---

output: github_document

eyesim

Installation

You can install development version from GitHub with:

#install.packages("remotes")
remotes::install_github("bbuchsbaum/eyesim")

Overview

Are two eye-movement fixation patterns similar? The goal of the eyesim package is to provide methods for answering this question.

For example, in studies of recognition memory, we might ask whether the eye-movement fixation pattern during the first presentation of an image ("encoding") is similar to the eye-movement fixation pattern during the second presentation of the same image ("recognition"). Furthermore, we can ask whether fixation similarity varies as a function of conditions in the experimental design or the behavioral responses (e.g. correct recognition or miss).

The experimental GazeWeave workflow asks whether a held-out gaze path identifies its correct template among permitted nonmatches. Directional Replay is designed for encoding-to-recall studies. Edge-normalized Transport is the symmetric explanatory alignment engine for local order and partial coverage; its dedicated gaze_transport_cv() interface accepts one or several separate study presentations. Every engine returns one primary endpoint, gaze_info_bits = log2(p_true / prior_true), using candidate calibration and warp estimation learned only from outer-training trials. Posterior Replay coverage and Transport's optimized correspondence, coverage, chronology, spatial residuals, and contraction remain explanatory diagnostics rather than additional outcomes.

Transport passed its frozen invariant, simulation, runtime, and full-cohort repeated-viewing gates. In the sealed 0--3000 ms court it had the largest response-blind retrieval point estimate, but the interval included zero and it did not improve held-out old-item response prediction. The package therefore makes no retrieval-prediction or general-superiority claim; Replay, registered density, and MultiMatch remain visible comparators. Earlier Transport v1/v2/v3 labels were internal development names: the edge-normalized estimator is now the sole public gaze_transport_* method. Version labels survive only in internal and frozen validation provenance; they are not selectable methods. See the GazeWeave vignette for the executable Transport example, episode normalization, plot semantics, validation evidence, and failure modes.

Vignettes

See examples of use of eyesim in the vignettes.

With some luck you might be able to make a cool image like the one below!