Corrections log

Changes to the method itself: how studies are selected, graded or read. Each entry has a date and the reason for the change. This is the log referenced from About & Methodology's correction policy.

Date	Change	Why
2026-10-18	One-off retrospective pass adding foundational papers	The corpus had grown mostly from the daily screen of new papers, so older work that established the pathway was thin wherever a branch had not been revisited. A deliberate pass branch by branch added the foundational papers that were missing: between the snapshots of 15 and 20 October 2026 the corpus went from 449 to 518 records, and 66 of the 69 new records were papers published between 1997 and 2018. They were selected by the same inclusion rules (IN1 to IN5) as every other record and graded the same way, and were attached as supporting studies to existing pathway links; no new link was created from them. This was a single catch-up, not a change in the normal update rate.
2026-09-23	Open Questions cards rewritten around what has changed since the question was written	A check of the ten cards against published human results found sentences and two titles saying that something had never been tested in humans when trials existed (for example, the combination of rapamycin with other drugs was tested by the mouse Interventions Testing Program in 2016 and 2022). Each card now states what level of evidence the question stands on, what has changed in the literature since it was written, and what is still open. The changed-since block is the only place a card may cite studies outside the corpus, and every number in it must carry a source a reader can look up (PMID, trial number, preprint or corpus code). Later edits to a card go into a separate collapsed revision history instead of the scientific text itself. Five Frontier Questions about how the pathway is measured got their own section, kept apart from the gaps computed against the corpus.
2026-09-14	Inclusion and exclusion criteria rewritten to match what the corpus actually does	The published rules asked for a systematic review or a large randomised trial. The corpus never followed that, and should not: its human evidence includes small trials of 8 to about 25 people, because nothing stronger exists for those questions. The rule is now the strongest evidence available for the claim, with sample size and limits shown in the record. A PMID is no longer required. A DOI or an NCT number is enough, because 29 of 380 records (7.6%) were papers still waiting for PubMed indexing and one is a registered trial. A fifth inclusion rule covers narrative reviews, about 10% of the corpus; a review never stands alone behind a causal link. The rules are numbered IN1 to IN5 and EX1 to EX4 so that any exclusion can be cited.
2026-09-13	Public self-audit of the evidence, and a record of how long and under what dosing a finding was measured	Two pages now audit the corpus against itself. /evidence/audit/ shows what the evidence can carry: at the time of publication 10.3% of studies were human, 72% of pathway links rested only on mechanistic work, 40% rested on a single study, and none of the links had been signed off as Confirmed by a reviewer. /pathway/timing/ records whether a finding was measured under chronic, intermittent, acute or withdrawal dosing. Not stated means nobody checked; not applicable means the question does not arise for that study, and the two are never merged.
2026-09-07	Evidence codes renamed from A–D to S / H / A / M / R, and reviews split out	The old codes ran A, B, C, D. However carefully the caption said “a tier is not a grade”, a lettered ladder is read as a school report — and 61% of the corpus sat at the bottom letter, including the experiments that established the pathway. The codes now name the system a finding was established in and carry no order: S = synthesis of human data, H = human study, A = animal model, M = molecular. Narrative reviews, previously grouped with primary molecular work under the same letter, now carry their own R and are drawn outlined, like preprints and registered trials, because a review is a different form of document rather than a weaker one. Nothing in the underlying data changed and no study was re-graded: the stored values are untouched, only what a reader sees. The change was prompted by a public comment from Prof. Brendan Manning (Harvard T.H. Chan School of Public Health), who pointed out that rigorous mechanistic work should not be weighted down merely for not yet having been shown in humans.
2026-08-03	Wording calibrated to what each statement is: fact, interpretation or hypothesis	Statements are labelled as fact, interpretation or hypothesis. Absolute wording such as none of them or the first was replaced with hedged wording that names the limit. The evidence pyramid carries a note that a lower rung is not worse science. The section once called Gaps and Hypotheses became Open Questions. The selection policy now says that no screening log of rejected papers exists, so no number of rejected papers is quoted.
2026-07-29	External scientific review: citations must carry the claim, and confidence split into three judgements	An external review checked all 93 pathway links at the time, sign by sign, and found no wrong sign. It found 16 problems elsewhere, mostly citations that did not support the claim made: a paper that measured only side effects was cited for a lifespan benefit and was replaced. Since then a citation has to support the specific claim it is attached to. Confidence is no longer one number but three separate judgements (strength of the mechanism, relevance to humans, degree of consensus). Relevance to humans is derived from the cited studies, and a curator can only lower it; on the first run it dropped for 29 of 100 links.

About & Methodology