: 0.4.
Canonical classification: materially_fragile.

Sampling estimates remained negative at all five target rates, but only 500 Hz and 250 Hz met the frozen 20% relative-deviation tolerance. At 125, 100, and 50 Hz, relative deviations were approximately 37.0%, 38.7%, and 72.7%, respectively. Recovery under added missingness also fell to 0.4 or lower for several 10%/20% families, which satisfies the predeclared material-fragility rule.

See docs/results/pedrotti_sampling_missingness_v1.md.

Interpretation boundary

These three outcomes intentionally cover different scientific states:

incomplete — the declared specification space cannot support a robustness classification;
robust — the prespecified scientific direction remains stable under the declared uncertainty model;
materially fragile — the endpoint fails the prespecified recovery criterion under defensible perturbations.

This is the scientific-MVP validation claim. GazeAudit is not validated merely because it produces a preferred or statistically significant result.