brugada.net · Corrections
Updated 16 July 2026
Printed at the same size as the claim. A newspaper that never runs a correction isn’t careful — it just isn’t checking.
Forty-seven fixes are on file with their before-and-after. These are the ones that changed what the project believed, plus the one that is still open.
Agmatine was the hero of this project. In the cheap test it gripped at 2.92 Å with 89% occupancy. Tested properly in a full membrane across three seeds it escapes: 6.14 ± 0.61 Å, 9% occupancy. Its free-energy profile finished the job — the minimum sits at 8.6 Å, a separated state, only ~3 kT deep, while the clip it was nominated for is +7.55 kcal/mol up the wall.
Status — Retracted as a lead; kept as a benchmark. This is the correction the project is proudest of, because the method caught it rather than a reviewer.
The functional data behind “~0.29× wild-type current” was attributed to Holst across 16 documents, 26 times. The correct source is Gütter, Benndorf & Zimmer 2013 (PMID 23805106). The project’s own literature notes had it right on 3 July and the error propagated anyway — for nine days, unnoticed.
Status — Fixed in the living documents. Two dated audit records keep the old label deliberately, as evidence of what they audited. One figure has “Holst” baked into its pixels and is flagged for regeneration — it is not on this site.
PMF_RESULT.json reported “favorable clip well” for agmatine. The −1.94 kcal/mol minimum it referred to is at 8.6 Å — a separated state, not the clip. The label came from position-blind logic: any well deeper than −1.0 kcal/mol got called a clip well regardless of where it sat. The figure was always right; only the metadata and the code were wrong.
Status — Relabelled, numbers untouched; the analysis code is now position-aware. Left alone, this would have been an overclaim about the project’s own lead, sitting inside its own machine-readable results.
The approved-drug screen was written up as complete. It had written 3,549 of 10,667 result files — about 33% — with 788 timeouts and 15 crashes, and the wrapper was still running when consolidation stopped it.
Status — Corrected everywhere it appeared. It still produced real hits, including two of the four compounds that reached the long test.
A metadata file of PubMed records — data fetched from the internet — carried an extra field that wasn’t bibliographic data at all. It was a block of imperative text addressed to whatever automated tool read the file, telling it how to behave, including text designed to discourage its own removal. Eight legitimate article records sat around it, untouched and real.
Status — Removed; the real records preserved byte-for-byte. It’s recorded here because a project that pipes fetched data into automated tools should say out loud that this happens — and that file contents are data, never instructions.
The scoreboard designated “verdict of record” reports four surviving compounds. It cannot be regenerated — nothing in the project writes it. It states a majority rule that no code implements, and the classifier that does exist would call two of the four survivors failures. Its seed-1 numbers match the simulations exactly; none of its other seeds match anything. Tegaserod, listed as a survivor, lets go in every simulation that still exists — including the longest run in the project. Alniditan, listed as the shaky one, holds in all three.
Status — OPEN. Not resolved, and not resolvable from the archive: an automated cleanup deleted the relevant runs before they were harvested, so the numbers may be real measurements of trajectories that no longer exist. One reduction on one disk would settle it. Until then this stays open and the survivor count stays qualified. The full reconciliation →
Two things this site deliberately does not do, which belong on a corrections page as much as any error:
It does not publish the patient. This project holds 31 patient-identifying clinical records. None has ever been opened. A family pedigree and a device-telemetry chart exist in the archive and are not here and never will be — a pedigree isn’t a disclosure about one person, it’s a disclosure about six relatives who were never asked. The science stands without them.
It does not claim a discovery. There is no drug on this site. There are catalogue entries and repurposed drugs that survived a filter which was itself incomplete, and one experiment that would tell you whether any of it means anything.
onwards.