# Pass 4 — 24 October 2026

Current table: measurement_audit_pass4.csv. Four records received additional source checks; none is newly certified analysis-ready. Existing corrections and platypus gap entry remain intact. No CFF numeric replacements were made.

## Findings

- H2025_064, pigeon, Delius et al. 2017: Table 1 reports 73 Hz (70–74) for green light, four birds. Compiled 70 matches the range low rather than the mean. Source mean is stored separately in source_supported_summary_Hz_pass4; retain the compilation value until the statistic-selection rule is applied consistently. Conditions and sample size now recorded. https://pmc.ncbi.nlm.nih.gov/articles/PMC5679541/
- H2025_005, goldfish, Behrend et al. 2007: abstract supports approximately 39 Hz, not exact 39.5. Behavioural and ERG experiments and drug manipulations require arm-specific extraction. https://doi.org/10.1017/S0952523807070691
- H2025_006, goldfish, Mora-Ferrer and Gangluff 2002: pre-injection range 33–39 Hz verified; exact 37.5 and summary type remain unverified. https://doi.org/10.1017/S0952523802196106
- H2025_008, goldfish, Bilotta et al. 1998: behavioural sinusoidal-flicker contrast-sensitivity procedure verified; 24 Hz and corresponding luminance/contrast still need full-text extraction. https://pubmed.ncbi.nlm.nih.gov/9474375/

Sources accessed 24 October 2026. Pigeon full text/table was available through PMC after the PLOS image endpoint failed. Goldfish evidence here is abstract-level; it is not a full-text audit.

## Working method clarification

For a baseline comparative analysis, use demonstrably untreated measurements. Record drug, lesion and recovery arms separately. A paper containing treatment arms is not automatically excluded; its eligible baseline must be identified. Do not collapse sinusoidal contrast thresholds, square-wave flicker, wavelength-specific means, or training variants into a single species maximum. Missing conditions remain missing, not imputed from another experiment. Reported ranges are not confidence intervals.

## Status and remaining work

280 measurement rows remain. The screening queue remains 117 measurements across 91 tree labels; these are workload counts, not a final sample. The earlier longevity-quality candidate count is not re-audited in this pass. Human review, tree/population reconciliation and final sample freeze remain pending. No final model, preregistration or public-release changes.

Next priorities: obtain exact goldfish baseline values and conditions; decide the pigeon statistic under a consistent estimand; continue remaining behavioural sources. The raw AnAge data and trees are not in this ZIP, so this pass does not claim to have rerun their checks.

## Reproduction and validation

Run `python continue_audit_pass4.py` from the package, or pass its directory as the argument. It reads the bundled pass-3 table; no raw external files are required for this incremental pass. Validation checks unique IDs, unchanged CFF/identity/longevity fields, preservation of untouched rows, deterministic regeneration, and ZIP integrity. Earlier passes still require their original external inputs.