Scientific Python · eye-tracking methodology
GazeAudit turns structural QC, gaze-measurement uncertainty, analysis-readiness decisions, and defensible alternative pipelines into an auditable robustness record.
pip install gazeaudit==0.2.0Choose your research task
GazeAudit is organised around methodological questions. Select a task to see the shortest route into the package.
Structural preflight
Map semantic columns with GazeStudy, identify non-finite coordinates and timestamps, missing identifiers, duplicate timestamps, and decreasing within-trial time.
Analysis-readiness governance
Declare researcher-owned thresholds, aggregate trial and participant diagnostics, compare policies, preview retained cohorts, and preserve a provenance-bound readiness record.
Measurement uncertainty
Represent gaze uncertainty transparently, propagate it into AOI membership, and inspect how measurement assumptions affect the endpoint instead of treating observed coordinates as exact.
Specification-space audit
Declare a finite decision space, execute a common endpoint across every specification, then inspect sign, magnitude, specification curves, and factor sensitivity without searching for a favourable pipeline.
Reproducible publication
Generate deterministic evidence artifacts and fingerprints, preserve declared decisions, and report what the audit supports without turning descriptive robustness summaries into causal claims.
Scientific boundary
Executable visual methods
The public plotting API produces deterministic SVGs for QC, readiness, robustness, sensitivity, and measurement uncertainty.
All figures in this section use deterministic synthetic demonstration data and are regenerated from executable Python in CI.
Frozen validation programme
Robustness is allowed to fail, survive, or remain incomplete. These labels are protocol-bound scientific records—not generic scores for the source datasets.
Five frozen detector specifications reached 322/322 finite participant estimates; two reached 0/322, so the predeclared completeness gate was not satisfied.
Read the frozen case →
The uncertainty-propagated effect moved toward zero, but all 2,000 prespecified measurement-model draws remained below zero.
Read the frozen case →
The endpoint retained its negative direction while magnitude recovery failed at lower sampling rates under the frozen perturbation protocol.
Read the frozen case →One auditable workflow
The workflow keeps measurement assumptions, readiness decisions, specification choices, and reporting artifacts visible as separate scientific decisions.
Start from a defensible question
Run the smallest example, inspect the public API, or reproduce the complete visual gallery from source.