# Source audit pass 3 — 23 October 2026

> Historical pass snapshot. Current status and methods: README.md; current measurement audit: measurement_audit_pass4.csv. Counts and pending items below describe this pass at the time, not the latest state.

Current measurement table: measurement_audit_pass3.csv. Seven additional records received source adjudication; none received a numerical CFF replacement. These are partial evidence decisions, not seven fully approved observations. Raw measurements, prior corrections and supplied trees remain preserved.

Read source_decisions_pass3.csv for row-specific evidence, URLs and holds. The current screening queue contains 117 measurements from 91 tree labels. It is a review workload, not a final sample; excluded endpoints remain in the full table. Prior crosswalk_pass2.csv remains the current identity crosswalk; this pass changes measurement classification only.

## Method decision before pooling

Define the estimand explicitly: a behavioural threshold under recorded test conditions, versus maximum temporal capacity over luminance. The working default is measured behavioural uniform-flicker thresholds with conditions retained. Analyse retinal ERG separately. Motion responses, reported maxima, means, medians and range midpoints must not be pooled as interchangeable estimates. No geometric averaging across incompatible statistics or protocols is authorised by this audit.

A threshold measured at a particular luminance remains a measurement even if the species peak was not reached. This is different from a tested frequency ceiling that supplies only a lower bound. The new maximum_capacity_status field records that distinction. tested_lower_Hz and tested_upper_Hz are descriptive source ranges/brackets, not confidence limits or automatically valid individual censoring intervals.

## Remaining work

Extract outstanding figures/tables and original reports, record light, temperature, age and threshold criteria, reconcile taxon/population and longevity quality, and document exclusions. Missing researcher replies remain open evidence requests; no response is assumed. Freeze the sample only after these gates are reviewed. No final models, preregistration or publication were performed.

## Reproduction

Run apply_audit.py DATA OUT, then continue_audit.py DATA OUT, then continue_audit_pass3.py DATA OUT using Python 3. The input directory must contain the supplied datasets in their existing layout. Scripts use the standard library. All three passes were rerun in a fresh temporary directory and generated CSV/JSON outputs were compared byte-for-byte. ZIP integrity was checked.