Workflows

Research workflows

Use these pages when you already understand the basic functions and need to structure a complete analysis or revision record.

First-study audit

Question: how do I take one real project from canonical data through preflight, declared robustness, evidence review, and publication provenance?

Typical sequence:

1. define the scientific endpoint and study-specific assumptions;
2. map one canonical GazeStudy;
3. run structural preflight and record decisions;
4. add readiness or measurement layers only when scientifically relevant;
5. declare and execute the defensible specification space;
6. inspect robustness and targeted sensitivity diagnostics;
7. preserve the complete evidence and software identity.

Measurement audit

Question: could plausible gaze-position error change AOI membership or the scientific endpoint?

Typical sequence:

1. preserve validation information;
2. fit global or grouped error models;
3. propagate uncertainty into AOI membership;
4. compare hard and probabilistic assignment;
5. calculate uncertainty-weighted endpoints;
6. stress-test the error model.

Robustness audit

Question: does the scientific endpoint remain stable across defensible analytical decisions?

Typical sequence:

1. define the common endpoint;
2. declare a PipelineSpace;
3. remove scientifically invalid combinations before execution;
4. execute all remaining specifications;
5. inspect specification curve and sign/magnitude stability;
6. screen marginal and pairwise sensitivity;
7. interpret instability without post-hoc optimisation.

Sensitivity audit

Question: how does one fixed endpoint behave when a specific ordered uncertainty dimension is deliberately perturbed?

Typical sequence:

1. freeze the endpoint and baseline representation;
2. name one perturbation dimension such as sampling rate, controlled missingness, or spatial-error scale;
3. justify the evaluated range independently of the observed endpoint;
4. declare units, model/mechanism, randomness, and software identity;
5. execute and preserve the complete curve;
6. interpret direction and magnitude separately;
7. integrate the result with the main robustness record without collapsing different uncertainty types into one score.

Peer-review revision toolkit

Question: how do I add reviewer-requested work without rewriting what existed at submission?

Typical sequence:

1. classify the reviewer request before changing the record;
2. preserve the submitted evidence and its original denominator;
3. record outcome-inspection timing for post-review analytical decisions;
4. execute reviewer-requested amendments under their own denominators;
5. bind each response item to manuscript and archive locations;
6. record material v1 → v2 changes in a version-change manifest;
7. scaffold and validate the structural revision package;
8. run the final resubmission consistency gate.

The toolkit routes to the amendment guide, response-letter guide, executable gazeaudit-revision-package workflow, worked failure/repair example, and final editor-handoff checks.

Reproducible publication

Question: can another researcher reconstruct what was declared, executed, revised, and reported?

Typical sequence:

1. freeze software and source identity;
2. freeze the scientific decision rule where appropriate;
3. execute the analysis;
4. build a deterministic audit bundle;
5. preserve scientific and execution fingerprints;
6. preserve review-stage amendments and their temporal provenance;
7. archive report, methods, manifests, complete results, and final evidence maps;
8. cite the exact release or commit.

Combine workflows deliberately

The first-study workflow is the project-level spine. Measurement, robustness, sensitivity, revision, and publication workflows are governed layers that can be nested where the study requires them.

For example, a PipelineSpace may contain hard versus probabilistic AOI branches, while each probabilistic branch is generated under a declared error model. If peer review later requests a stricter threshold, that analysis belongs in a separately timed revision layer rather than being silently inserted into the submitted specification space. The publication workflow can then bind the submitted and post-review records without collapsing them.

The important constraint is semantic and temporal clarity: measurement assumptions, analytical choices, perturbation analyses, reviewer-requested extensions, corrections, and publication rules should remain distinguishable in the provenance record.