Scientific figure atlas¶

Filter and search selected figures — a source-linked curated view with representation, inference and missingness filters. The full atlas remains below.

Figures are grouped by the question they answer, rather than by the order in which they were created. Example plots are deterministic synthetic illustrations. Research plots document specifically frozen experiments and should not be generalized beyond their archived assumptions.

A. Understand the functional gaze representation¶

Time-indexed planar trajectories — what we preserve instead of collapsing immediately to counts.

Representation guide

Sparse x/y covariance structure — illustrates modeled covariance, not tracking hardware performance.

Sparse FPCA/MFPCA methods

B. Functional inference under stress¶

R78 time-varying AR(1) functional inference diagnostic

R80 independent-unit Student-t5 functional inference diagnostic

R78 temporal covariance · R80 heavy-tailed units · Independent figure-ledger archive

C. Missingness, audit and exact identified sets¶

Original R82 selective missing-gaze retention and support refusal

R93 exact binary latent mean sharp identified sets

R94 sharp completion-floor assumptions

R95 known direction and equality response assumptions

Missing gaze study · Identification methods · All W3 studies

D. Compare conditional missing-outcome assumptions¶

R93–R95 exact normalized sharp-set widths across conditional return assumptions

Why different assumptions change the sharp set. These ratios are analytical population identities, not fitted uncertainty or measured gains.

Reproducibility¶

Read the source-specific evidence ledgers before quoting numeric proportions: R93, R94, R95. See also the full synthetic workflow visual gallery.

Do not treat simulation truth-inclusion as validated coverage

Whole-slope inclusion under a specified generator is not proof of general nominal frequentist calibration. A population identified set is not a confidence band.