Overview
Archive note: The extracted image thumbnails are retained from itrace-0.4.1-manuscript.pdf and are not figures from the corrected 2026-10-16 manuscript. Use itrace-manuscript-2026-10-16.pdf or its pinned source archive for current figures.
# iTrace manuscript revision, 2026-10-16 This deposit archives the corrected 30-page manuscript from reviewed commit `310be17ace8ae61621afe01329356427a3e6c861`. Exact source and figure artifacts are at https://github.com/docxology/itrace/tree/310be17ace8ae61621afe01329356427a3e6c861. It is a manuscript revision after the immutable v0.5.0 release, not a new package release. The package version in that source remains 0.5.0. The previous release remains at https://doi.org/10.5281/zenodo.23403300. References in the manuscript to that DOI identify the archived baseline. Changes correct overlap scoring, resampling and fit uncertainty, fitted-null diagnostics, uncertainty figures, scholarship, and scientific interpretation. The manuscript distinguishes synthetic verification from real-device validation. No new participant or reference-device data are claimed. Verification: 737 tests passed, 2 skipped, 91.05% combined statement-and-branch coverage; 44 final artifact tests passed, independent review approved the exact source commit, and all ten hosted CI jobs passed. The included figure manifest inventories 41 artifacts in the linked source tree. `provenance.json` binds this PDF to the Git commit; `SHA256SUMS` inventories every payload.
Overview source: Curated paper metadata.
Methods and contributions
Read the source for the full argument, qualifications, and evidence.
Findings and contributions
- In the specified synthetic landmark-noise sweep, gaze remains below 2 deg RMS to sigma approximately 0.0048, about 3 pixels at 640-pixel width.
- One-to-one overlap-based saccade recovery falls below F1 = 0.8 near sigma 0.0014, about 0.92 pixels, in that synthetic sweep.
- The overlap result measures event recovery and does not establish timing accuracy.
- Pupil noise-sweep robustness is conditional on the assumed pupil-noise injection model.
- These results verify specified algorithms and internal geometry; they do not validate webcams as eye trackers.
Methods
- Hardware-decoupled analysis core — The optional capture shell emits typed records consumed by a headless scientific-analysis core.
- Constructed-ground-truth verification — Headless tests exercise the algorithms on synthetic signals whose truth is supplied by construction.
- Explicit caller-supplied benchmark truth — Benchmark outputs report their truth boundary and distinguish comparator agreement from reference measurement.
Summary sources: Paper metadata and evidence · Extracted source text.
Document downloads
PDFs are hosted here; supporting sources may use pinned archives.
- itrace-0.4.1-manuscript.pdf PDF · 5.08 MiB
- itrace-manuscript-2026-10-16.pdf PDF · 5.47 MiB · declared primary
These files may be different versions or companion documents. The archive does not identify a latest edition.
Citation
Citation metadata follows the unified bibliography.
Catalog details and resources
Related works
Ranked by recorded methods, topics, contributors and associated software, with contributor roles and domain-only matches labelled. Metadata overlap is not a citation or a scientific equivalence.