Correction to TFP submission sub_536xxz341y, prepared 2026-10-02

1. Changed 1,978 explicit internal-standard absence flags to unspecified. The paper does not report internal-standard use; source silence does not prove absence. AREA and CUSTOM SELECTIVITY values are unchanged.
2. Qualified reconstructed algorithm/iteration labels and physical-plate claims. Coverage of published suggestions/results does not establish how many physical reruns occurred or whether unpublished measurements were excluded.
3. Added runnable supporting checks, reconstruction inputs, dependency versions and evidence to the sources archive. The original dataset is included solely as a baseline for a reproducible comparison of the correction.
4. Corrected the README publication URL description; all reactions deliberately retain the arXiv DOI and URL.
5. Qualified drawing-based phosphine-form assignments and identified the separate-upscale source ambiguity. No chemical structure was replaced.
6. Added explicit agent/tool attribution and separated historical evidence from checks freshly executed for this correction.
7. Preserved original record_created values and added dated record_modified events using the same confirmed public curator identity.

Source experimental values, source compounds and reaction coverage are preserved. The exact semantic comparison is in analysis/change_comparison.json. The dataset is proposed for review; no ORD acceptance is claimed.

The included 0–199 seed scan supplies executable evidence: seed 9 matches 96/96 measured sulfonamide conditions and the next-best overlap is 8/96. The report distinguishes the mean across all 200 seeds (3.64) from the mean excluding seed 9 (about 3.18). The old 3.6 figure is consistent with the former and is not treated as a demonstrated calculation error. The inferred 96-row membership is unchanged.
