Governed end-to-end analysis pipelines

The pipeline layer links import, measurement quality, preprocessing, feature construction, modeling, diagnostics, sensitivity analysis, and reporting while preserving the researcher’s declared choices. Pipeline steps are explicit functions with declared dependencies; eyeprocess does not silently choose preprocessing or statistical specifications.

spec <- eye_analysis_spec(blink_correction="linear", pupil_baseline=c(-500,0), fixation_algorithm="ivt", aoi_rule="probabilistic")
p <- eye_analysis_pipeline(list(
  eye_pipeline_step("import", read_fun),
  eye_pipeline_step("quality", quality_fun, requires="import"),
  eye_pipeline_step("model", model_fun, requires="quality")
), spec = spec)
validate_eye_pipeline(p)
r <- run_eye_pipeline(p, context=list(path="study.csv"))
audit_eye_pipeline(r)
plot(p)

eye_targets_manifest() and write_eye_targets_template() provide interoperability scaffolding without pretending arbitrary closures can be losslessly translated into another pipeline engine.