Articles
996102e0
Articles
These articles explain the methodological ideas behind GazeAudit. They are not substitutes for the API guides, researcher-owned study decisions, manuscript-readiness checks, reviewer-amendment provenance, or the frozen validation records. If the need is procedural, factual, or learning-oriented rather than explanatory, use the Documentation compass.
Choose by question
Why treat gaze as uncertain?
Understand why measured gaze coordinates are not automatically known true points and why AOI boundaries can amplify spatial error.
Why audit more than one pipeline?
See how defensible analytical alternatives become a declared, inspectable specification space rather than a sequence of post-hoc choices.
What does fragility mean?
Learn why sensitivity is evidence about the scope of a conclusion rather than a software defect or a reason to search for a preferred branch.
I already have outputs. What next?
Move from the complete execution record to bounded interpretation by checking denominator, endpoint, direction, magnitude, sensitivity, and untested uncertainty.
Can a reviewer reconstruct the claim?
Check whether the manuscript, decision log, execution denominator, failures, limitations, archive manifest, and software identity tell one recoverable story.
The reviewer asked for another analysis. Now what?
Extend the record transparently: keep the submitted audit recoverable, record outcome-inspection timing, and give post-review evidence its own denominator and provenance.
How should I design a sensitivity analysis?
Separate a scientifically justified perturbation protocol from a post-outcome search for settings that preserve the preferred result.
Measurement uncertainty is a modelling problem
Why a gaze coordinate should not automatically be treated as a known true point, why AOI boundaries magnify small spatial errors, and why Monte Carlo propagation is an assumption-aware representation rather than a claim of certainty.
Continue with: the AOI uncertainty guide and AOI boundary example.
From one pipeline to a robustness audit
Why a single reasonable pipeline does not answer whether a conclusion depends on other reasonable pipelines, and how specification spaces turn analytical choice into an auditable object.
Continue with: the specification-space guide and end-to-end robustness example.
How to read a fragile result
Why fragility is not a software error, how to localise its source, and how to narrow scientific claims without turning robustness analysis into post-hoc optimisation.
Continue with: Interpret an audit result and the four synthetic result-pattern reporting examples.
Sensitivity analysis is not a search for a stable answer
Why the perturbation protocol—not the most reassuring point on a curve—is the scientific object, and why sampling, missingness, spatial measurement error, and ordinary specification choices should retain distinct meanings.
Continue with: the sensitivity-analysis design guide, sensitivity protocol example, and sensitivity-audit workflow.
From concept to action
If you are no longer asking why robustness or uncertainty matters and instead need to decide what a completed result licenses you to say, use this practical sequence:
- Interpret an audit result — verify completeness, endpoint consistency, direction, magnitude, sensitivity, and unresolved uncertainty.
- Result-pattern reporting example — compare complete/stable, magnitude-sensitive, sign-sensitive, and incomplete synthetic patterns.
- Reporting robustness — turn the bounded interpretation into precise Methods and Results language.
- Manuscript readiness — test whether an independent reader can reconstruct endpoint, denominator, failures, decisions, wording, limitations, and software identity.
- Reviewer reconstruction — practise finding an incomplete valid denominator and repairing an over-strong claim from outside the analysis team.
- Reviewer-requested amendments — preserve the submitted record when revision adds exclusions, thresholds, endpoints, measurement assumptions, or sensitivity analyses.
- Reviewer-requested reanalysis — practise keeping a synthetic submitted 8/8 audit separate from a post-review 4/4 amendment.
- Publication/archive handoff — preserve the decision trail, branch accountability, provenance, limitations, revision evidence, and manuscript record.
Relationship to the validation records
The articles discuss general methodology. The canonical GazeBase, Korthals, and Pedrotti/de Chambrier outcomes remain defined by their frozen protocols and evidence. Use the validation matrix for those claims. Synthetic teaching patterns and general methodological articles do not create or modify case-study classifications. Reviewer reconstruction and reviewer-requested reanalysis exercises are also synthetic teaching material and do not create or modify case-study classifications.