eyetrajectoriespy 1.2.0rc2 — published prerelease (24 October 2026)¶

Intended purpose¶

This candidate is a compatibility-preserving, prerelease-stage update for the existing ten typed, explicit workflow orchestration routes under eyetrajectoriespy.workflows. It is not a release of new inferential confidence guarantees or a reclassification of research-only W3 methods.

Candidate functionality¶

• Ten versioned W1–W4 routes for FPCA, sparse PACE, paired/asynchronous planar sparse multivariate analysis, sparse multilevel FPCA, partial prediction, function-on-scalar, Gaussian mixed effects, generalized functional regression, and explicit recurrence/RQA.
• Workflow configuration, ordered decisions and pre-processing boundaries remain visible; execution does not silently interpolate, infer families, select analyses or delete observations.
• Portable bundles, reports, reproducible figure dispatch, and per-route provenance are retained.
• A separate opt-in experimental eyetrajectoriespy.workflows.experimental module containing twelve E1–E4 helpers (preprocessing, preflight, declared sensitivity and reporting) is importable from the candidate wheel. These helpers are not among the ten qualified workflow entry points and were not included in stable 1.1.0. Exact-source E5 engineering checks passed on Python 3.11–3.13; independent E5 content approval remains pending. No promise of API stability, nominal inferential coverage or missingness correction accompanies them.

Compatibility¶

The historical root API is frozen at 458 names, of which 455 are stable and three explicitly experimental. Existing 1.1.0 stable root exports and supported 1.1 sparse/irregular module APIs must remain compatible. The workflow candidate APIs remain module-scoped and are not promoted into eyetrajectoriespy root.

Supported Python versions: 3.11–3.13. Optional FDApy scientific-backend compatibility retains its narrower supported Python constraint.

Scientific and inferential boundaries¶

The following are not supported release promises:

• nominal 90% simultaneous function-on-scalar regression bands or universally calibrated fitted-covariance/CR2 inference under extreme predictor leverage (#289);
• identification of missing-not-at-random gaze loss from observed-only data, automatic reconstruction of unreturned gaze, or response-dependent audit completion without externally justified assumptions (#296);
• converting original R1–R95 simulation studies and independent negative findings into production statistical estimators.

Synthetic research findings, including calibration failures, remain available as labelled research evidence rather than being silently omitted or interpreted as positive coverage guarantees.

Verified production publication¶

The literal 1.2.0rc2 prerelease was published on 24 October 2026 from protected main a45d159f0ba6246736752575b9ede1520140016e. The release workflow #25 passed its build, governance, version uniqueness, GitHub Release, PyPI Trusted Publishing, checksum and fresh install stages. Wheel SHA256: 0641db2e634221d9e077b9f4c0837623eb97dc2dd245583ec9af5ba16396a134. Sdist SHA256: ea810de67bda6019b72db7e05145fdf36d700c5d23bbe4d6cb07d3f51c74adf8.

This RC2 is a prerelease, not stable 1.2.0. Stable 1.1.0 remains the default pending separately qualified stable 1.2.0 publication.