Corrections

A public record of every factual correction: what the record said, why it was wrong, how the error was found, and the commit that fixed it. Errors are expected in a corpus this size; hiding them would be the failure.

Every factual correction to the OnCo corpus, newest first. Format: date · entity · what was wrong · how it was found · fix.

2026-10-10

Date	Entity	What was wrong	How found	Fix
2026-10-10	UCSF and 58 breast-study records	58 of the 120 studies in UCSF's female/adult breast list had no OnCo trial record; the institution's direct trial list did not contain the complete set.	Compared the filtered UCSF card inventory with the assembled graph by NCT, searched aliases and links, and resolved all 120 exact identifiers in ClinicalTrials.gov.	Added 58 sourced records and linked all 120 canonical records to UCSF. Preserved local versus overall availability, observational versus interventional design, and expanded access. Complete dated audit; triage issue #247.
2026-10-10	NCT03990896 and COGNITION-GUIDE	Generic PARP summaries did not expose metastatic/assay gates or Germany-only molecular assignment.	Current ClinicalTrials.gov records and UCSF listing; issue 245.	State current circulating-DNA gate, prior-treatment constraints and site uncertainty; distinguish the postoperative umbrella's separate arms and somatic-BRCA assignment.
2026-10-10	PARP breast trial and paper graph	Closed postoperative studies and the olaparib/axatilimab study were absent; smaller somatic-basket and negative postoperative evidence lacked paper records.	Registry identifiers and resolved primary publications, linked in issue 245.	Add seven sourced trials, four papers and the missing axatilimab drug record; retain current status, cohort/endpoint limits, pulmonary exclusion and the difference between somatic metastatic activity and adjuvant benefit. Link the records from existing PARP drug owners and classify axatilimab as supportive care for its current GVHD approval.

2026-10-07

Date	Entity	What was wrong	How found	Fix
2026-10-07	ascot-jcog1202	The UMIN receipt-number URL resolved to a different registry identifier from the source label.	Trial participation inventory; both old and replacement pages checked against their displayed UMIN identifiers.	Replaced R000013683 with R000013644; retained the intended study identifier and trial findings.
2026-10-07	create-x	The UMIN receipt-number URL resolved to a different registry identifier from the source label.	Trial participation inventory; both old and replacement pages checked against their displayed UMIN identifiers.	Replaced R000001004 with R000000989; retained the intended study identifier and trial findings.
2026-10-07	jaspac-01	The UMIN receipt-number URL resolved to a different registry identifier from the source label.	Trial participation inventory; both old and replacement pages checked against their displayed UMIN identifiers.	Replaced R000000794 with R000000791; retained the intended study identifier and trial findings.
2026-10-07	jcog0802	The UMIN receipt-number URL resolved to a different registry identifier from the source label.	Trial participation inventory; both old and replacement pages checked against their displayed UMIN identifiers.	Replaced R000002809 with R000002300; retained the intended study identifier and trial findings.
2026-10-07	triumph	The UMIN receipt-number URL resolved to a different registry identifier from the source label.	Trial participation inventory; both old and replacement pages checked against their displayed UMIN identifiers.	Replaced R000031885 with R000031949; retained the intended study identifier and trial findings.
2026-10-07	QUARTZ	ISRCTN3826061 was malformed and the registry API rejected it.	The ISRCTN API resolved QUARTZ as ISRCTN13826061, with acronym QUARTZ and ClinicalTrials.gov cross-reference NCT00403065.	Replaced the link with ISRCTN13826061.

2026-10-06

Date	Entity	What was wrong	How found	Fix
2026-10-06	TROP2 expression biomarker	The summary retained the withdrawn U.S. Trodelvy urothelial indication and omitted Dato-DXd's 2026 TNBC approval.	Checked FDA withdrawal, sacituzumab TNBC and Dato-DXd TNBC notices.	Scope availability to U.S. advanced-disease indications, add the 2026 TNBC updates and separate PD-L1 eligibility from the absence of a required TROP2 test.
2026-10-06	TROP2 target and TNBC molecular landscape	A copy-number range of 2-4% was labelled as protein expression/amplification and pointed first to an IHC paper; a separate ASCENT note overstated expression independence.	Compared the row's own TNBC-filtered cBioPortal count notes and study sources with the ASCENT biomarker publication.	Label the 2.5-4.4% row as gene amplification, lead with its copy-number study citation, and retain uncertainty in the low-expression ASCENT subgroup. No new cBioPortal query or protein-positivity estimate is claimed.
2026-10-06	LCB84	Modality and mechanism labelled it a monoclonal antibody even though the cited intervention text described an antibody linked to an MMAE prodrug.	Read the official title and intervention description in NCT05941507.	Classify as ADC, describe its TROP2 antibody/MMAE construct and phase 1/2 study, and replace the search link with the direct registry citation.
2026-10-06	ASP2998 study	The trial's drug links and TL;DR omitted ASP2998 and named only its combination partners.	The current registry and Astellas's public mechanism description, checked while reviewing TROP2 dual-payload programs.	Linked ASP2998 and TROP2, named the TOP1/STING payloads, and distinguished current estimated enrollment from the earlier trial-in-progress abstract. Added the missing ASP2998 drug record without duplicating its existing trial.
2026-10-06	LCB84	Listed as an unconjugated monoclonal antibody despite an ADC trial	ClinicalTrials.gov NCT05941507	Classify as ADC and retain the phase 1/2 registry qualification; overlaps the focused correction in PR #174
2026-10-06	JSKN016	Target and mechanism described as not publicly stated.	Resolved ASCO and SABCS publisher abstracts and registry intervention.	Add TROP2/HER3, TOP1 payload and a dated phase I cohort distinct from phase III registration.
2026-10-06	ESG401 and FDA018-ADC	Target and mechanism missing despite primary clinical abstracts.	Resolved ASCO abstracts and FDA018-ADC registry alias.	Add cited TROP2 mechanisms, early cohort denominators and toxicity; retain separate later development.
2026-10-06	EB-NK-301	Presented as phase 2 without making the combined phase 1/2 protocol clear.	NCT07589530.	Explicit phase 1/2 wording and unproven efficacy; conservative phase 1 headline.
2026-10-06	TROP2	Blanket low normal-tissue expression in the summary and TL;DR.	Human Protein Atlas TACSTD2 normal-tissue protein annotation and separate RNA analysis.	State normal epithelial expression and distinguish RNA, protein, membrane localization and response prediction.
2026-10-06	SHR-A1921	Publicly reported TROP2 target omitted.	Resolved Cancer Cell first-in-human paper and PubMed abstract.	Add target and cited pooled phase I results without presenting them as TNBC-specific efficacy.
2026-10-06	TROPION-Breast03	Summary retained 1,075 participants and an undifferentiated primary comparison.	NCT05629585 updated June 15, 2026.	Use actual 1,174 participants and Dato plus durvalumab versus investigator choice as primary iDFS comparison.
2026-10-06	ASCENT-05 and TROPION-Breast05	Speculative readout wording and implied future efficacy.	Current registry records.	State registered settings, status, planned enrollment, primary endpoints and estimated primary completion.
2026-10-06	ASCENT-03 / ASCENT-04 / TROPION-Breast02	Structured outcomes did not distinguish earlier publication response rates from the FDA approval analysis	FDA June 24 and May 22, 2026 approval notices	Add FDA-analysis confirmed ORR of 50% vs 47%, 61% vs 55%, and 64% vs 30%, respectively; identify source populations and preserve PFS/OS precision without claiming those response rates equal the earlier publication analysis
2026-10-06	Sacituzumab govitecan	HR-positive indication narrowed systemic therapy to chemotherapy; first-line combination omitted subcutaneous pembrolizumab and FDA-authorized PD-L1 test.	Current label and June 24, 2026 FDA notice.	Exact retained and new indications, primary links and urothelial withdrawal.
2026-10-06	Datopotamab deruxtecan	TNBC approval row added a first-line restriction absent from the FDA indication; summary relied on patient-facing OS summaries.	May 22, 2026 FDA notice and current label.	Separate indication wording from first-line supporting trial and cite randomized results.
2026-10-06	Sacituzumab tirumotecan	Chinese approvals conflated years and omitted third and fourth indications; licensing appeared as a regulatory filing with an FDA source.	Kelun-Biotech primary announcements and Merck pipeline.	Four country-specific sponsor-reported approvals, correct sources and announcement-month timeline; remove the licensing-as-filing event.
2026-10-06	olaparib	Original withdrawn ovarian treatment and current disease-specific scopes not clearly separated	Current FDA label	Separate maintenance, germline-selected breast/pancreas and HRR/BRCA prostate populations
2026-10-06	niraparib	TL;DR implied current all-comer ovarian maintenance; Akeega BRCA2 mCSPC indication absent	Current Zejula label, FDA CDx supplement and December 2025 Akeega notice	State HRD first-line and germline-BRCA recurrent scope; add specific prostate indications
2026-10-06	rucaparib	Historical taxane requirement presented without regular-approval update; withdrawn ovarian treatment called narrowed; bankruptcy coded as regulatory withdrawal	Current FDA label and December 2025 approval	Separate current from historical indications and remove business event from regulatory withdrawals
2026-10-06	PARTNER and OlympiA	PARTNER merged wild-type 2024 and germline 2025 results into one positive claim; OlympiA denominator and latest outcome source stale	Primary cohort publications and 2026 OlympiA paper	Separate populations/endpoints; preserve negative wild-type result and cite six-year primary paper with 1836 randomised
2026-10-06	imp1734	EIK1003 licensing alias and PARP target link absent	Primary Eikon statement and NCT06253130	Keep one canonical asset with EIK1003 alias and PARP relationship
2026-10-06	vb15010	Molecular target relationship absent and study simplified to phase 2	NCT06819215	Link PARP and retain phase 1/2 protocol
2026-10-06	saruparib and palacaparib	Missing selective PARP1 context; saruparib actual enrolment called planned; PETRA access stale	Current registry captures and primary publications	Refresh trial states, actual/estimated counts and clinical evidence boundaries
2026-10-06	parp	PARP2 listed as an alias despite PARP1 identifiers; implication that healthy cells simply survive	Molecular identifier ownership and clinical evidence	Clarify class scope, separate enzymes and preserve toxicity and biomarker limits

2026-10-01

Date	Entity	What was wrong	How found	Fix
2026-10-01	Lirafugratinib	Recorded as "an experimental small-molecule drug from Elevar Therapeutics in phase 2 trials". The FDA approved it on 23 September 2026 for previously treated cholangiocarcinoma with an FGFR2 fusion or other rearrangement. The record also named no cancer.	The weekly FDA feed, proposed by the update bot and checked against the FDA notice.	Status, brand, summary, the cancer, the approval row and the regional row all corrected from the FDA notice. The bot's proposed note omitted the FGFR2 restriction; the restriction is in the row.
2026-10-01	The update bot's FDA matcher	A product whose name is a combination matched any notice naming one of its partners, so "accelerated approval to vusolimogene oderparepvec with nivolumab for melanoma" was proposed as a regulatory event on Relatlimab + nivolumab. Proposed twice, on 30 September and 1 October, and caught by hand both times.	Reviewing the proposals queue.	A combination is matched only when every component is named. `scripts/propose-updates.test.ts` holds the rule.

2026-09-30

Date	Entity	What was wrong	How found	Fix
2026-09-30	Red cards on every antibody-drug conjugate (brentuximab vedotin)	The class card matched every product whose modality is "ADC", so a reader on brentuximab vedotin was told to watch for the Stevens-Johnson reaction that enfortumab vedotin carries, and a reader on an investigational ADC with no warnings got the same card.	Writing the lymphoma standard-of-care rows and reading the rendered page.	The named warnings attach by product id; the class set keeps only the neutropenic fever card. `src/lib/red-flags.test.ts` fails against the old code.
2026-09-30	Differentiation syndrome on bexarotene	The set matched the fragment "retinoid", so bexarotene, an RXR agonist given for cutaneous T-cell lymphoma, carried a warning that belongs to ATRA and arsenic trioxide in acute promyelocytic leukaemia. Bexarotene's label carries no such warning.	Same read.	The set now matches the modality phrase "differentiation agent".
2026-09-30	NICE TA892, mosunetuzumab	Recorded as available through the Cancer Drugs Fund. Recommendation 1.1 reads "is not recommended".	Reading every lymphoma appraisal on nice.org.uk.	Changed to not recommended.
2026-09-30	NICE TA524	Described as the cutaneous T-cell lymphoma appraisal. TA524 is the Cancer Drugs Fund review carrying the Hodgkin lymphoma recommendation; the CTCL appraisal is TA577.	Same sweep.	Corrected to TA577.
2026-09-30	NICE TA649, polatuzumab vedotin	Recorded as Cancer Drugs Fund; recommendation 1.1 is routine commissioning. TA933 was recorded as "not recommended" when it is a terminated appraisal.	Same sweep.	Both corrected.
2026-09-30	Fentanyl and resminostat structures	The weekly maintenance run added PubChem CID rows for two compounds that already had name-lookup rows, and the duplicate key would not compile.	Typecheck after merging the weekly maintenance branch.	The name-lookup rows removed; the explicit CID kept, because it cannot drift when PubChem renames a synonym.

2026-09-25

Date	Entity	What was wrong	How found	Fix
2026-09-25	Red flags, urgent advice (symptoms)	The `UKONS` source constant in src/data/red-flags.ts pointed at `ukons.org/site/assets/files/1134/oncology_haematology_24_hour_triage.pdf`, which 404s with an HTML page after a site move. Nine urgent-advice rows cited it.	Source sweep of every URL in src/data/red-flags.ts, checked on content type and size rather than status	Repointed at the current UKONS Oncology/Haemato-oncology 24-hour triage toolkit version 3 (`application/pdf`, 652,152 bytes), linked from ukons.org resources
2026-09-25	Red flags, urgent advice (symptoms)	Six thresholds attributed to the UKONS triage tool were not in it. "Signs of sepsis" listed cold or mottled skin and very low urine output; "Breathlessness" said blue lips; "Uncontrolled vomiting or diarrhoea" said unable to keep fluids down for 24 hours and seven or more stools a day; "Bleeding" said 15 minutes of pressure, blood in vomit, urine or stool and a sudden severe headache; "Confusion or drowsiness" said difficulty speaking and a seizure; "Rash that blisters" said sores in the mouth, eyes or genitals.	Each row read against the toolkit's own grading table after the URL was repaired	Rows reworded to the toolkit's wording (breathlessness at rest or any chest pain; 6 or more vomits in 24 hours; an increase of 7 or more bowel movements a day over pre-treatment normal; bleeding that is not self-limiting; severe confusion or an altered level of consciousness; a rash over 30 percent of the body with blistering or ulceration). The sepsis row now cites the NHS sepsis page, which states those 999 signs
2026-09-25	Red flags, urgent advice (symptoms)	Five more source URLs silently redirected: NHS blood clots, diarrhoea and vomiting, vomiting blood, and three Pancreatic Cancer UK pages under `/information/`.	Same sweep	Updated to the pages the redirects land on
2026-09-25	62 FDA label citations (accessdata.fda.gov)	The weekly link check recorded every accessdata.fda.gov label as gone: the site redirects non-browser clients to `/apology_objects/abuse-detection-apology.html` and serves it with a 404. The documents are live.	A new test comparing red-flag sources against public/links.json	`scripts/check-links.ts` now classifies a bot-wall interstitial as blocked rather than gone

2026-09-24

Date	Entity	What was wrong	How found	Fix
2026-09-24	pixantrone	Status "approved" and the text said the EU authorisation "remains conditional pending further confirmatory data"; the EMA register shows Pixuvri's authorisation expired on 12 June 2024 after Les Laboratoires Servier did not apply for renewal, and it was never approved elsewhere.	Proposals bot residue (PR 74) read by hand against the register page	status "withdrawn", TL;DR and summary state the expiry, EU row W(2024) added
2026-09-24	imp4927	The record said IMP4927's target and mechanism were "not stated"; the EMA register page for Sepalna names the substance senaparib, a PARP inhibitor (ATC L01XK07), with a positive CHMP opinion on 17 September 2026 for maintenance in advanced ovarian cancer.	Proposals bot residue (PR 74): the register named a product no corpus name matched exactly	record renamed Senaparib (IMP4297) with aliases, PARP target, EMA link and an EU under-review row
2026-09-24	pald1	Generated summary was one line ("UniProt has no function text"), although four primary papers describe paladin as a lipid phosphatase in endothelial cells, an antiphosphatase in the neural crest and a driver of colon cancer cell migration in mouse models.	Reader proposal (issue 64), checked against Europe PMC	record rewritten by hand in src/data/targets-reader-edits.ts with the papers as sources
2026-09-24	setmelanotide	The record said hypothalamic obesity after craniopharyngioma was "being tested" and listed only the 2020 genetic-obesity approval; the FDA approved the acquired hypothalamic obesity indication on 19 March 2026.	Contributor report (GitHub issue 89); confirmed against Drugs@FDA (supplement 9, approved 2026-03-19) and the Imcivree label indications via openFDA.	Added the 2026 approval row, rewrote the TL;DR and summary, and linked the TRANSCEND trial (`transcend`) and its NEJM 2026 paper.

2026-09-17

Date	Entity	What was wrong	How found	Fix
2026-09-17	checkmate-548	One record named "CheckMate 548 & CheckMate 143 & CheckMate 498" carried a single registry id (NCT02667587, which is CheckMate 548 only) and `yearReported` 2020 for all three. They are three separate trials: CheckMate 143 (NCT02017717, JAMA Oncology 2020), CheckMate 498 (NCT02617589, Neuro-Oncology 2023) and CheckMate 548 (NCT02667587, Neuro-Oncology 2022). The glioma history also dated CheckMate 143 to 2017 while the record said 2020.	Reader report by Erion Junior de Andrade, MD, MSc, neurosurgeon, Hospital São José, Santa Casa de Porto Alegre (PR #25), checked against ClinicalTrials.gov and the PubMed records (143, 498, 548)	split into `checkmate-143`, `checkmate-498`, `checkmate-548`, each with its own registry id, publication, year and outcomes; history entry moved to 2020
2026-09-17	checkmate-548	The structured overall-survival outcome (28.9 vs 32.1 months) cited `academic.oup.com/neuro-oncology/article/25/1/123`, which is the CheckMate 498 paper (Omuro et al.), not CheckMate 548 (Lim et al., Neuro-Oncology 2022;24:1935-1949).	Reader report by Erion Junior de Andrade, MD, MSc, neurosurgeon, Hospital São José, Santa Casa de Porto Alegre (PR #25), checked against PubMed (source)	`source` set to https://doi.org/10.1093/neuonc/noac116

2026-09-13

Date	Entity	What was wrong	How found	Fix
2026-09-13	nct05208047	`status` was "recruiting"; the registry says RECRUITING; primary completion 2025-09-30.	Weekly registry fact check (source)	set to "active" via apply-factcheck
2026-09-13	alphabreak	`status` was "recruiting"; the registry says ACTIVE_NOT_RECRUITING.	Weekly registry fact check (source)	set to "active" via apply-factcheck
2026-09-13	alphabreak	`phase` was "3"; the registry says PHASE2.	Weekly registry fact check (source)	set to "2" via apply-factcheck
2026-09-13	cambria	`status` was "recruiting"; the registry says ACTIVE_NOT_RECRUITING.	Weekly registry fact check (source)	set to "active" via apply-factcheck
2026-09-13	mountaineer	`status` was "recruiting"; the registry says COMPLETED.	Weekly registry fact check (source)	set to "completed" via apply-factcheck
2026-09-13	mountaineer	`phase` was "3"; the registry says PHASE2.	Weekly registry fact check (source)	set to "2" via apply-factcheck

2026-09-11

Date	Entity	What was wrong	How found	Fix
2026-09-11	acupuncture-aromatase-inhibitor-arthralgia, retifanlimab, b-combination-space, basement-membrane-tissue-barriers, kenneth-kinzler, cbt-fatigue-distress, multidisciplinary-tumour-board	Seven cited DOIs did not resolve (mistyped or belonging to another article); the intended papers were re-identified through Crossref with journal and DOI prefix pinned, for example POD1UM-303 to 10.1016/S0140-6736(25)00631-2 and the FAP locus paper to 10.1126/science.1651562. Two more (rEECur, NUT carcinoma registry) could not be identified and were removed; the small bowel adenocarcinoma guideline DOI now points at the ASCO guidelines index.	Link check (doi.org 404)	URLs corrected
2026-09-11	46 records with a Wikipedia link, 13 with a Wikipedia-only source	The Wikipedia article title in `wikipedia` or `links` no longer exists (for example Oncotype_DX, PSMA_PET, Immuno-PET). Titles were checked against the Wikipedia API; one was renamed (Biosecure Act), the rest removed. BCLC staging and cardio-oncology now cite their primary guideline papers instead.	Link check (404)	references removed or replaced
2026-09-11	33 dead pages (fda.gov, cancer.gov, institutional and company pages)	Pages returned 404 to a browser as well as to the checker; replaced with the nearest Wayback Machine snapshot and labelled "(archived copy)". 45 pages whose publisher moved them now point at the nearest live section of the same site, labelled "(page moved; nearest live section)"; 101 dead links on records that keep another source were removed. accessdata.fda.gov and doi.org answer non-browser clients with 404 or 403, so those were verified with a browser user agent and left as they were.	Link check	URLs replaced or removed
2026-09-11	elevate-tn	`nct` was NCT02475097, which does not exist on ClinicalTrials.gov; the acalabrutinib ELEVATE-TN study is NCT02475681 (acronym ElevateTN).	Registry fact check (nct-not-found), confirmed against the registry entry	set to NCT02475681
2026-09-11	ace-breast-02	`nct` was NCT04829604, which is the separate US phase 2 study ACE-Breast-03 (72 patients); ACE-Breast-02 is registered in China only.	Registry fact check: phase and enrolment disagreed, registry title names ACE-Breast-03	id removed; the primary publication is the source
2026-09-11	preopanc	`nct` was NCT02172976, a 40-patient German gemcitabine versus FOLFIRINOX study, not the Dutch PREOPANC trials.	Registry fact check: enrolment 40 versus 246 and a different title	set to NTR3709 (Netherlands Trial Register) and cite the JCO 2020 and JCO 2022 papers
2026-09-11	iberdomide	Status "phase-3" with no US approval; openFDA holds the Zenbexus label, effective 13 August 2026, for relapsed myeloma with daratumumab and dexamethasone.	Registry fact check (openFDA label exists, not recorded as approved)	status approved, approvals row and DailyMed link added; regional table updated
2026-09-11	daraxonrasib	regional-approvals.ts still said "under review" while the record and the FDA novel approvals page record approval on 26 August 2026 (Rasonque).	Audit (approvals disagree with the regional table)	regional row set to approved 2026
2026-09-11	nintedanib	Status "negative" although the EMA authorised Vargatef for lung adenocarcinoma with docetaxel in 2014; the mesothelioma failure is one programme, not the molecule's status.	Audit (status disagrees with approvals)	status approved, EMA EPAR linked, TL;DR reworded
2026-09-11	atrt, chordoma, epithelioid-sarcoma, epigenetics-roadmap	Standard-of-care rows, pipelines and a roadmap step still presented tazemetostat as available; Ipsen withdrew it from all markets in March 2026.	Audit (standard of care and pipeline cite a withdrawn item)	rows reworded to state the withdrawal, removed from pipelines, roadmap step notes it
2026-09-11	mds	Pipeline listed magrolimab, stopped in 2024.	Audit (pipeline cites a withdrawn item)	removed
2026-09-11	hairy-cell-leukemia	Row said moxetumomab pasudotox "where still available"; it was withdrawn from sale in 2023.	Audit (standard of care cites a withdrawn item)	reworded
2026-09-11	docetaxel, durvalumab, ibrutinib, lenalidomide, pegylated-liposomal-doxorubicin, temsirolimus, dabrafenib-trametinib	approvals[] began later than the first approval: docetaxel 1996 (Taxotere, breast), durvalumab 2017 (urothelial, accelerated), ibrutinib US 2013 (MCL) and EU 2014, lenalidomide 2005 (MDS del(5q)), Doxil 1995 (Kaposi sarcoma); temsirolimus EU year was 2009 (Torisel authorised November 2007, MCL added 2009); the dabrafenib plus trametinib combination was approved in the US in January 2014 (regional table said 2013, the single agents).	Audit (approvals disagree with the regional table) and Drugs@FDA earliest original approval	rows added or corrected
2026-09-11	lacc	Text quoted a death hazard ratio of 6.00 while the only structured outcome carried the disease-free survival HR 3.74.	Audit (number in the text disagrees with the structured field)	overall survival outcome added from NEJM 2018
2026-09-11	peace-1	Result gave only the overall-population OS HR 0.82 while the structured outcome carried the ADT plus docetaxel population HR 0.75.	Audit (number in the text disagrees with the structured field)	result states both populations
2026-09-11	bevacizumab, cabozantinib, everolimus, ivosidenib, lenvatinib, letrozole, ramucirumab, regorafenib, sunitinib, trabectedin, Eisai, Exelixis, Incyte	Each was defined in full by two or three cancer deep dives with different summaries, mechanisms and approval rows; only the first copy was ever shown and the rest were silently hidden.	Audit (duplicates with different values)	second copies became supplements: relationship arrays, extra approval rows and cancer-specific facts as notes
2026-09-11	alfa-0701	`enrolled` was "271"; the registry says 280 (ACTUAL).	Registry fact check (source)	set to "280" via apply-factcheck
2026-09-11	alter-0303	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	amplify-7p	`phase` was "2"; the registry says PHASE1/PHASE2.	Registry fact check (source)	set to "1/2" via apply-factcheck
2026-09-11	apt-trial	`enrolled` was "410"; the registry says 406 (ACTUAL).	Registry fact check (source)	set to "406" via apply-factcheck
2026-09-11	ascertain-v	`phase` was "2"; the registry says PHASE1/PHASE2.	Registry fact check (source)	set to "1/2" via apply-factcheck
2026-09-11	belinda	`enrolled` was "322"; the registry says 330 (ACTUAL).	Registry fact check (source)	set to "330" via apply-factcheck
2026-09-11	camel	`enrolled` was "412"; the registry says 419 (ACTUAL).	Registry fact check (source)	set to "419" via apply-factcheck
2026-09-11	cartitude-5	`status` was "recruiting"; the registry says ACTIVE_NOT_RECRUITING.	Registry fact check (source)	set to "active" via apply-factcheck
2026-09-11	checkmate-9dw	`enrolled` was "668"; the registry says 732 (ACTUAL).	Registry fact check (source)	set to "732" via apply-factcheck
2026-09-11	checkmate-9er	`enrolled` was "651"; the registry says 701 (ACTUAL).	Registry fact check (source)	set to "701" via apply-factcheck
2026-09-11	cll14	`enrolled` was "432"; the registry says 445 (ACTUAL).	Registry fact check (source)	set to "445" via apply-factcheck
2026-09-11	anbl0532	`enrolled` was "652"; the registry says 665 (ACTUAL).	Registry fact check (source)	set to "665" via apply-factcheck
2026-09-11	compete	`enrolled` was "309"; the registry says 324 (ACTUAL).	Registry fact check (source)	set to "324" via apply-factcheck
2026-09-11	desktop-iii	`enrolled` was "407"; the registry says 408 (ACTUAL).	Registry fact check (source)	set to "408" via apply-factcheck
2026-09-11	destiny-breast12	`enrolled` was "504"; the registry says 506 (ACTUAL).	Registry fact check (source)	set to "506" via apply-factcheck
2026-09-11	dreamseq	`enrolled` was "265"; the registry says 267 (ACTUAL).	Registry fact check (source)	set to "267" via apply-factcheck
2026-09-11	emerald-3	`enrolled` was "725"; the registry says 760 (ACTUAL).	Registry fact check (source)	set to "760" via apply-factcheck
2026-09-11	epcore-nhl-1	`phase` was "2"; the registry says PHASE1/PHASE2.	Registry fact check (source)	set to "1/2" via apply-factcheck
2026-09-11	foenix-cca2	`phase` was "2"; the registry says PHASE1/PHASE2.	Registry fact check (source)	set to "1/2" via apply-factcheck
2026-09-11	golseek-1	`status` was "recruiting"; the registry says ACTIVE_NOT_RECRUITING.	Registry fact check (source)	set to "active" via apply-factcheck
2026-09-11	her2climb-02	`enrolled` was "463"; the registry says 466 (ACTUAL).	Registry fact check (source)	set to "466" via apply-factcheck
2026-09-11	herizon-gea-01	`enrolled` was "914"; the registry says 920 (ACTUAL).	Registry fact check (source)	set to "920" via apply-factcheck
2026-09-11	horizon-breast01	`enrolled` was "366"; the registry says 381 (ACTUAL).	Registry fact check (source)	set to "381" via apply-factcheck
2026-09-11	imroz	`enrolled` was "446"; the registry says 475 (ACTUAL).	Registry fact check (source)	set to "475" via apply-factcheck
2026-09-11	inavo120	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	keynote-483	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	ion-trial	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	izabright-breast01	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	lacc	`enrolled` was "631"; the registry says 636 (ACTUAL).	Registry fact check (source)	set to "636" via apply-factcheck
2026-09-11	legend-2	`phase` was "1"; the registry says PHASE1/PHASE2.	Registry fact check (source)	set to "1/2" via apply-factcheck
2026-09-11	lidera	`enrolled` was "4200"; the registry says 4170 (ACTUAL).	Registry fact check (source)	set to "4170" via apply-factcheck
2026-09-11	litespark-005	`enrolled` was "746"; the registry says 755 (ACTUAL).	Registry fact check (source)	set to "755" via apply-factcheck
2026-09-11	lume-meso	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	maps	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	matterhorn	`enrolled` was "948"; the registry says 957 (ACTUAL).	Registry fact check (source)	set to "957" via apply-factcheck
2026-09-11	mevpro-1	`status` was "recruiting"; the registry says ACTIVE_NOT_RECRUITING.	Registry fact check (source)	set to "active" via apply-factcheck
2026-09-11	orient-32	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	orient-32	`enrolled` was "571"; the registry says 595 (ACTUAL).	Registry fact check (source)	set to "595" via apply-factcheck
2026-09-11	outback	`enrolled` was "919"; the registry says 926 (ACTUAL).	Registry fact check (source)	set to "926" via apply-factcheck
2026-09-11	persevera	`enrolled` was "978"; the registry says 992 (ACTUAL).	Registry fact check (source)	set to "992" via apply-factcheck
2026-09-11	phergain	`enrolled` was "356"; the registry says 377 (ACTUAL).	Registry fact check (source)	set to "377" via apply-factcheck
2026-09-11	quilt-3-032	`phase` was "2"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	rxponder	`enrolled` was "5083"; the registry says 5018 (ACTUAL).	Registry fact check (source)	set to "5018" via apply-factcheck
2026-09-11	senticol-iii	`status` was "recruiting"; the registry says ACTIVE_NOT_RECRUITING.	Registry fact check (source)	set to "active" via apply-factcheck
2026-09-11	stellar	`enrolled` was "80"; the registry says 82 (ACTUAL).	Registry fact check (source)	set to "82" via apply-factcheck
2026-09-11	viale-a	`enrolled` was "431"; the registry says 443 (ACTUAL).	Registry fact check (source)	set to "443" via apply-factcheck
2026-09-11	waveline-003	`phase` was "3"; the registry says PHASE2/PHASE3.	Registry fact check (source)	set to "2/3" via apply-factcheck
2026-09-11	her2climb	`phase` was "3"; the registry says PHASE2.	Registry fact check (source)	set to "2" via apply-factcheck
2026-09-11	imcgp100-202	`phase` was "3"; the registry says PHASE2.	Registry fact check (source)	set to "2" via apply-factcheck
2026-09-11	ken-she	`phase` was "3"; the registry says PHASE4.	Registry fact check (source)	set to "4" via apply-factcheck
2026-09-11	amplify-7p	`enrolled` was "144"; the registry says 158 (ACTUAL).	Registry fact check (source)	set to "158" in trial-outcomes.ts
2026-09-11	eortc-26981	`enrolled` was "573"; the registry says 575 (ACTUAL).	Registry fact check (source)	set to "575" in trial-outcomes.ts
2026-09-11	keynote-355	`enrolled` was "847"; the registry says 882 (ACTUAL).	Registry fact check (source)	set to "882" in trial-outcomes.ts
2026-09-11	litespark-012	`enrolled` was "1688"; the registry says 1854 (ACTUAL).	Registry fact check (source)	set to "1854" in trial-outcomes.ts
2026-09-11	pathfinder-2	`enrolled` was "35878"; the registry says 35883 (ACTUAL).	Registry fact check (source)	set to "35883" in trial-outcomes.ts
2026-09-11	rasolute-302	`enrolled` was "460"; the registry says 500 (ACTUAL).	Registry fact check (source)	set to "500" in trial-outcomes.ts
2026-09-11	tropion-breast03	`enrolled` was "1075"; the registry says 1174 (ACTUAL).	Registry fact check (source)	set to "1174" in trial-outcomes.ts
2026-09-11	tropion-lung01	`enrolled` was "604"; the registry says 605 (ACTUAL).	Registry fact check (source)	set to "605" in trial-outcomes.ts
2026-09-11	flaura2	`enrolled` was "557"; the registry says 587 (ACTUAL).	Registry fact check (source)	set to "587" in trial-outcomes.ts
2026-09-11	impassion131	`enrolled` was "651"; the registry says 653 (ACTUAL).	Registry fact check (source)	set to "653" in trial-outcomes.ts
2026-09-11	latitude	`enrolled` was "1199"; the registry says 1209 (ACTUAL).	Registry fact check (source)	set to "1209" in trial-outcomes.ts
2026-09-11	olympia	`enrolled` was "1836"; the registry says 1837 (ACTUAL).	Registry fact check (source)	set to "1837" in trial-outcomes.ts
2026-09-11	psmafore	`enrolled` was "468"; the registry says 469 (ACTUAL).	Registry fact check (source)	set to "469" in trial-outcomes.ts
2026-09-11	splash	`enrolled` was "412"; the registry says 455 (ACTUAL).	Registry fact check (source)	set to "455" in trial-outcomes.ts
2026-09-11	therap	`enrolled` was "200"; the registry says 201 (ACTUAL).	Registry fact check (source)	set to "201" in trial-outcomes.ts
2026-09-11	vision	`enrolled` was "831"; the registry says 861 (ACTUAL).	Registry fact check (source)	set to "861" in trial-outcomes.ts
2026-09-11	xalute	`enrolled` was "675"; the registry says 707 (ACTUAL).	Registry fact check (source)	set to "707" in trial-outcomes.ts

2026-09-07

Date	Entity	What was wrong	How found	Fix
2026-09-07	nhs-galleri	Trial summary said final results were pending; the trial had reported at ASCO 2026 (primary endpoint not met, stage IV incidence down ~14%).	Pipeline-intelligence review while writing the ASCO 2026 digest	ac8b9f3

2026-09-06

Date	Entity	What was wrong	How found	Fix
2026-09-06	ezh2, epigenetic-drugs, sarcoma	Described tazemetostat (Tazverik) as approved in epithelioid sarcoma and follicular lymphoma. Ipsen withdrew it from all markets and indications on 9 March 2026 after SYMPHONY-1 showed excess secondary haematologic malignancies.	Companies-expansion review; verified against the FDA alert and Ipsen's press release	1d0eb06
2026-09-06	pubmed-europepmc	TL;DR was a two-word fragment that failed the readability check.	Automated test (`tldr` length)	27e0eb6

Proposed by the update bot (pending review)

2026-09-16: 44 proposals drafted from the feeds (32 high confidence); review in docs/proposals/latest.md. Nothing here is a confirmed correction until a human moves it above.
2026-09-24: PR 74 carried 21 EMA rows a person had to read; 6 were applied by hand (olaratumab, pixantrone, alitretinoin, cediranib, ensartinib, imp4927) and 15 left as no change (non-cancer medicines, a withdrawn biosimilar application, a pending re-examination); the table is in the PR comment.

How corrections are logged

Anyone can report an error via a GitHub issue or pull request.
A correction entry names the entity, states precisely what was wrong, says how it was found (reader report, automated audit, fact-check job, congress update, reviewer), and links the fixing commit.
The automated audit and weekly registry fact-check surface candidates; only confirmed errors are logged here.

Found an error? Open an issue or a pull request on GitHub; the fix gets logged here. Source: CORRECTIONS.md. See also the automated audit.