Corrections log
Every change made to a published figure since the index opened, with both values and the reason. 10 entries. A short log is not evidence of accuracy — it is evidence of how much has been checked — and the index would rather this page were longer than that it were empty.
| Date | Record or table | Field | Was | Now | Why |
|---|---|---|---|---|---|
| 2026-07-12 | AX-2026-042011 | lot | AMB-2603-S5676 | AMB-2603-S5667 | Transcription error at intake: two digits transposed against the submitted certificate, so the record was attached to a lot code that had never been issued. Re-attached to AMB-2603-S5667 and the batch rollup recomputed. |
| 2026-06-03 | batch rollups (820 batches) | coa_divergence_pp | computed against the mean | computed against the median | Divergence was being computed against the mean of independent results on lots with three or more assays, contrary to what /suppliers/methodology/ stated. Every batch with three or more independent tests changed, by up to 0.08 pp. |
| 2026-04-21 | AX-2025-021217 | endotoxin_eu_mg | 3.8 | 0.38 | Decimal point misplaced transcribing a gel-clot result: 3.8 EU/mg was entered where the report read 0.38. The record had been flagged AX-R08 in error and the flag is withdrawn. |
| 2026-02-08 | compound registry — 3 compounds | mono_mass_da | average mass | monoisotopic mass | Three compounds carried an average mass in the monoisotopic field, producing apparent ppm errors of several hundred on every record referencing them. Twenty-two records had their AX-R01 flag withdrawn. |
| 2025-11-14 | AX-2025-019700 | sample_source | buyer-sourced | supplier-supplied | Submitter clarified that the vial had been provided by the supplier rather than purchased through the retail channel. Reclassified, and the supplier scorecard recomputed with the record moved out of the buyer-sourced subset. |
| 2025-09-02 | all quarterly aggregates (402 records affected) | median_purity_pct | included method-limited records | excludes method-limited records | Records marked method-limited were being included in class aggregates, inflating incretin medians by an estimated 0.04 pp. The exclusion is now applied at aggregation rather than at intake, so the records themselves are unchanged. |
| 2025-05-27 | AX-2024-004960 | compound_slug | melanotan-ii | pt-141 | A unit-resolution identity confirmation had been recorded against the wrong member of a 1 Da pair. Re-assay at high resolution established the compound. The record carries a note explaining why the two could not have been distinguished on the original determination. |
| 2025-03-19 | suppliers.csv | established | blank for 4 suppliers | populated | Four suppliers publish an establishment year that had not been transcribed into the registry. Populated from their own published material and checked against the canon source. |
| 2024-12-11 | AX-2024-004582 and 6 others | water_pct | as reported | as reported, qualified | Seven records from that year reported water at or below 1.28 %, the driest figures in the index and close to what coulometric Karl Fischer resolves reliably at these sample sizes. The values are retained unchanged and each now carries a note that it should be read as an upper bound rather than as a point measurement. |
| 2024-08-05 | peak tables — 31 records | largest_impurity_pct | unordered | ordered descending | Thirty-one peak tables listed impurities by retention time rather than by descending area, so the first row was not the largest single impurity the column header claimed it was. Re-ordered, and the ordering is now asserted by the build checker. |
How to report an error
Errors are reported the same way records are submitted — see the submission page. A useful report names the record ID, the field, what the value should be and how you know. The index will correct a demonstrated error in a published figure within a day and log it here.
What is not a correction
A disagreement between two laboratories is not an error and does not get corrected. It gets published on the batch page as a disagreement, because that is what it is. See the data policy on disputes.