Changelog

All notable changes to OnCo are recorded here. The format follows Keep a Changelog. Dates are release dates; the corpus itself is kept current continuously.

[Unreleased]

Regenerated from the commit log each time the site ships (scripts/changelog-sync.ts); the version sections below are written by hand when a release is cut.

3 October 2026

• How recovery is measured, and who is reached by it
• The recovery front gets its hub, its history and its open problems
• Rehabilitation, reconstruction, and a matrix of what comes back
• Recovery: transplant, second cancers, the mind, and the markdown nobody was rendering
• Childhood and young adult survivorship: the cohorts that made the field
• Six summaries that described the page instead of the thing
• chore: EU regional rows read from the EMA register pages (#127)

2 October 2026

• How recovery is measured, and who is reached by it
• Name the England quality of life survey as a gap rather than describe it
• How recovery is measured, and who is actually reached by it
• The recovery front gets its hub, its history and its open problems
• Recovery: the front's hub rewritten, the history of survivorship, and the research agenda
• Rehabilitation, reconstruction, and a matrix of what comes back
• What comes back after treatment: a sourced recovery matrix, treatment by treatment
• Cancer rehabilitation and reconstruction: the services that put function back, and the gap between them and what people are offered
• Recovery: transplant, second cancers, the mind, and the markdown nobody was rendering
• Graft-versus-host disease, late effects after transplant, and the immune system afterwards
• Fear of recurrence, mood, body image, work, money and sleep after cancer
• Second cancers after treatment, and the screening that follows
• Childhood and young adult survivorship: the cohorts that made the field
• Childhood, adolescent and young adult survivorship: the cohorts, the burden, and the follow-up that governs care
• Six summaries that described the page instead of the thing
• Six TL;DRs were talking about the page instead of the subject
• Recovery and rejuvenation: a new front, and the hair spike the owner asked for
• Biological ageing after cancer treatment, and what is sold against it
• Recovery and rejuvenation: the body, system by system
• AGENTS.md learns the five days since it was written
• The home page stops drawing a flow the corpus does not have
• Hair: the number for your own regimen, and an honest grade for everything sold for regrowth
• The home page stops drawing a flow the corpus does not have, and two labels come off
• A twentieth front: Recovery and Rejuvenation
• Lymphoma, and a home page with less above the grid
• The hero stops promising a plain-English TL;DR
• The home page loses the record count, the JSON chip and the body pill above the grid
• chore: EU regional rows read from the EMA register pages (#125)

1 October 2026

• Lymphoma: the biology of the germinal centre, and what a person actually decides
• Lymphoma: the UK and NHS pathway, and the world geography layer
• Lymphoma evidence: 29 trials, 42 papers, a roadmap and eight open questions, each resolved against a registry or Europe PMC
• Nineteen lymphoma entity pages, a rewritten non-Hodgkin hub, and both 2022 classifications quoted where they disagree
• The oldest three DLBCL rows retire, because the record now says the same thing three times
• Lymphoma molecular layer: the antigens and what they cost, the lesions, the classifications, resistance and the tests
• Lymphoma, the person: six decisions, six red cards, three first-sixty-days lists
• The weekly check now fails when a page grows past what a reader downloaded last week
• A ceiling on what a reader actually downloads, measured on the live site
• Lirafugratinib is approved, not phase 2, and the bot stops proposing one combination's approval for another
• chore: monthly identifier enrichment (HGNC, Wikidata) (#124)
• chore: refresh HTA decisions and survival snapshots (#123)
• chore: refresh pulse, abstract and citation snapshots (#122)
• chore: weekly roadmap registry check (public/roadmap-watch.json)
• chore: refresh institution research output from OpenAlex (#121)
• chore: draft change proposals that need review

30 September 2026

• Four verified approval rows from the proposals queue, and two structure keys that stopped the build
• Page titles, second attempt: a noun phrase naming the thing and the dimension that orders it
• The unattended gates run the tests the way the ship chain does, so a weekly data refresh can merge itself again
• chore: refresh preprint tracker from Europe PMC
• chore: refresh universe lists (NCI drugs and types, NCI centres, OECI, NHS alliances, NLM journals, OpenAlex, KEGG, ChEMBL, ClinicalTrials.gov, FDA OCE)
• chore: check regional approvals against the EMA register
• chore: weekly citation link check
• chore: refresh FDA approvals feed (OCE notifications, openFDA drugsfda)
• chore: draft change proposals that need review

29 September 2026

• Lymphoma treatment rows, and red cards that no longer name a drug the reader is not taking
• Lymphoma treatment: 70 standard-of-care rows across 21 diseases, each citing its trial
• The software of oncology, and the top of a record page
• chore: weekly maintenance (logos, structures, trial links, acronyms, orphan links)

28 September 2026

• Every cancer-page section named for what is in it, in nine languages
• The block inside Symptoms and diagnosis is not called that too, and four pill labels follow the new names
• The page that shows what is waiting, with the queue read from the open pull requests
• A command for a person to type is not a change to the product
• The queue is a rendering of the open pull requests, refreshed on every ship
• The admin page waiting in its own queue
• The page that shows what is waiting, and says plainly what it is not
• The record page top waiting in the queue
• The top of a record page is the record again: a family strip of six, two paragraphs of summary, the clinic names readable, the model panel last
• The software map waiting in the queue
• The map a product proposal belongs in: 136 pieces of oncology software, each row saying whether anything outside the company backs it up
• The title review waiting in the queue
• Thirty page titles that named a format, could belong to any site, or needed a term first
• Product changes now wait for the owner, and the plan for the cancer page is written
• The approval mechanism belongs on main; the renames stay in the queue
• The cancer page measured before it is redesigned, and a plan the owner can approve part by part
• The first product change waiting in the queue
• Product changes wait for the owner; data ships
• Each section of a cancer page is now named for what is inside it, and the years index says which years it holds
• The front page links to 140 sections and shows almost none of them
• The prevalence matrix: 23 columns wide and growing with the corpus
• Visual first, and the measurement that would be misleading if taken naively
• Every page title, with the count of how many are one or two words
• The machine links, and the admission that seven patches are not a design
• The years index title, and the question of how it differs from the timeline
• The section names and the wall of links, with what each section holds and how many links a page carries
• The section called Evidence is called Trials and papers, which is what is in it
• Three more record-page asks on the roadmap, with what the print control actually does
• The fourth ask: the also-known-as line is the smallest type on the page
• Three owner asks on the roadmap: what a reader meets first on a record page
• Five weekly runs merged, the agent rules written down with the scripts they point at, and a card that can name no record no longer counts as one that lost one
• A card that can name no record is not a card that lost one
• Provenance regenerated with the private name redacted after the weekly run reintroduced it
• A locally installed skill directory is not part of this project
• The rules an agent needs, in the file every agent reads, with the scripts they point at now in the repository
• chore: render this week's issue to public/newsletter/
• chore: weekly fact check, audit, and provenance refresh
• chore: refresh public/votes.json from Discussions reactions
• chore: refresh trial counts from ClinicalTrials.gov
• chore: draft change proposals that need review

27 September 2026

• A total review of fatalistic language: 153 phrases rewritten, the survival guard now says what a median is, and the red-flag card leads with calling
• A cancer with no NHS pathway page now says so in development as it does in production
• A word that is the source's is shown as the source's
• The half a keyword search cannot find: order, framing and the numbers printed as verdicts
• Say what the cancer does, not how to feel about it: fatalistic language out of OnCo's own voice, every quotation left exactly as its author wrote it
• Three dated fields the corpus lacked: accelerated approvals with their confirmations, a sourced first-description year on targets, and a trial start date
• Adenoid cystic carcinoma described without the word relentless

25 September 2026

• Three dates the timeline was missing: every accelerated approval and what became of it, when each target was first described, and when each trial opened
• Measure what a reader downloads, not what the tests can render
• A record for every year and one timeline over all of them, with ten questions asked of the corpus and two it cannot answer
• Oncolytic virotherapy in full: four approved replicating viruses and not one randomised survival benefit, and the 2024 self-experiment read from its case report
• A record for every year in oncology, one timeline over all of them, and what the shape of the dates shows
• The for-me budget measures its own tiles, not the chrome every page carries
• Oncolytic virotherapy as a subject, not scattered records: the interferon defect, the approvals and what they really showed, and the virologist who treated herself
• Clicking a value in a table now filters by it: thirteen cells across eight tables, a year facet on trials, and a guard that a filter link can never come back empty
• Click the year to filter the table: thirteen cells across eight tables become filters
• Three owner asks on the roadmap: cells as filters, a record per year, and oncolytic virotherapy
• Six countries in depth, each read from its own registry and regulator: the United States, Germany, Japan, the United Kingdom, Israel and Russia
• Wave one of the country deep dives is done: six countries
• Four country pages said the figure excluded non-melanoma skin cancer; it never did
• The benchmarking partnership is hosted by Cancer Research UK, not founded by it
• Israel: a committee decides in public, once a year, inside a fixed sum, which cancer drugs the state will pay for
• One label over four health services: the United Kingdom deep dive, and the cost-per-QALY threshold the corpus had wrong
• Japan reads tenth in our own count and fifth in the world, so the Japan page is written from Japanese sources
• Russia read from Russian sources: the Herzen register and what it does not count, who pays, and the Eurasian approvals route
• Two country deep dives: the United States, where the question is who pays, and Germany, where a drug is paid for on the day it is approved and assessed afterwards
• The first two country deep dives: the United States and Germany
• In America the pill has a ceiling and the drip does not, and 239 accelerated approvals show what happens when nobody checks
• A red-card pattern can no longer match inside an unrelated word
• Compugen, which found PVRIG by predicting it, and the paper that first measured what a liposome does in a patient
• Germany pays for a new cancer drug on the day it goes on sale and asks what it is worth afterwards
• A blood-clot warning for a myeloma drug was reaching every fluoropyrimidine page, because the pattern matched inside the word pyrimidine
• Remove the scratch script used to survey what the corpus held for Israel
• Israel: the people and papers behind the country page, and the term for what a founder variant is
• The skin review: a page about a cream warned a reader about a blood clot, and an appraisal that refused a drug was recorded as one that funds it
• A second front page at /v2 for review, and table filters that find what you asked for: family cancers, phases in time order, search ranked by what you typed
• Country deep dives on the roadmap, ranked by what the corpus holds and by what that ranking gets wrong
• Filtering a table by a cancer family finds its subtypes, the phase filter runs in time order, and the facet search ranks what you typed
• The front page opens on the list again; the graph is one click away
• Four page-render tests were measuring how busy the machine is, not the code
• Four launch-list items were already done: tick them so nobody does them twice
• The answer to which page is mine now sits on the page a reader lands on, not the last tab
• Skin cancer on the NHS: nearly a third of England's urgent cancer referrals are for skin, and the commonest skin cancer is excluded from the standard that measures them
• Two layers of the skin round wrote the same trial: the ratchet caught it and the merge tool fixed it
• Record which of the skin layer's sources a program cannot open, checked one URL at a time
• The skin cancer NHS pathway, and the two exclusions that mean the published figures are not what they look like
• A reader on a breast subtype page now gets the cord compression card, and never two sepsis cards
• Skin cancer: what a person with a basal cell or squamous cell cancer faces, the family tree given its reason, and the counting rule that hides the commonest cancer there is
• The breast family review: the family page duplicates nothing, but the two receptor pages it routes to said everything twice
• What a person with a basal cell or squamous cell skin cancer actually faces, and a rule that lets a row cite the leaflet that carries its sentence
• The treatment of the keratinocyte cancers: the margin, Mohs surgery, the creams and the light, radiotherapy, the hedgehog pathway and the transplant recipient
• Three review lessons from the skin layer: a citation that does not contain the claim passes every gate
• The rows about living with a skin cancer now cite the leaflet that actually contains the sentence, and the photodynamic therapy figures move to the cream the trials used
• The skin family given its routing, the keratinocyte cancers given their subtypes and staging, and the counting rule that governs every figure on them
• A patient leaflet may be the source for a row about living with a disease, if the row names who wrote it
• What a person with a basal cell or squamous cell skin cancer actually faces: the choice between surgery and its alternatives with the trial figures, the operation on the face, Mohs as a day, and the ones that are not cured
• The family roll-up does not assume who is reading, and stops printing the cancer's name twice in one sentence
• Family cancer pages now reach what their subtypes hold: 3,683 trials, 1,079 medicines and 3,534 centres that no family page could reach before
• A family page now shows the trials, medicines and centres its subtypes hold, grouped by the subtype each came from
• The breast family page given its surgery, its NHS pathway and its shared decisions, and 83 duplicate records retired onto the ones that carry the reading
• The prostate review found an appraisal that refused a drug recorded as merely restricting it, and the breast family page given the routing a reader needs
• Breast cancer in the NHS: the screening programme the site never had, the family history service, both referral routes, and four wrong or stale funding claims corrected
• One thing, one id: 83 duplicate records merged, with the rule, the helper and the ratchet the three reviews asked for
• Breast cancer, the shared treatment spine: 22 trials on the operation, the axilla, radiotherapy and reconstruction, read at twenty and twenty-five years
• Breast cancer, living with it: the surgery choice with the twenty-year figures, the axilla and lymphoedema, fertility, sensation, hair, and four red cards
• The breast family page: the routing a reader needs in one click, the shared story of grade, stage and the receptor tests, and the classification edition nobody had noticed changed
• Prostate review: the appraisal that refused abiraterone was recorded as one that restricted it, eleven papers written twice, and the screening trials the whole section rests on
• The lung review: 33 lines the layers had written twice folded away, four contradictory figures resolved against their sources, and two duplicate-key bugs the lung data exposed
• The lung review: a placeholder idea rewritten from the UK page's own gaps, 33 duplicated base lines folded, three papers that existed twice merged, and two duplicate-key bugs the lung data exposed
• Prostate cancer in depth: 57 trials from their publications, the receptor landscape from eighteen deposits, the March 2026 screening recommendation, and four wrong NHS funding rows corrected
• Wire the two prostate pathology terms into the papers that found them
• Prostate molecular: the androgen receptor's five escape routes, the repair genes the four PARP labels disagree about, and what a PSMA PET cannot tell you
• The explained page budget measures its furniture, not how much the corpus knows
• Prostate glossary: fifteen new terms, with the screening figures, the two endpoint definitions and PI-RADS against the NICE Likert scale
• Prostate cancer on the NHS: the screening committee's March 2026 recommendation from its own coversheet, the risk in Black men with its cohorts, and the 62-day standard prostate fails worst
• Prostate treatment from its trials: 57 randomised comparisons, what each option costs a man, and the four NICE refusals
• Prostate cancer: five orthogonal classifications settled, the Cambridge Prognostic Groups added, and the decisions written with the trial figures rather than the word side effects
• Register the twelve prostate glossary drafts and link them from the decisions that use them
• Prostate cancer evidence: the 44 papers that made the field, a roadmap in twelve eras, seven ideas and nine open problems
• Prostate cancer, living with it: the ProtecT trade-off in numbers, hormone therapy for the long run, and cord compression named as the emergency
• An agent that has only the origin can now find the corpus: MCP and llms descriptors under .well-known, licence and provenance on every API response
• Prostate cancer has five classifications at once, and only one of them makes pages
• How to read the UK bodies that answer a script with a wall
• Lung cancer on the NHS: the screening programme from its own protocol, 34 centres, 21 appraisals and 8 refusals, the deprivation gradient from four national sources
• The UK lung pathway hangs off the family, not one branch of it
• Lung cancer gets its UK and NHS layer: the screening programme with its risk thresholds, the deprivation gradient in four nations, 108 funding decisions and the reflex testing clock
• Lung cancer given its trials, its molecular landscape and its evidence: 36 trials written from source, three landscapes, 44 papers, a roadmap of eleven eras
• A standard-of-care row for the antibody-drug conjugates and bispecifics entering lung cancer
• Lung molecular layer: adenocarcinoma, squamous and small-cell landscapes read separately, PD-L1 by assay, why tumour mutational burden failed, and resistance mechanism by mechanism
• Approval dates cut back to what the FDA notification page shows
• Lung cancer evidence: 44 landmark papers from Doll and Hill to tarlatamab, a roadmap in eleven eras with a 2032 registry watch, and seven open questions
• Registry routing reads the title and conditions, not the eligibility text
• Roadmap page quoted the old review gauge label
• Write down the three gauges no script can move, and what the owner holds
• Lung registry snapshot gets a breadth rule, later lines get their trials, and Ask keeps its floor
• Lung cancer given a family and a patient side: five tiers under lung-cancer, ten decision rows, 86 appointment questions, six red-card sets
• Europe PMC snapshots for 1,545 records, 42 logos, and a collection that had borrowed a charity's mark
• Take the backlink fill back out: it costs 0.002 of Ask recall, and the floor does not move
• Lung cancer: the patient side, written so a person who smoked and a person who never smoked are both addressed
• Read the three new lung pages in the dev server and drop one redundant alias
• Three more lung glossary terms the other layers will need, and an outlook paragraph on the family page
• Glossary terms: four Wikipedia links found, 58 absences recorded as checked
• Lung taxonomy settled and the core layer written: the family now hangs together, and lung cancer is described from its sources
• Lung cancer treatment layer: 884 registry trials, 31 landmark trials and the England access rows
• A test that fails when a gauge's denominator moves and its label does not
• Backlink fill for 1,002 weak records: two acronym collisions fixed first
• Provenance was blind to the generated files: 1,486 records get their commit line back
• Three gauges were measuring the wrong population; fix the questions, not the targets
• Colorectal reviewed against its sources: duplicate treatment rows folded, eleven NICE errors corrected, RAS wild-type recorded as a biomarker, 1,406 links checked
• Review the colorectal deep spike: fold the duplicated rows, settle the contradictory figures, fix eleven NICE errors and add the RAS wild-type readout
• Colorectal cancer deep spike complete (core, 710 trials, molecular landscape, evidence and roadmap); red cards re-sourced against the live toolkit; pages no longer grow with what is known about a reference; weekly jobs merge only when the gates pass
• Roadmaps and dossiers made robust to corpus growth: a reference costs the same whatever is known about it, the dossier export is a file, and no bot merges a tree the tests have not seen
• Check every cited URL and fix the one that was dead
• Register the colorectal treatment spike, its helper files and the sponsor supplement with the audit
• De-duplicate the NICE NG151 link across the three colorectal patches, and the CMS paper on its own term
• Thirteen glossary terms the evidence, living-with and molecular layers asked for, and the eight living-with drafts registered
• Colorectal: the historical trials the evidence and molecular layers cite, and targets for ten pipeline drugs
• Colorectal cancer deep spike, round one: the UK and NHS pathway across four nations with the national audit indicators, decisions, first 60 days, red cards and the adjuvant chemotherapy aid
• Red flags re-sourced: the UKONS triage tool found at its new URL, nine thresholds quote-checked against it, six redirects fixed, soft 404s caught by the weekly link check
• Colorectal: company links for the trials the registry names, and the sponsor backfill that keeps the rankings honest
• Colorectal molecular layer: 38-row genomic landscape from eight cBioPortal cohorts, sidedness as biology, the adenoma and serrated routes, CMS, immunity and ctDNA, with 47 paper records
• Colorectal cancer, the UK and NHS layer: the screening programme and its four-nation differences, the faecal immunochemical test that now governs referral, the National Bowel Cancer Audit, 13 NICE appraisals with the 2026 bevacizumab reversal, and the UK trial legacy
• Colorectal cancer deep spike: the evidence, the roadmap and the ideas
• Colorectal core, second pass: sidedness and consensus molecular subtype supplements, the screening standards source, and three claims tightened against their sources
• Colorectal: anti-EGFR rechallenge and the microsatellite-stable immunotherapy attempt
• Colorectal trials: cross-links into the records already in the corpus, and two corrections
• Colorectal cancer, living with it, care and decisions: ten decision rows, 74 appointment questions, a 19-item first 60 days checklist, six red-card sets and a NICE NG151 adjuvant decision aid
• Fold the duplicates the base colorectal record and spike carried: five history entries, seven biomarker lines, three state-of-the-art points and three open problems written twice
• Colorectal treatment layer: 38 hand-written trials from MOSAIC to ALTAIR, 661 registry trials, verified NICE and FDA approval rows, palliation
• Colorectal core layer: sourced epidemiology, the early-onset rise, risk, referral, screening, staging, pathology and prevention on the colorectal record; four subtype pages and eleven glossary terms

24 September 2026

• LAUNCH: pancreatic spike and the cooperative-group home ticked
• Pancreatic reviewed against its sources: duplicates folded, figures settled, 1,006 links checked, mobile overflow fixed
• Pancreatic deep spike review: duplicates merged across seven files, daraxonrasib approval and NICE terminations settled, 13 paper duplicates supplemented, 47 redirected URLs rewritten, seven glossary terms and two journals, UK and tool pages lay out at 390 px; docs/PANCREATIC-QA.md
• Pancreatic cancer deep spike, round four: the molecular landscape with allele-level KRAS shares, subtype classes, precursor sequence and 82 papers
• Pancreatic molecular layer: the genomic landscape, subtypes, precursors, stroma and monitoring markers with their cohorts
• Pancreatic cancer deep spike, round three: 30 trials with outcomes, 284 registry trials, verified NICE and label rows, fifteen standard-of-care rows; Ask ranking weights curated records
• Ask concept index: inverse document frequency over curated records only; extractive recall 0.408 to 0.427, floor raised to 0.42
• Pancreatic treatment layer gates: cooperative groups under companies, pancrelipase flagged supportive, deep-dive tag described, stroma-enzyme schematic rule
• Pancreatic cancer deep spike, round two: the record extended with sourced epidemiology, risk, diagnosis and surveillance, six subtypes, 22 terms, 34 papers and a twelve-era roadmap
• Pancreatic core: classical versus basal-like (GATA6) and clinical benefit response terms for the evidence layer, high-risk individual folded into familial pancreatic cancer, disease-free survival note on the EFS term; the other requested ids resolve to existing records
• Pancreatic treatment layer: ESPAC-1 record, shorter TL;DRs, duplicate label links removed, explained-page budget note
• Pancreatic evidence: UNICANCER is a cooperative-group company, listed under companies
• Pancreatic treatment layer: 32 hand-written trials, 284 registry trials, drug approvals by regulator, standard-of-care export
• Pancreatic core: register the seven living-with glossary drafts (enzyme draft folded into pancreatic-enzyme-replacement, new-onset diabetes alias repointed), drop URLs the UK and living patches already link, subtype rows cite the neoadjuvant term rather than PRODIGE 24 so /explained/ stays under budget, audit list and stale zh summary updated
• Pancreatic cancer evidence: 34 Europe PMC paper records, a roadmap from Whipple 1935 to the 2031 registry watch, nine ideas
• Pancreatic cancer deep spike, round one: the UK and NHS pathway with the national audit findings, decisions, first 60 days, red cards and the first-treatment aid
• Pancreatic core layer: sourced epidemiology, risk factors, presentation and emergency share, diagnosis, resectability and TNM 8, pathology and precursors, surveillance programmes and outlook on the pancreatic record; six WHO subtype pages; thirteen glossary terms; five supplements
• Pancreatic cancer living with it, care and decisions: ten decision rows, 90 hand-written questions, first 60 days checklist, six red-card sets with the 111 and 999 wording, first-treatment decision aid from NICE NG85, sourced living-with notes on 26 side-effect and procedure records
• Pancreatic cancer UK and NHS page: NG12 diabetes rule, 28/31/62-day standards, NPaCA 2026 findings, 23 HPB hubs, NICE TA476/TA440/TA750/TA1052 and SMC decisions verified, PERT audit gap, EUROPAC and Precision-Panc legacy, four-nation registry figures
• Cooperative groups have one home: 35 trial groups move from institutions to cooperative-group companies with redirects
• One home for cooperative groups: 35 consortium institutions become cooperative-group companies (row 251)
• TNBC reviewed against its sources: figures reconciled, duplicates folded, 822 links checked
• Roadmap watch row: text column shrinks and wraps on phones
• TNBC QA: de-duplicate the six patches, correct KEYNOTE-522, BRCA share, NG101 1.3.6, TA992 and TROPION-Breast05, fix four dead links, roadmap and UK page fit 390 px; docs/TNBC-QA.md
• Wave 4: 99 cancer entity pages under the taxonomy rule; roadmaps page their eras; subtype pages rank below their parents in Ask
• Ask OnCo retrieval: subtype capped below its parent, provenance weight shared by both stages; extractive recall 0.403 to 0.411, floor held at 0.41
• LAUNCH: decisions and shipping of 25 Sept 2026
• Section planner test allows eight paged sections for NSCLC after the lung histologies; wave 4 pages weigh 0.7 in the Ask index
• Wave 4 cancer pages, family 7: sixteen pituitary, thymoma, salivary, cervical, vulvar and vaginal entities; roadmap Progress row
• Cancer hubs with section sub-pages, TNBC deep spike complete (core, treatments and 235 trials, molecular landscape), Ask recall repaired, route progress bar and edge caching, Methods and models glossary with 129 terms
• Roadmap pages: eras collapse behind Show era pills fed by /api/v1/roadmaps/<id>.json; watch table paged past 20; 300 KB markup budget
• Ask OnCo retrieval: drop function words and rank by kind and name in the word search; extractive recall 0.329 to 0.416, floor 0.41 with a ratchet
• Wave 4 cancer pages, family 6: fourteen rare tumour entities; cancer map SVG budget 320 KB
• TNBC treatment: reword a code comment to the page name
• TNBC treatment: LOTUS, PAKT, SCAN-B and ZEST detail, registry trials linked to subtype pages, test fixtures for the new rows
• Wave 4 cancer pages, family 5: sixteen lymphoma and leukaemia entities from WHO-HAEM5
• EntityDetail: remove the conflict markers committed with the architecture merge, drop the unused Tip import
• TNBC molecular layer: sourced genomic landscape, Lehmann and Burstein subtypes, ancestry and founder mutations, ctDNA studies, assay-specific PD-L1 and HER2-low thresholds, 55 landscape papers
• Wave 4 cancer pages, family 4: twenty colon, kidney and testicular entities; sibling linker so no new page arrives orphaned
• TNBC core: register the eight living-with glossary terms drafted by the living file, drop the brand-name aliases that hijacked Ask, trim two TL;DRs and a jargon phrase
• TNBC treatment deep dive: trial and drug records, approvals by regulator with NICE appraisal numbers, 229 registry trials, standard-of-care rows
• record-fold test: TNBC ceiling follows the new layers until the hub architecture lands, then drops to 300 KB
• CanSim terms, methods group: 58 "Methods and models" glossary terms with one-line definitions and why they appear in cancer papers; gap document gets a Done column
• CanSim terms, oncology group: 70 sourced glossary terms and the cancer-ai-vocabulary hub, 11 HGNC-sourced gene targets, TCGA cohort codes as cancer aliases, 12 data-source rows, Methods and models category
• Loading experience: navigation progress bar, route skeletons, intent prefetch on dense lists, pages hot at the edge, above-the-fold checks
• Decisions page holds 390 px: referenced-record and guideline pills truncate with tooltips, grid children may shrink
• TNBC core layer: sourced epidemiology, risk, presentation, diagnosis, pathology, staging, screening and outlook on the tnbc record; twelve subtype pages; nineteen glossary terms
• Modality hubs for twelve formats, TNBC deep spike round one (UK and NHS pathway, evidence and roadmap, decisions, first 60 days, red cards, after-chemotherapy aid), supportive care flag on 55 drugs, 22 cooperative-group records
• Geography section fits 390 px: grid items may shrink and source labels wrap
• Wave 4 cancer pages, family 3: fourteen breast histologies under breast-cancer from the WHO 2019 breast classification
• TNBC living with it, care and decisions: eleven decision rows, 77 hand-written questions, first 60 days checklist, three red-card sets with the 111 and 999 wording, after-chemotherapy decision aid, sourced living-with notes on 20 side-effect and procedure records
• TNBC UK and NHS layer: pathway page for tnbc with the screening caveat, NG12 and NG101 steps, every NICE and SMC decision including the TA992 refusal against SMC2608, R208 and M234 tests, eight UK-open trials with sites, TNT, POSH/OlympiA and PARTNER legacy, registry figures read from PHS, WCISU and NICR downloads, Breast Cancer Now and three researcher records, named gaps
• WIP: tnbc-core spike draft before merging main
• Cooperative groups as company records: cooperative-group company type with browser facet, glyph pill and tooltip; 22 group records (GORTEC, German CLL Study Group, French Sarcoma Group, AGO, AIO, MRC CTU, HORG, DPCG, DBCG, DAHANCA, GERCOR, ISG, SSG, EMN, IFM, ALFA, AMC, OCOG, KGOG, SARC, FIL, FILO) linked to 41 corpus trials as sponsor; sponsor aliases carry their ids and the IFM and KGOG mis-resolutions are fixed; the trials fetchers link and never give products to them
• Supportive care medicines: supportive drug flag on 55 records, "Supportive care" pill with glyph and tooltip, Purpose facet on /drugs/, excluded from the home treatments count, the product rankings and the treatment references on standard-of-care rows; EMA and FDA fetchers flag candidates by a stated indication rule
• Cancer records as a hub with ten sections: registry, section pages, shared strip, sections.json, budgets and deep-link tests
• Wave 4 cancer pages, family 2: ten lung histologies under nsclc from the WHO 2021 thoracic classification
• Front navigation: the kinds drawn as a linked graph, the body map as the entry point on phones; ship chain builds on Vercel only
• TNBC evidence: 45 papers verbatim from Europe PMC (guidelines, receptor and basal-like discovery, BRCA1 founder mutations, disparities, subtypes, pCR and RCB, platinum, CREATE-X, OlympiA updates, PD-L1 assays, TILs, TROPION-Breast02, CAPItello-290, HER2-low, ctDNA), tnbc-roadmap with 14 eras and a registry watch list to 2033, nine ideas
• Modality lens: /modalities/ and twelve format hubs assembling what the corpus records per shape of medicine
• Home page: the counts grid becomes a living graph of the kinds, with the body map as the entry point on phones
• Wave 4 cancer pages, family 1: entity pages under the seven parents without subtype records, taxonomy rule in docs/CANCER-PAGES.md
• LAUNCH: the ship chain no longer builds locally; Vercel builds and the chain verifies the alias
• LAUNCH: issue and PR sweep of 24 Sept, checkpoints, target statuses, ivermectin
• Ivermectin: the trial evidence so far, with the label toxicity table and the papers under expressions of concern stated
• Ivermectin evidence page: drug record with label toxicity table, four registry trials, 22 papers with expressions of concern and retraction stated, compact fenbendazole and mebendazole records, per-file citation test
• Checkpoint families mapped with immune and cell-cycle hubs, 26 new targets; 31 reader proposals applied as 23 trial records and 27 papers; Cancersim terms gap filed
• Cancersim terms gap: 378 of 504 terms not yet in OnCo, listed by bucket with the decision the methods terms need; IMPROVEMENTS row 250
• Target specificity and distribution on 904 targets with where each is found, gallbladder decision aids and comparison, world map of gallbladder cancer rates, six EU register rows decided by hand, PALD1 rewritten
• Issue 98: LITESPARK-011 (NCT04586231) readout written onto the record: Lancet 2026 dual primary progression-free and overall survival rows and toxicity as stated in the abstract, ASCO GU 2026 response, duration and treatment-related toxicity rows marked as conference data, Motzer 2026 paper record linked, plain-English summary, simple sentence and Chinese TL;DR
• Issue 97: INSITE (NCT03686215) trial record for pegulicianine with the NEJM Evidence 2023 coprimary endpoints and the existing Smith paper linked, pegulicianine.trials wired; status mixed because the sensitivity threshold was not met
• Issue 96: HBI0101 / NXC-201 multiple myeloma phase 1 (NCT04720313) trial record with the Haematologica 2023 response, survival and cytokine release endpoints, Asherie 2023 paper record, nxc-201-car-t wired to the trial and to multiple myeloma
• Issue 95: ScreenTrustCAD (NCT04778670) trial record for Lunit INSIGHT MMG with the Lancet Digital Health 2023 paired-reader detection endpoints, Dembrower 2023 paper record, lunit-insight-mmg.trials wired
• Issue 94: PSMA-PreRP (NCT03368547) trial record for Illuccix and 68Ga-PSMA-11 with the JAMA Oncology 2021 nodal accuracy endpoints, Hope 2021 paper record, illuccix.trials and ga68-psma-11.trials wired
• Issue 80: Vistogard Studies 1 and 2 (NCT01432301) trial record and Cancer 2017 paper; uridine-triacetate.trials wired with label link
• Issue 79: Study 0201 (NCT00003991) trial record and Blood 2006 paper; histamine-dihydrochloride.trials wired
• Issue 78: Study 201 (NCT00168064) trial record and JAMA Dermatology 2013 paper; mechlorethamine wired to the trial and to cutaneous-t-cell-lymphoma
• Issue 77: Study 2005-01 (NCT00358501) trial record and Blood 2016 paper; defibrotide.trials wired with label link
• Issue 76: CT041-ST-01 (NCT04581473) enriched in place with Lancet 2025 outcome and paper; satricabtagene-autoleucel.trials wired, false rivoceranib link and stale skip dropped
• Issue 75: MSB-GVHD001 (NCT02336230) trial record and BBMT 2020 paper; remestemcel-l.trials wired
• Issue 93: ROMAN (NCT03689712) trial record for avasopasem manganese with the eClinicalMedicine 2025 severe oral mucositis endpoints, Anderson 2025 paper record, avasopasem-manganese.trials wired; status mixed because the paper records the FDA complete response
• Issue 72: FOCUS (NCT02678572) enriched in place with Ann Surg Oncol 2024 outcomes; hepzato.trials wired with paper and label links
• Issue 71: eXalt3 (NCT02767804) enriched in place with JAMA Oncology 2021 outcomes; ensartinib cites the paper and label; stale zh TL;DR dropped
• Issue 70: Act.In.Sarc (NCT02379845) trial record and Lancet Oncology 2019 paper; nbtxr3.trials wired
• Issue 69: ROCKstar (NCT03640481) trial record and Blood 2021 paper; belumosudil.trials wired
• Issue 92: VISION (NCT02864992) enriched in place for tepotinib with the Paik NEJM 2020 response, duration and toxicity outcomes and the existing paper linked; capmatinib-tepotinib.trials wired to GEOMETRY mono-1 and VISION; geometry-mono-1 leads capmatinib.trials; simple sentence and Chinese TL;DR follow the record
• Issue 68: ARCHER 1050 (NCT01774721) enriched in place with separate Wu 2017 and Mok 2018 rows and papers; dacomitinib cites both
• Issue 67: ASCEND-4 (NCT01828099) enriched in place with Lancet 2017 outcomes and paper; ceritinib cites the paper
• Issue 66: CK-301-101 (NCT03212404) trial record and JITC 2023 paper; cosibelimab.trials wired, label and paper links replace the generic FDA listing
• Issue 65: ALTA (NCT02094573) trial record and JCO 2017 paper; brigatinib.trials wired to ALTA and ALTA-1L
• Glossary and record pages: checkpoint disambiguation term owns the bare word, Checkpoint families pills on target, term, pathway, technology and mechanics stage pages
• Checkpoint hubs: /checkpoints/ with a pill switch, /checkpoints/immune/ with the interactive synapse, /checkpoints/cell-cycle/ with the cell-cycle ring, class tables, JSON companions, nav entry in nine languages
• Checkpoint map: typed taxonomy of immune and cell-intrinsic checkpoints, 26 new targets from HGNC and UniProt, derived rows and tests
• Issue 91: Alliance A071401 (NCT02523014) trial record for GSK2256098 with the JCO 2023 six-month progression-free survival endpoints, Brastianos 2023 paper record, gsk2256098.trials wired
• Issue 90: PREEMPT CRC (NCT04369053) trial record for SimpleScreen CRC with the JAMA 2025 accuracy endpoints, Shaukat 2025 paper record, simplescreen-crc.trials wired
• Issue 89: TRANSCEND (NCT05774756) trial record for setmelanotide with the NEJM 2026 body-mass index and hunger endpoints, Miller 2026 paper record, setmelanotide refreshed with the 19 March 2026 US approval for acquired hypothalamic obesity and a CORRECTIONS row
• Issue 88: UX023T-CL201 (NCT02304367) trial record for burosumab with the JBMR 2021 serum phosphorus and fracture-healing endpoints, Jan de Beur 2021 paper record, burosumab.trials wired
• Issue 87: PATRICIA (NCT00122681, HPV-008) trial record for the bivalent HPV vaccine with the Lancet 2009 vaccine-efficacy endpoints, Paavonen 2009 paper record, hpv-bivalent-vaccine.trials wired
• Issue 85: XM01-22 (ISRCTN08063129) trial record for epoetin theta with the Archives of Drug Information 2011 haemoglobin response and transfusion endpoints, Tjulandin 2011 paper record, epoetin-theta.trials wired
• Issue 84: rolapitant Study 3 (NCT01500226, TS-P04834) trial record with the Lancet Oncology 2015 delayed-phase complete response endpoint, Schwartzberg 2015 paper record, rolapitant.trials wired and label links made explicit
• Issue 64: PALD1 (paladin) rewritten from the primary papers instead of the proposed GeneCards text; hand-written target records from reader proposals live in src/data/targets-reader-edits.ts and replace the generated gene record of the same id
• PR 74 residue decided by hand: six EU rows from the EMA register pages (olaratumab revoked 2019, pixantrone expired 2024, alitretinoin withdrawn 2021, cediranib application withdrawn 2016, ensartinib and senaparib positive CHMP opinions as under review); pixantrone status corrected to withdrawn, IMP4927 identified as senaparib (Sepalna) with PARP target and aliases, cediranib aliases; 15 rows closed as non-cancer medicines or pending
• Issue 83: XM22-03 (EudraCT 2009-015999-10) trial record for lipegfilgrastim with the BMC Cancer 2013 cycle 1 neutropenia endpoint, Bondarenko 2013 paper record, lipegfilgrastim.trials wired, id allowlisted as a phase 3 without an NCT
• Issue 82: MNTX 302 (NCT00402038) trial record for methylnaltrexone with the NEJM 2008 coprimary laxation endpoints, Thomas 2008 paper record, methylnaltrexone.trials wired and label links made explicit
• Gallbladder cancer geography: GLOBOCAN 2022 rates by country and sex for 185 countries, a reusable CancerGeography choropleth with incidence, mortality and sex pills, the high-burden regions with their papers, Chile's GES prophylactic cholecystectomy programme, India's NCRP registry rates and Japan's registry and guideline practice, prevention evidence, India and Chile spotlights, a geography.json companion and a gallbladder section on /countries/in/
• Target specificity explainer links the Human Protein Atlas collection record, whose licence now reads CC BY-SA 3.0
• Target specificity tests: enums cover every populated value, every row has a note and an https source, the graph carries the fields, HER2 reads tumour-associated and many types with a tumour-agnostic mark, BCR::ABL1 tumour-specific and one type, CD19 lineage antigen, PD-1 immune; browser facet counts match the data; the explainer and the pills render
• Target specificity in the pages: pills with glyphs and tooltips on target pages, a "Where it is found" section from the Human Protein Atlas rows, specificity and cancer-type facets on the target browser and the gene hub, and the explainer at /targets/specificity/
• Target specificity: blood-cancer family regex no longer swallows "(all types)" names, family names for the merged blood families, own-record sources for targets the atlases do not cover
• IMPROVEMENTS: rows 150 to 249, the 100 upgrade ideas of 24 September 2026
• Target specificity and distribution: schema, fetcher and data for 904 targets with an approved or clinical-stage drug
• Record pages stay inside a phone screen at 390 px; 130 more trials carry posted registry results
• Gallbladder decision aids and comparison: polyp aid on the 2022 European guideline, incidental cancer aid on AHPBA, Søreide, Ethun and CAPBIL, gallbladder compared with cholangiocarcinoma and ampullary cancer
• Record pages fit a 390 px phone and lose the Tabs key warning: grid children shrink (min-w-0 on Neighbours and What changed cards, *:min-w-0 on EntityDetail grids), GuidelineChip and DrugChip wrappers cap at max-w-full, tables in EntityDetail, InvestorPanels, StartupPanels, BiomarkerDetail, PrevalenceTable, ToxicityTable, AccessTable and ResearchOutput scroll inside ScrollRow with the year bars; each tab's content is a fragment keyed on the tab id so Flight no longer hands Tabs a keyless array; one record of every kind is a width-only check in scripts/mobile-audit.ts
• Registry passes over the gallbladder spike: outcomes for every NCT trial without its own, status re-read for all 4,102 NCT trials, key papers for 10 gallbladder trials
• LAUNCH: gallbladder deep spike of 24 Sept 2026
• Gallbladder cancer deep spike, final round: review reconciled the six facets to their sources, Wales and Northern Ireland figures, Cancer Drugs Fund status, prescription and benefit rules, lighter cancer pages
• Gallbladder deep spike review: duplicate terms, trials and history entries become supplements, HER2 and NICE figures reconciled, dead and redirected links fixed, backlinks for the ctDNA roadmap and the chemo-immunotherapy pairing, long relation lists capped with deep links so the cancer page drops from 997 KB to 652 KB, QA checklist in docs/GALLBLADDER-QA.md
• Gallbladder UK layer, second pass: Wales and Northern Ireland registry figures with stage, Scotland checked and found not to publish C23, CDF list v1.365 status on every biliary drug, four-nation prescription and special-rules benefit pages, GLH turnaround statements, CAPBIL tied to BILCAP, Hassan Malik record
• Gallbladder cancer deep spike, round three: treatments and trials, 131 registry trials, approval rows with NICE numbers, standard-of-care rows
• Gallbladder evidence: ARTEMIDE-Biliary01 points at its existing registry record instead of a second trial record
• Gallbladder cancer deep spike: the record rewritten with UK and world figures, eight subtypes, molecular landscape, UK and NHS pathway page, 40 papers, roadmap and ideas, decisions, first 60 days and red cards; weekly bot refreshes merged
• Gallbladder cancer treatments and trials: surgery, adjuvant, systemic, HER2 and palliation records with 131 registry trials
• Gallbladder cancer deep spike, first round: the record rewritten with UK and world figures, eight subtypes, molecular landscape, decisions, first 60 days and red cards
• Citation counts for the gallbladder evidence papers
• Gallbladder cancer roadmap, ideas, guideline and trial papers, terms and history
• Gallbladder cancer evidence: surgical, staging, UK, polyp, prevention, HER2 and ctDNA papers from Europe PMC
• UK and NHS page per cancer: /cancers/<id>/uk/, cancer-page pill, sitemap and JSON companion
• Gallbladder cancer: UK and NHS layer data, HPB centres, AMMF and UK trialists
• Gallbladder cancer record rebuilt with sourced UK, world and US figures, TNM 8th edition in plain words, risk factors with evidence, screening rules and patient FAQ; eight subtype pages and ten glossary terms
• Spike patches reach every cancer record wherever it lives, and fail loudly when the cancer id is missing; gallbladder landscape guard test
• Spike patches also merge into NCI coverage cancer records
• Gallbladder cancer molecular layer: sourced genomic landscape, regional cohorts, premalignant sequence, immune findings, HER2 testing, 30 landscape papers; spike supplements onto records other files own
• Gallbladder cancer: cancer-scoped red cards, dated first-60-days checklist, incidental-cancer journey
• Pipeline engine: a biparatopic antibody with one declared target resolves both arms at high confidence; zovodotin is a named auristatin payload
• Gallbladder cancer living-with spike: decision rows, question set and glossary records
• chore: weekly roadmap registry check (public/roadmap-watch.json)
• chore: refresh pulse, abstract and citation snapshots (#86)
• chore: refresh institution research output from OpenAlex (#81)
• chore: EU regional rows read from the EMA register pages (#73)

23 September 2026

• Top bar fits at 1280 px with a classic scrollbar
• LAUNCH: owner asks of 23 Sept 2026
• Drugs browser fits its card with capped cells and clearer approval years; NHS-Galleri performance paper, trial outcomes and Galleri test page
• Header fits a 1280 px window with a classic scrollbar: star count folds into its icon, tighter nav gaps and 13 px labels between xl and 2xl
• NHS-Galleri test-performance paper (Nature Medicine, 22 Sept 2026) as a paper record, trial rebuilt on the registry and the abstract, result on Galleri, MCED and the ctDNA roadmap
• Drugs browser fits at 1280 and 1440: capped list cells with a "+N more" pill, Approved as first year plus muted latest with region flags, results tables scroll inside their card only when measured too wide
• Biomarker readouts under their parent genes at /biomarkers/, genome hub pages its genes
• Gene hub pages its role sections: first 30 genes per role in the HTML, the rest from one tables file, role and evidence deep links over the full set
• Proposals bot merges register-verified rows itself, mechanics of cancer rebuilt with 56 stage pages, mobile patterns for views where a control drives a visual
• Regenerate target cross-references for the seven new parent targets (HGNC and ChEMBL, 23 Sept 2026)
• Dependency map gets a touch path, and wide matrices scroll inside their own box
• Biomarker pages: /biomarkers/ browser with gene, measurement, cancer and approval facets; readout pages with a Part of strip; target, drug, term and tumour-test pages link their readouts
• Biomarker readouts: 80 records under 40 parent genes, thresholds quoted from FDA labels and the companion diagnostic list
• Add the biomarker kind: schema, KIND_META, glyph, colours, nav and kind labels in nine languages
• Mechanics atlas: a page per stage and a light journey hub
• Cancer gene layer of 1,447 genes, open-source oncology map, evidence-backed combination proposals on the engine, daily spotlight, iodine target correction
• LAUNCH: proposals bot, what merges itself and what waits
• Nightly proposals: auto branch squash-merged after the gates, review PR labelled needs-review with a reason on every EMA row
• Proposals bot decides EU rows from the EMA medicine pages: page status, exact attribution, biosimilar rule, drift check, apply script
• Regional approvals io helper: parse and write the hand-written file with a round-trip test, and R() takes a source
• Two-column tools get a phone layout: sticky preview, Choose and View pills, inline results
• Cancer genes, tier 4: 389 genes associated with cancer without a proven role
• Cancer genes, tier 3: 332 drivers called by cohort mutation analysis
• Cancer genes, tier 2: 507 genes with clinical evidence on their variants
• Cancer genes, tier 1: 219 genes with an approved or late-stage drug recorded against them
• Cancer gene layer: target roles, evidence tiers and sources, fetch-cancer-genes script, genome hub and browser facets
• Home spotlight rotates by kind each day: one hero record per kind (most connected), build day's set in the HTML, reader's day swapped in from /api/v1/spotlight.json without layout shift
• ADC payload fields and setidegrasib note: record only what the records say
• Open pipeline engine, tag power search pages and full-width tabs, Edge type and For you filters, API licence wording, For me browser-only notice
• Open-source catalogue tests: unique ids, https URLs, SPDX or placeholder licences with notes, every corpus id exists, 200 sources on the generation date, curated entries generated or explained
• Open source in oncology page: paged browser with category, licence, language, openness and activity facets, intro on how oncology is done in the open, and cross-links from technology, collection, organisation and data-source pages
• Open-source oncology catalogue: schema, curated list of 379 projects and the fetcher that reads licence, stars and dates from the GitHub API
• Radioactive iodine no longer linked to SSTR2: I-131 enters thyroid cells through the sodium-iodide symporter, which has no target page yet
• Document the open pipeline: how proposals are scored, searched and contributed
• Sixty combination proposals, each searched against trials, papers and patents
• Schema for combination proposals: a scored, evidence-backed cell of the modular-drug grid
• api layout: drop the entries duplicated by the engine merge
• Record pages: full-width tab strip above both columns; tag chips open /tagged/ power-search pages
• API page licence wording matches the footer and about page; For me says it is browser-only with a signed-in version coming
• Open drug engine: every medicine taken apart into its modules, and the permutation grid per format with each cell classified from records
• Edge: filter by type or For you, with URL state, per-cancer deep links and per-type feeds
• Search ranks cancers and treatments first and groups results by kind
• Search ranks cancers and treatments above journals and papers, with kind-grouped dropdowns
• Top bar without the tagline, with the weekly bot refreshes merged
• Top bar without the tagline
• Top bar: drop the tagline beside the wordmark; it took space the navigation needs
• chore: refresh universe lists (NCI drugs and types, NCI centres, OECI, NHS alliances, NLM journals, OpenAlex, KEGG, ChEMBL, ClinicalTrials.gov, FDA OCE) (#61)
• chore: check regional approvals against the EMA register (#60)
• chore: refresh preprint tracker from Europe PMC (#59)
• chore: refresh FDA approvals feed (OCE notifications, openFDA drugsfda) (#58)
• chore: weekly citation link check (#57)

22 September 2026

• Health gauges explain people without papers by role and exempt the closed section list
• Kind-size gauge holds open kinds only: the 19 treatment fronts are a closed OnCo taxonomy and are named in the note
• People-papers gauge counts roles that produce no papers as explained: 1,528 of 1,547 with a note on who is still without
• LAUNCH: chains 101 to 104 ticked
• Kind browser pages ship 60 rows and load the rest on demand, 402 registry trials for drugs and technologies, glossary terms without a Wikipedia article say so
• Kind browsers page their rows: first 60 in the HTML, the rest from /api/v1/tables/kind-<route>.json
• Content roadmap: wave 5 trial run recorded, headline numbers refreshed from content:gaps
• Wave 5 technologies: 32 technologies without a trial gain 65 registry trials and 7 corpus links from ClinicalTrials.gov
• fetch-entity-trials: hold technologies whose aliases name something else, stop words for ex vivo, Catalyst, STRIDE and hormone names
• Wave 5 drugs: 142 drugs without a trial gain 337 registry trials and 19 corpus links from ClinicalTrials.gov
• 75 glossary terms marked wikipediaChecked 2026-09-22; NO_ARTICLE comment lists retired
• fetch-entity-trials: ClinicalTrials.gov trials for drugs and technologies without one (wave 5, second half)
• Terms record a checked-and-absent Wikipedia state: wikipediaChecked field, gauge counts it as explained, page says so
• fetch-term-wikipedia: Wikipedia links for glossary terms, exact title or redirect match with a sense check
• Paper pages for 736 cited DOIs, 1,381 bare references now link to them
• Wave 7 batch 4: the last 155 paper pages for cited DOIs, 736 in all, 1,161 citing records linked; Progress row
• fetch-cited-papers: a preprint is "posted on bioRxiv", not "published in bioRxiv (preprint)"
• Trial statuses follow the registry with dated notes, iobenguane records merged
• Wave 7 batch 3: 194 more paper pages for cited DOIs, 581 in all, 1,042 citing records linked; intismeran autogene gains its aliases
• Wave 7 batch 2: 192 more paper pages for cited DOIs, 387 in all, 905 citing records linked
• Wave 7 batch 1: 195 paper pages for the most-cited DOIs without one, 785 citing records linked
• fetch-cited-papers: name citing pages by kind and count only, no year in TL;DR or id
• Registry status data: 13 trial status changes from 3,542 ClinicalTrials.gov records (active -> completed 5, active -> recruiting 3, recruiting -> active 2, completed -> withdrawn 2, active -> withdrawn 1) with dated notes and why-stopped text; iobenguane-i-131 (Azedra) merged into i131-mibg with its approval, brand, aliases, links and a redirect stub
• Registry status pass: scripts/fetch-registry-status.ts reads overallStatus, whyStopped and the registry dates for every trial with an NCT id and writes src/data/trial-registry-status.ts, applied in src/data/index.ts through applyRegistryStatus with a dated note; result statuses kept, hand-written statuses newer than the registry never downgraded; iobenguane-i-131 listed as a duplicate of i131-mibg in dedupe MERGES
• Wave 7 fetcher for cited DOIs: scripts/fetch-cited-papers.ts writes a paper page per DOI a record cites without one
• Explained page pages its sections, four duplicate companies and one duplicate drug merged with redirects
• Page /explained/ sections at 10 rows: 1,135 KB to 561 KB of markup, budget lowered to 600 KB
• LAUNCH: content waves through chain 100
• Merge five duplicate records: Orca Biosystems into orca-bio, InxMed (Shanghai) into inxmed, Cellectis S.A. into cellectis, the second Valius Sciences into valius, and the gap-fill nogapendekin alfa inbakicept into nogapendekin-alfa; trials, cancers, drugs, investors, aliases, links and a logo move to the kept records, 30 references repointed, sponsor aliases updated, stale simple text and translations dropped, redirects and static stubs added; biocon/biocon-biologics, jw-pharmaceutical/jw-therapeutics, astrazeneca/fusion-pharma and gtx/oncternal-therapeutics reviewed and kept apart
• Duplicate-record finder and merger for companies: scripts/dedupe-companies.ts flags pairs by normalised name (legal suffixes, descriptors and city words stripped), shared website domain, or shared Wikidata or Wikipedia identifier, prints hand-reviewed MERGE and SKIP verdicts, and with --apply moves the loser's fields onto the winner, removes the loser literal, rewrites every reference across src/data (dropping duplicates inside id arrays, renaming or dropping keyed simple, translation and wikidata entries), moves or deletes the logo and registers the redirect build-redirect-stubs.ts expects; institutions and drugs are report-only unless listed as obviously identical; the source-editing helpers in scripts/orphan-links.ts are exported for reuse
• Papers for people and ideas, trials for ideas and drugs for companies from wave 6
• Content roadmap: wave 6 progress row (people without a paper 318 to 300, ideas without a key paper 1,104 to 1,026, ideas without a trial 1,047 to 1,044, companies without a drug 720 to 718) with what was held back and why
• Wave 6 company drugs: 2 companies gain a drug through the ClinicalTrials.gov lead-sponsor field (Impact Therapeutics IMP4927, R-Pharm RPH-051), each verified on the registry record; 75 other sponsors held back because their corpus trials name only comparators with a maker on record
• Registry results for 144 trials, 24 more standard-of-care trials with outcomes, papers and standard-of-care rows linked
• Wave 6 idea evidence: 78 ideas gain key papers (124 new Europe PMC records, 98 existing reused) and 3 gain trials (4 ClinicalTrials.gov phase 2 or 3 records written, 4 corpus trials linked), matched on the drug or target the idea's text names plus its cancer
• heavy-pages test: explained budget follows the corpus after the trial waves, with the paging rule for the next rise
• Content gaps: the trial-acronym measure treats the programme stem before a hyphen as a known trial name; roadmap notes the 33 remaining tokens are all in the wave SKIP list
• Content wave 3, rare cancers: FIRSTMAPPP, InPACT, GETUG 13, ADIUVO, ADIUVO-2, TROPHIMMUN and BALLAD, 4 with structured outcomes, 4 primary papers, 13 standard-of-care rows linked
• Content wave 3, myeloma: DRAMMATIC (SWOG S1803, MRD-guided maintenance after transplant) recorded from the registry and linked from the myeloma maintenance row; NCI sponsor-of-record institution links removed across the wave
• Content wave 3, endometrial: RAINBO programme (four molecular-class trials including POLEmut-BLUE de-escalation) recorded from the registry; POLE-ultramutated standard-of-care row linked
• Roadmap Progress: wave 7 registry outcomes row (144 of 2,845 ingested trials gain 272 outcome rows from 153 records with posted results; 114 requests)
• Wave 7 data: registry outcomes for the first 2845 registry-ingested trials by family burden (144 trials with 272 outcome rows from 153 records with posted results)
• Content wave 3, ovarian: AURELIA (bevacizumab with single-agent chemotherapy in platinum-resistant ovarian cancer) with structured outcomes and its JCO 2014 paper; two standard-of-care rows linked
• Wave 7 data: registry outcomes for the first 2500 registry-ingested trials by family burden (144 trials with 272 outcome rows from 153 records with posted results)
• Content wave 3, head and neck: CONTINUUM (sintilimab added to induction-concurrent chemoradiotherapy in locoregionally advanced nasopharyngeal carcinoma) with structured outcomes and its Lancet 2024 paper; standard-of-care row linked
• Wave 7 data: registry outcomes for the first 2000 registry-ingested trials by family burden (111 trials with 209 outcome rows from 117 records with posted results)
• Wave 6 people papers: 18 people gain up to three most cited papers each from Europe PMC by author and affiliation (46 new paper records, 6 existing reused), administrators filtered by role before searching, 2 ambiguous names skipped with reasons
• Content wave 3, colorectal: CAPP2 (aspirin in Lynch syndrome, 10-year follow-up) with structured outcomes and its Lancet 2020 paper; Lynch syndrome standard-of-care row linked
• Wave 7 data: registry outcomes for the first 1500 registry-ingested trials by family burden (85 trials with 171 outcome rows from 91 records with posted results)
• Wave 7 data: registry outcomes for the first 1000 registry-ingested trials by family burden (61 trials with 126 outcome rows from 65 records with posted results)
• Content wave 3, lung: IMpower010 is the existing registry record nct02486718, which gains the IMpower010 alias, its positive status and disease-free survival outcomes from the registry results section and the Lancet 2021 report; JCOG0802 skipped (UMIN only)
• Content wave 3, anal family: ANCHOR, POD1UM-303/InterAACT-2, InterAACT and NCI9673 with structured outcomes, 5 primary papers, 6 standard-of-care rows linked
• Explained page: the Open explanation pill carries only its label; glyph and Close state move to CSS; row classes and cross-reference titles shortened
• Content wave 3, sarcoma family: ACTG A5263/AMC 066 (paclitaxel against etoposide or bleomycin-vincristine with antiretroviral therapy for AIDS-associated Kaposi sarcoma) with structured outcomes and its Lancet 2020 paper; Progress table rows for gastric, pancreatic and sarcoma
• Content wave 3, pancreatic family: 7 standard-of-care trials (ESPAC-3, ESPAC-4, CONKO-001, NORPACT-1, LAP07, ESPAC-5, PRECEDE), 6 with structured outcomes, 6 primary papers, 7 standard-of-care rows linked
• LAUNCH: chains 97 and 98 ticked
• Wave 7 data: registry outcomes for the first 500 registry-ingested trials by family burden (39 trials with 78 outcome rows from 40 records with posted results)
• Wave 7 script: fetch-registry-outcomes.ts copies structured outcomes from the ClinicalTrials.gov results section into a side file merged by index.ts
• Content wave 3, gastric family: CLASSIC (adjuvant CAPOX after D2 gastrectomy) with structured outcomes and two primary papers; SKIP list extended for UMIN- and ISRCTN-only trials
• Wave 1 trial papers from Europe PMC, wave 3 standard-of-care trials, wave 5 targets and EMA products and sponsors, seven heavy table pages and idea rankings load rows on demand
• Paged tables: the heavy StaticTable pages carry 30 rows and fetch the rest from /api/v1/tables/
• Content wave 3: structured outcomes for 8 hand-written trials from the ClinicalTrials.gov results section (ARANOTE, CheckMate 915, PACIFIC-2, INTELLANCE-1, ATEMPT, ASPEN, DESTINY-Gastric02, AREN0532); roadmap Progress notes the families not reached
• Content wave 3, lymphoma family: 7 standard-of-care trials (FoRT, GALLIUM, RELEVANCE, TRIANGLE, SYMPATICO, BMT CTN 0803, PTLD-1), all with structured outcomes, 7 primary papers, 9 standard-of-care rows linked
• Content roadmap: wave 5 progress row (targets 30 to 0, EMA products 12 to 0, sponsor companies 21 to 15) with what was not written and why
• Wave 5 sponsor companies: 6 trial sponsors and 2 EMA authorisation holders from Wikidata and the ClinicalTrials.gov sponsor field (Criterium, Artios Pharma, Ambrx, Nerviano Medical Sciences, Qurient, Immunovaccine, Grünenthal, 4SC); drugs and companies waves registered
• Wave 5 EMA products: 7 register products written from their EPAR pages (crisantaspase, burosumab, epoetin theta, human normal immunoglobulin, fentanyl, resminostat, Tacquell), 3 matching faults fixed with aliases, vedolizumab excluded
• Wave 5 targets: 26 pages for the symbols two or more drug records name without a target page (NRAS, CD28, HLA-A, MPL, SHP2, STAT5, IKZF1, JAK1, CD80, ABL1, FAK, IKZF3, PTCH1, CD137, HIF-1α, IDO1, BRD4, IRF4, B7-H4, TET2, NF1, ACVR1, HOXA9, MEIS1, BIM, BAX)
• Content wave 3, childhood family: 9 standard-of-care trials (ANBL1221, ANBL0531, ANBL17P1, AEWS0031, SIOPEL-3, SIOPEL-4, AGCT1531, AHOD1331, EuroNet-PHL-C2), 7 with structured outcomes, 6 primary papers, 11 standard-of-care rows linked
• Content wave 3, brain and CNS family: 10 standard-of-care trials (ACNS1123, ACNS0122, ACNS0333, ACNS0121, NOA-08, SJMB12, ACNS1422, IELSG32, MATRix/IELSG43, INTUITT-NF2), 8 with structured outcomes, 8 primary papers, 12 standard-of-care rows linked
• Content wave 3, leukaemia family: 18 standard-of-care trials with registry records and structured outcomes, 12 primary papers, standard-of-care rows linked
• Fetched paper records link only their trial and journal; wave 1 progress row; citation counts for the new papers
• Idea rankings page carries the first 30 rows of each view and fetches the rest per view from /api/v1/ideas/rankings/<view>.json
• Prostate trials: key papers from Europe PMC for the prostate family
• Skin trials: key papers from Europe PMC for the skin family
• Colorectal trials: key papers from Europe PMC for the colorectal family
• Breast trials: key papers from Europe PMC for 77 trials, 85 new paper records, 22 long-term follow-ups
• Lung trials: key papers from Europe PMC for 125 trials, 134 new paper records, 25 long-term follow-ups
• Cancer map DAG, every hand-written table on the shared header filter, explained page loads explainer bodies on demand, 3,056 corpus-only cross-links
• GuidelineConcordance keeps the guideline body's website link: a small external-link glyph beside each stance chip in the row's cell (sibling anchor, Tab-reachable, tooltip "Open the body's website"), with the test asserting one per body cell and no nested anchors
• Wave 2 corpus-only joins applied: 3,056 links from fields the records already carry, links still to make down from 12,295 to 10,335 Trial to technology 2,743 links on 1,739 trials (modality technologies of the studied agents); technology to trial 211 links, 26 technologies gaining their first trial; idea to trial 133 links on 30 ideas and idea to key paper 46 links on 32 ideas, only where the idea's text names the drug or target in the idea's cancers; drug to key paper 34 links on 25 drugs; drug to trial 35 links where the title names the drug; target drugs arrays 47 links on 21 targets; company to drug 6 links on 5 companies. No EU approval rows written: 59 of the 60 snapshot candidates already had rows and the last is a withdrawn application with no date. No target-to-drug mechanism links: both measured hits were substrings.
• Corpus-only joins script: scripts/link-joins.ts plans and applies wave 2 links from fields the records already carry, with every rule stated; content-gaps measures corrected Ten rules, dry run by default and --apply to write through the editor in scripts/orphan-links.ts (append only, never removes a link, sorted and de-duplicated): trial to technology through the studied agents' modality technologies; idea to trial and idea to key paper only where the idea's own text names the drug or target, in the idea's cancers, for actors that trials test, capped at five and three; company to drug through sponsored trials where the drug has no maker; drug to key paper and drug to trial through titles that name the drug; technology to trial; target drugs arrays mirroring drug.targets; target to drug by whole-word mechanism mention; EU approval rows from the EMA register snapshot with the file's A, C, V and W helpers, product name and snapshot date checked.
• Header filters for the toolbar facets that had no column: CountryRanking gains a Region column, ScorecardTable Type and Country columns (chips that select the value) and a Regions header filter, PipelineFunnel filters the crowding table by approved product and population completeness, ResearchRanking by parent university and five-year coverage; docs/TABLES.md carries every table's state
• Header filters on the client tables that keep their own controls: TrialFinder, TrialFinderGeo, AssistanceBrowser, GuidelineConcordance, ManufacturingMap, InteractionChecker, AutoPulse, SurvivorshipPlan, MarketEstimator, TumorBoard, Navigator and PlanRankings filter from their column headers through useHeaderFilters and FilterHead, sharing state with the existing facet controls where one exists; CaregiverPanel skipped (eight rows, no categorical column)
• Trial key papers fetcher: Europe PMC by registry id, one primary and one follow-up per trial, links merged from trial-key-papers-wave1.ts
• Header filters on every server-rendered table: StaticTable draws plain rows from the page (links, chips, bars, avatars as small objects) so deals, evidence, funding, universities, payloads, costs, status, freshness, eval questions, survival, cited papers, papers pulse, startup investors, audit, review, China trials and deals, India sites, pathway matrix and dossier trials all filter from their column headers with a N of M count and a Clear pill; docs/TABLES.md records every table's state and the skips under ten rows
• Trials in plain words keeps every trial's heading, status, summary and page link in the HTML and fetches the explanation on demand: one file per cancer section, prefetched as the section scrolls near and opened with an Open explanation pill
• LAUNCH: owner asks of 22 Sept 2026 and what shipped in chain 96
• Column headers filter every browser table, tumour testing redesign with laboratory logos, Edge feed, idea and corpus rankings, explore pagination, content roadmap, PR 54 and 55 rows, signed-in bridge
• Cancer map: a layered DAG of every cancer type at /cancers/map/, with organ systems, histology groupings, cancers and subtypes, badges switchable between trials, products, approvals and ideas, an accessible outline, graph.json for agents, and links from the cancer browser and every family strip
• Rankings hub: fourteen league tables computed from corpus counts and dates, one page each with basis, coverage and a JSON companion
• Content roadmap: pages still to write and cross-connections still to make, measured by scripts/content-gaps.ts; waves 143 to 149 on the roadmap
• Edge: one feed of the freshest signals in cancer, new papers, trial results, approvals, withdrawals, law, bot proposals and weekly issues, ranked by recency with a modest importance weight, with Atom and JSON feeds at /edge/feed.xml and /edge/feed.json
• Tumour testing: By scope and By sample become card grids with company logos, and the table filters by sample, scope, laboratory and status from the URL query
• Idea rankings: six views of the same ideas at /ideas/rankings/, every score from existing fields with its formula shown
• Explore, For me and the navigator page their sections: the first 30 rows of each kind ship in the HTML and the rest load per section from static JSON as the reader scrolls or presses Show 30 more; For me and the navigator fetch one file per cancer instead of carrying all 328
• Filter from the column header: every filterable column in the results tables is a button with a funnel glyph that opens a popover of value pills (counts, search past eight, Clear, Close; bottom sheet on phones), wired to the same setter and URL as the toolbar facets; sortable headers keep sort on the label and put the filter on a sibling glyph
• Four EMA application withdrawals recorded as withdrawn rows, melflufen US withdrawal row restored, FDA Tazverik communication dated 11 May 2026
• Bridge to the signed-in site: header greets a signed-in reader, Watch keeps pages on the account with a sign-up offer, a star on every list row, For me can be saved to the account; consent copy
• chore: weekly roadmap registry check (public/roadmap-watch.json)
• Upkeep runs on its own: weekly logos, structures, trial links and acronyms, monthly identifiers, and the bots' refreshes now deploy without a manual ship
• Weekly maintenance and monthly identifier workflows; bot pushes now deploy through the Vercel ignore test
• chore: refresh literature snapshot from Europe PMC (#56)
• Add FDA safety communication link and SPM rate to epithelioid sarcoma record

21 September 2026

• Trial phases read as plain labels everywhere: Platform trial and Observational study instead of Phase platform, with old filter links still working
• Phase facet test asserts the string key and the shared normaliser
• Facet normaliser is a string key resolved in the client, not a function crossing the server boundary
• Trial phases read "Observational study" and "Platform trial" instead of "Phase observational", from one phaseLabel helper
• A JAMA Oncology 2026 comparison of tissue-free and tumour-informed residual-disease blood tests joins the ctDNA roadmap's sources
• c-TRAK TN tissue-free residual-disease assay paper added and linked from the tumour-informed assay term
• Trial pages say what their enrolment number counts, randomised, analysed or enrolled, with the registry figure explained on hover
• Four new publications found by the registry check are in, including the eight-year ADAURA survival update with its figures, and the check no longer flags screening studies
• Registry check: a non-drug interventional study matches the corpus's observational phase, clearing the PATHFINDER 2 flag
• Four publications flagged by the roadmap registry check recorded as key papers: the ADAURA eight-year overall survival update (J Thorac Oncol 2026) linked from the trial and the EGFR-mutant and resectable NSCLC pages with its stage II-IIIA and IB-IIIA survival figures added to the trial's outcomes, and the FLAURA2 long-term safety analysis (Lung Cancer 2026), the KRYSTAL-12 plain language summary (Future Oncol 2026) and the TROPION-Breast01 China cohort (ESMO Open 2026) linked from their trials only; citation counts fetched from Europe PMC
• Trial enrolment figures now say what they count: 25 landmark trials explain why their randomised or analysed numbers differ from the registry, and five stale figures are corrected from their sources
• Enrolment gaps on 30 roadmap trials reviewed one by one against ClinicalTrials.gov and the primary papers: 25 keep the paper's randomised, analysed, treated or registered count and now say so in a new enrolledBasis and enrolledNote, 5 stale or unsourced figures corrected from the registry or the paper; the roadmap watch reports explained gaps separately and its offline mode re-compares against the committed registry records
• MIMic-01 trial status corrected to active against ClinicalTrials.gov, the first fix from the new roadmap registry check
• Every roadmap now carries a registry check: a weekly script compares each referenced trial with ClinicalTrials.gov and Europe PMC, flags contradictions and overdue readouts, and shows the check date under What to watch
• MIMic-01 status and PATHFINDER 2 design corrected against ClinicalTrials.gov, as flagged by the roadmap watch
• Roadmap watch for every roadmap: scripts/roadmap-watch.ts compares each roadmap's referenced trials with ClinicalTrials.gov (status, phase, enrolment within 5 percent, moved completion dates), lists Europe PMC papers since asOf, flags watch items past their expected date, writes public/roadmap-watch.json for a weekly Thursday workflow and a "Registry check" line on roadmap pages
• A roadmap and timeline for circulating tumour DNA tests, from the 1948 discovery of cell-free DNA to the residual-disease and multi-cancer detection trials reading out next, with a watch list and a script that checks it against the registries
• PATHFINDER 2 status and ECLIPSE enrolment corrected against ClinicalTrials.gov, as flagged by the ctDNA watch script
• ctDNA tests roadmap: ten eras from the 1948 sighting of DNA in plasma to the 2029 randomised readouts, a "What to watch" list with registry and meeting dates, six trials the corpus lacked (TRACC, MERMAID-1 and -2, BESPOKE, CIRCULATE-US, NCI Vanguard), a tumour-informed versus tumour-naive term, and npm run ctdna:watch to keep it current
• Twenty-six landmark papers for the pancreatic cancer pages, from PRODIGE 24 and PREOPANC to the KRAS inhibitor cohorts and the IPMN guidelines, each checked against Europe PMC
• Every pancreatic subtype page now rests on two to four landmark papers: 26 new records (PRODIGE 24, ESPAC-4, PREOPANC, ESPAC5, Alliance A021501, LAP07, CodeBreaK 100 and KRYSTAL-1 pancreatic cohorts, POLO and its survival analysis, the BRCA/PALB2 platinum trial, KEYNOTE-158, the NRG1 fusion discovery and zenocutuzumab papers, Fukuoka, Kyoto and European cyst guidelines, and the acinar cell carcinoma and pancreatoblastoma reference series), each DOI checked on Europe PMC and doi.org, with Le 2017 reused for the MSI-high page and citation counts fetched
• Seventy-four registry trials linked to the newest pancreatic and nasopharyngeal cancer pages, with a guard that keeps neuroendocrine trials off the pancreatic adenocarcinoma pages
• Registry trials now reach the 50 subtype pages added this week: 74 trials linked to the pancreatic disease-state and driver pages and to recurrent or metastatic nasopharyngeal carcinoma, with a neuroendocrine guard so GEP-NET trials stay off the ductal adenocarcinoma pages
• Pages load about three times less script: the header no longer drags in the molecule viewer, every language dictionary and the search engine, so slow connections reach a working page far sooner
• Bundle split follow-ups: stars JSON out of the guarded data folder, layout test reads dictionaries from the full set, row 140 done
• Seven European approvals corrected to withdrawn against the EMA register, the PARTIQoL proton trial gains its reported result, and the KEYNOTE-A18 papers are split into their two Lancet reports
• Client JavaScript on every page cut from 3.0 MB to 1.0 MB: the layout no longer pulls the animated schematics, the structure index, all eight chrome dictionaries, zod or MiniSearch
• Paganetti 2014 RBE reference points at the real paper (Physics in Medicine and Biology)
• Visit counting now asks first: Google Analytics loads only after you allow it, the browser's privacy signal is honoured, and the choice can be changed in the footer
• Data follow-ups: seven withdrawn EU authorisations, PARTIQoL readout, KEYNOTE-A18 split into its two Lancet papers
• Roadmap row 141 marked done for onco.cc
• Google Analytics now loads only after the visitor presses Allow: a consent bar at the foot of the viewport, the choice kept in the browser, Global Privacy Control read as No thanks, an Analytics choice control in the footer and on /privacy/, strings in nine languages
• GitHub star count is baked in at build time, so the top-bar badge is full from the first paint and never flickers
• Launch doc: two dated account notes marked as removed
• Launch doc: old account instructions cut to one-line history notes pointing at me.onco.cc
• Sign in/up opens in the same tab: links to me.onco.cc are treated as part of the site rather than external
• Sign in/up in the top bar now opens the WorkOS sign-in directly, with no stop on a landing page
• Sign in/up on onco.cc now hands you to me.onco.cc, the signed-in side; onco.cc itself keeps no account data, in the browser or the cloud
• Sign in/up now links to me.onco.cc; onco.cc holds no account data
• Sign in/up now goes straight to the WorkOS sign-in page instead of stopping at the newsletter page
• Greptile review config encoding the corpus rules for community pull requests
• Your role, cancer choice and preferences now save to your OnCo account so they follow you across devices, and the outlook block is held back pending a wording review
• Merge cloud profile sync; outlook block held pending wording review
• Community fixes: alphagenome model corrected, epithelioid sarcoma record completed with verified facts, RMC-9805 pancreatic pairing and ubamatamab mechanism corrected, plain-English outlook for ten cancers behind the survival disclosure, and 15 hidden FDA events restored
• Cloud profiles stage one: Supabase tables under the WorkOS token, sync, delete, privacy text
• PR round 2: alphagenome weights gated, FDA regulatory events surfaced, epithelioid sarcoma withdrawal history
• Issues 43, 45, 46, 47: prognosis outlook on ten cancer pages, ubamatamab as MUC16xCD3 bispecific with corrected citation, g12d-plus-pan-ras phase and links from the registry and company release
• Roadmap rows 141 and 142: GDPR compliance for accounts and HIPAA readiness
• chore: render this week's issue to public/newsletter/ (#52)
• chore: refresh public/votes.json from Discussions reactions (#51)
• chore: weekly fact check, audit, and provenance refresh (#50)
• chore: refresh trial counts from ClinicalTrials.gov (#49)

20 September 2026

• Fix duplicate entry for 'alphagenome' in model registry

18 September 2026

• Health data compliance analysis: HIPAA does not apply today, GDPR and state health-data laws do; stay browser-only and prepare encrypted sync
• Top bar sign-in works even before the page script loads: it is now a link that upgrades to the direct WorkOS start
• Top bar button shortened to Sign in/up in every language, and cached pages refreshed so everyone sees the new header
• After sign-in a welcome page asks who you are and which cancer, the top bar shows you are signed in, four community pull requests and four issue fixes land, and secret scanning guards every commit
• Fix issues 42, 39, 37, 36: chip ellipsis and titles, daraxonrasib approval, clickable tooltips, urothelial basics
• Welcome step as its own page: sign-in leaves /signup/ for /welcome/ or the return path, signed-in header pill, preferences kept with the account profile
• Secret scanning: gitleaks on every push and pull request, a local pre-commit hook, and an allowlist for public ids and data prose
• Sign up or log in button in the top bar, a medical disclaimer in the footer, and new Terms of Use and Privacy Policy pages
• Trastuzumab carries the June 2026 FDA approval of palbociclib with trastuzumab as HER2-positive maintenance, the one bot proposal from PR 35 not already in the corpus
• Pages are far lighter: the shared header, footer and cancer list are no longer copied into every page's data, and the sign-in id ships with the code
• Terms of use and privacy policy pages, medical disclaimer in the footer, copyright to OnCo
• Header: a Sign up or log in pill when signed out; region, language and search controls sit on the centre line
• Stop serialising the cancer list and the shared chrome into every page's payload
• Sign-in is live: first-time visitors choose whether they are a patient, caregiver, researcher or clinician and can set a cancer, kept only in their browser
• Each ship now builds once: Git-triggered Vercel builds are skipped and only the verified deploy runs
• First sign-in welcome step: who you are, optional cancer, browser-only profile per WorkOS user
• Vercel: skip Git-triggered builds so each ship builds once, from the verified CLI deploy
• add: Strand AI (W26) to the YC oncology list
• Owner list: proposal for six technology sections to lift the section kind past the size gauge
• Every reference DOI on the site checked against Europe PMC: four wrong links corrected, and a checker script added for future waves
• DOI checker for citation links: four wrong DOIs corrected and two rare-cancer key papers added
• Sixty-six landmark and classification papers for the rare-cancer pages, each checked against Europe PMC, plus two corrected reference links
• Two rare-cancer reference links corrected to the verified POD1UM-303 Lancet 2025 and MEK inhibition Nature 2019 papers
• Key papers for the 39 rare-cancer subtype pages: 66 verified landmark trials, classifications and guidelines from ANCHOR and ACT II to FIRM-ACT, PATHFINDER, HD21 and FIRSTMAPPP, linked through keyPapers
• Fifth Chinese quality pass over the new rare and pancreatic cancer pages: 14 corrections
• Four pancreatic Chinese TL;DR fixes synced from the wave export into the dictionary
• Chinese TL;DR pass 5: fix 14 rare and pancreatic subtype entries
• Eleven pancreatic cancer pages by stage and driver, from resectable and borderline disease to KRAS G12C, BRCA-mutant and MSI-high tumours, each linked from its trials and drugs
• Add 11 pancreatic cancer subtype pages: stage, driver and histology
• Thirty-nine subtype pages for rarer cancers, from anal and vulvar cancer by HPV status to Hodgkin lymphoma by stage and mastocytosis, plus an IndexNow script for changed pages
• Add 39 subtype pages for the rare cancer families that had none: anal, penile, vulvar, vaginal, thymic epithelial, adrenocortical, appendiceal, small bowel, cancer of unknown primary, histiocytoses, Langerhans cell histiocytosis, systemic mastocytosis, Hodgkin lymphoma, gestational trophoblastic neoplasia, pheochromocytoma and paraganglioma, nasopharyngeal and sinonasal cancers, with organ-drawing mappings, reverse links from 76 existing trial and drug records, Chinese TL;DRs registered for fold-zh and parent phrases for the cancer-families test
• IndexNow script: submit record pages changed since a date, batched, with a dry run
• Target pages with a mutation hotspot map no longer throw a hydration error: chart tooltips are written as single strings, and the page test now checks markup structure
• Fix hydration error 418 on target pages: multi-child SVG titles rendered empty by the server
• Fourth Chinese quality pass: 22 corrections across researcher, institute, trial and company summaries, including arsenic trioxide and enzalutamide names
• Chinese quality pass 4 over 243 folded TL;DRs
• Thirty-five key papers gain their DOI and PubMed ids, so every key paper but one now shows its citation count
• Add DOIs and PMIDs to 35 key papers that had neither
• Drug cards no longer nest a link inside a link, and a permanent test renders every page to forbid nested anchors site-wide
• Gene targets carry HGNC, Ensembl, UniProt and NCBI Gene identifiers, and key papers show how often they have been cited
• Guard every page against nested anchors; unnest the product card
• Roadmap rows 22 and 97 marked done
• TRACERx 2023 paper DOI corrected (ends -5, verified at doi.org)
• Fill HGNC, Ensembl, UniProt and NCBI Gene ids on single-gene targets
• Citation counts for key papers from Europe PMC (improvement 97)
• Dependency map hydration error fixed: vendor tiles no longer nest a link inside a link, with a test that forbids nested anchors
• Fix React 418 on the dependency map: no anchor inside an anchor
• Health pass: 369 paper links for 270 researchers, PubChem structures for 43 drugs, six glossary Wikipedia links, refreshed literature snapshots for all 2,093 tracked records, provenance for every record, and 28 logos
• Six more glossary terms gain Wikipedia links; 74 recorded as having no article
• Molecules for 43 more small-molecule and peptide drugs via PubChem name lookup
• Link 369 key papers to 270 people from existing corpus links, people-papers gauge 73.8% to 79.4%
• Trial acronym aliases completed for every registry trial that has one (115 more), so searches by study name find them all
• Add 115 registry trial acronyms from the CT.gov acronym field
• Eighty more drug companies gain Wikipedia and Wikidata links
• Add Wikidata ids and Wikipedia links for 80 more companies
• Sixty-three more institutes and 48 drug companies gain Wikipedia and Wikidata links for readers and machines alike
• Add Wikidata ids and Wikipedia links for 48 companies
• Add Wikipedia links and Wikidata ids for 67 more institutions from the Wikidata API
• Sixty-six hospitals and institutes gain Wikipedia and Wikidata links, which also feed the machine-readable sameAs data for agents
• Add Wikipedia links and Wikidata ids for 66 institutions from the Wikidata API
• Final researcher identifiers pass: 64 ORCID ids and 60 Wikipedia links, taking 373 people to ORCID and 181 to Wikipedia, plus STRASS trial results
• Sixty-four more researchers carry ORCID ids and 60 gain Wikipedia links in the fourth identifier pass
• Seventy-seven landmark trials now show their results as structured outcomes transcribed from the published abstracts, and 146 more researchers carry ORCID ids
• Add structured outcomes to STRASS from its Lancet Oncology 2020 abstract
• Add identifiers pass three: 146 ORCIDs and 7 Wikipedia links for people
• Add structured outcomes to 77 subtype-wave landmark trials from Europe PMC abstracts
• A further 121 researchers carry ORCID ids and 18 gain Wikipedia links, each checked against their ORCID employment record
• Add 121 ORCIDs and 18 Wikipedia links in a second identifier pass
• Subtype cancer pages now link their sibling subtypes (667 links) and 188 researchers carry ORCID ids with 22 Wikipedia links
• Add 188 ORCIDs and 22 Wikipedia links to people records
• Forty-one hospitals and institutes added and 78 researchers linked to where they work
• Link cancer subtypes to their siblings and name parents in tldrs
• Fill institutions for 78 trial investigators and add 41 institution records
• Five duplicate person pages merged into one each with redirects, and a duplicate finder for people added
• Merge five duplicate person records and add a duplicate-person finder
• Update weights and URLs for alphagenome model
• A hundred and eleven trial leads and first authors get person pages, linked from the landmark trials and key papers they led, plus 160 more trial acronyms
• A further 160 trial acronyms added as aliases to registry trial records
• Add 111 trial investigators and lead authors named by the subtype trials and key papers
• Trial acronyms (MARIPOSA-2, SPARTAN, CheckMate 9DX and 700 more) added as aliases to registry trial records so searches by the name patients hear find them
• Trial acronyms for registry records: scripts/trial-acronyms.ts adds 711 aka aliases
• Seventeen more landmark trials named on subtype pages (ATTRACTION-2, MAIA, RESONATE-2, NeoSphere and others) recorded with confirmed registry ids, and acronyms added to three registry records
• Third Chinese quality pass: 19 corrections to new drug, company and trial summaries, including wrong drug names and Korean text left in Chinese sentences
• Confirm registry ids for 24 subtype-page trials: 17 new records, 3 acronyms on registry records, 4 links filled
• Third Chinese quality pass: 19 of 236 TL;DRs from the biomarker, drug, maker and trial waves corrected, and the 73 trial TL;DRs folded into zh.ts
• Seventy-three trials named on subtype pages now have their own records with confirmed registry ids, plus logos for the new drug makers
• Add 73 trials named by the cancer subtype pages (trials-subtypes-wave)

17 September 2026

• Seventy-four drug makers added and 210 drug-to-company links filled, so pipeline drugs now show who is developing them
• Link 193 company-less drugs to their makers and add 74 maker records
• Key papers for 147 subtype pages: 364 pivotal trial, classification and guideline papers, each checked against Europe PMC
• Add key papers to 147 cancer subtype pages
• Thirty-nine drugs named by the new subtype pages now have their own records, from hormone therapies for pituitary tumours to antibody-drug conjugates in trials
• Add drugs named by the cancer subtype pages (drugs-subtypes-wave)
• Fifty biomarker glossary terms named by the new subtype pages (molecular groups, risk scores, resistance mutations), each linked from the pages that use it
• Add 50 biomarker terms named by the cancer subtype pages
• Registry trials linked to 109 subtype pages by driver and stage (1,018 links), with a re-runnable matcher script
• Link 1,018 registry trials to cancer subtype pages by driver and disease-state tokens
• Plain-English sentences for the 170 new subtype pages and trials, so every record carries one again
• Add simple sentences for the subtype-wave cancer pages and trials (part-s)
• Seven duplicate trial records folded into their named trials with redirects; second Chinese quality pass corrects 64 TL;DRs from tonight's new pages, including one inverted meaning, and adopts NMPA drug names
• Review Chinese TL;DRs of the subtype waves: standard drug names, terms and alignment with the English
• Merge duplicate trial records onto their named keepers
• Head and neck cancers by HPV status and site: HPV-positive and HPV-negative disease, recurrent or metastatic squamous cell carcinoma, hypopharyngeal cancer, three salivary gland histologies and three oral cavity subsites, tied to KEYNOTE-048, RTOG 1016, De-ESCALaTE, ECOG 3311, PATHOS and NIVOPOSTOP
• Head and neck cancers by HPV status and site: HPV-positive oropharyngeal, HPV-negative squamous, recurrent or metastatic state page, hypopharyngeal, adenoid cystic, salivary duct, mucoepidermoid, oral tongue, gingivobuccal (India) and lip, with NIVOPOSTOP, ECOG-ACRIN E3311, PATHOS and CAPTAIN-1st
• Neuroendocrine neoplasms by site and grade: small intestinal, pancreatic and lung neuroendocrine tumours, grade 3 NET and extrapulmonary neuroendocrine carcinoma, tied to NETTER-1 and 2, CLARINET, PROMID, RADIANT and CABINET; NETTER-2 refiled under neuroendocrine cancer
• Add neuroendocrine subtypes by site and grade
• Childhood cancers by type and risk: acute lymphoblastic leukaemia by risk group, Philadelphia status, infancy and relapse, paediatric AML, neuroblastoma by risk group, medulloblastoma by molecular group, and a page for adolescents and young adults
• Add paediatric cancers by type and risk group
• Sarcomas and bone tumours by type: chondrosarcoma, angiosarcoma, undifferentiated pleomorphic sarcoma, myxofibrosarcoma, alveolar soft part sarcoma, PEComa, epithelioid haemangioendothelioma, nerve sheath tumours, retroperitoneal and extremity soft tissue sarcoma, with STRASS, EORTC 62012, AMPECT and ANGIOTAX
• Add sarcoma and bone tumour subtype pages with landmark trials
• Cancer pages by state for the common cancers: gynaecological cancers by molecular class and stage, skin cancers by state, lung cancer by driver and stage, breast cancer by state, 44 pages tied to their landmark trials and drugs; the My cancer chip leaves the header until accounts are on
• Breast cancer by state: high-risk early and post-CDK4/6 HR-positive, HER2-low, early HER2-positive and HER2-positive brain metastases, early and metastatic triple-negative, inflammatory breast cancer, Paget disease of the nipple and phyllodes tumour, each linked from the trials and drugs that treat them
• Lung cancer by driver and stage: 14 subtype pages and 10 landmark trials
• Add skin cancers by state: melanoma by mutation, stage and site, advanced cSCC, advanced BCC, DFSP
• Gynaecological cancers by molecular class and stage: endometrial cancer by POLE, MMR, p53 and NSMP class, advanced or recurrent endometrial cancer, uterine carcinosarcoma, cervical cancer by stage (early and fertility-sparing, locally advanced, recurrent or metastatic) and ovarian cancer by platinum sensitivity, each tied to its landmark trials and drugs
• Blood cancers by state: acute myeloid leukaemia by driver and fitness, acute promyelocytic leukaemia, myeloma from smouldering to relapsed, MDS by risk, CLL by line and Richter transformation, CML by phase and primary myelofibrosis, nineteen pages tied to 111 drugs and trials
• Blood cancers by driver, state and phase: nineteen subtype pages
• Upper gastrointestinal and liver cancers by state: gastric cancer by HER2, CLDN18.2, PD-L1 and MSI status and early gastric cancer, hepatocellular carcinoma by BCLC stage, intrahepatic and extrahepatic cholangiocarcinoma, and GIST by mutation, each tied to its landmark trials and drugs
• Colorectal cancer by state: rectal, MSI-high, BRAF V600E, HER2-amplified, KRAS G12C and early-onset pages with OPRA, PROSPECT, BEACON and other landmark trials; primary mediastinal B-cell lymphoma
• Add upper GI and liver subtypes: gastric by biomarker, early gastric, HCC by BCLC stage, cholangiocarcinoma by site, GIST by mutation
• Brain tumour subtypes: IDH-mutant astrocytoma, oligodendroglioma, meningioma, brain metastases, paediatric high-grade glioma, spinal cord tumours, vestibular schwannoma and CNS germ cell tumours, each linked from the drugs and trials that treat them
• Add colorectal subtype pages and primary mediastinal B-cell lymphoma
• Add brain and spinal tumour subtype pages
• Plain-English sentences for every remaining record: 1,167 early-phase trials, 138 people, 65 companies, 8 institutions and 3 ideas, so all 11,306 records now carry one
• Simple layer: part-r, fourth and final wave to full coverage
• Websites for 15 organisations and nine more logos, with archived addresses replaced by live sites; roadmap row 139 records the open dependency map hydration mismatch
• Add official websites to 15 sponsor records and fetch nine logos
• Roadmap row 139: dependency map hydration mismatch
• 969 more plain-English sentences for early-phase trials, key papers and collections, taking the simple layer to 9,930 records; one non-cancer registry record removed
• Simple layer wave 3: part-q with 969 sentences (trials, papers, collections)
• Orphan trials given content: ten drug matches, 17 sponsor aliases, 36 sponsor companies and 130 trials tied to their makers, two non-cancer records removed; backlinks round: 303 weakly linked records gain true relations from their own text; orphans down to 214
• After the trial removals: no dangling references, floor re-run, sponsor wave three Chinese folded
• Backlink fill: text-supported relations for 303 weakly connected records
• 1,024 more plain-English sentences: 279 companies, 104 institutions and 640 phase 3 trials, taking the simple layer to 8,961 records
• Orphan trials: link sponsor-only trial records to drugs, companies, institutions and cancers; 36 new sponsor companies (wave 3); remove two myasthenia gravis records
• Papers for 36 more researchers matched by institution; trial snapshots refreshed for all 1,024 products with 136 registry changes detected; paper snapshots refreshed for all 1,924 records
• Add simple part-p: 1024 plain sentences for companies, institutions, the theories pathway and phase 3 trials
• One papers block per person after the two paper waves met
• People papers: 36 more research-trained people via OpenAlex and Europe PMC
• docs(corrections): name the reporting clinician on the two glioblastoma rows
• Company content and sources for 33 makers, 26 sponsor corrections and papers for twelve researchers; 7,065 reverse links take orphaned trials from 2,467 to 279; Wikipedia links for six terms; structured outcomes with sources for 26 trials; 153 new logos; the schematic gauge counts aliased drawings
• Add structured outcomes to 26 reported trials flagged by the trial-outcomes gauge
• Schematic gauge counts aliased drawings as specific
• Add Wikipedia links to six terms lacking one
• Fold a doubled relation line after the sponsor merge
• Reverse links for orphaned trials: their drugs, companies (or sponsor via alias), technologies and cancers now list them back; orphan floor 2572 to 384
• Link orphan sponsor companies, repoint AAA trials to Novartis, add Europe PMC papers to twelve people
• 661 plain-English sentences: every cancer, 201 drugs, 179 technologies, 120 terms, 50 phase 3 trials and 63 targets; 451 reverse links give 126 orphaned drugs and 105 companies their first inbound link, with three maker corrections
• Reverse links for orphaned drugs and companies: their companies, trials, cancers, technologies and institutions now list them back; orphan floor 2827 to 2572
• Simple layer: part-o with 661 sentences, layer test
• Machine translation pipeline for record summaries with hash-checked cache and a report link; 157 Chinese summaries live (every cancer page and nine glossary terms) and two Spanish; model review panel populated for twenty cancers by Fable, each verdict tied to the record's own sources
• Every record has its own RDF Turtle file at /api/v1/rdf/<id>.ttl, linked from the page head, the hidden machine links, JSON-LD, the OpenAPI document and llms.txt
• Machine-translated summaries: cached per record and language under public/i18n/summaries, shown only while the English hash matches, marked as machine translated with a report link and an English toggle; prioritised batch script with a fetch call to the Messages API and a validator for NCT ids, doses, gene symbols and leftover English; hand-written Spanish and Chinese examples for pCR; coverage script counts them
• Per-record RDF: one Turtle file per record at /api/v1/rdf/<id>.ttl, linked from every record page
• For me from a real situation: setting, biomarkers, treatments had, country and trial interest drive what is standard, what your biomarkers change, trials that fit, warnings and questions, all from record fields with sources; dependency map renders identically on server and client
• Theme stays light even when React recovers from a hydration mismatch; the counts grid fills its rows again with twenty kinds
• Fix hydration mismatch on the dependency map
• For me from a real situation (roadmap item 101): setting, biomarkers, treatments had, country and trials drive record-backed sections
• Machines of oncology, second wave: 23 more machine families from boron neutron capture and FLASH beams to hyperthermia, fluorescence surgery, positioning, gating and dosimetry, with 32 vendors; roadmap rows marked done for the day's work
• Decisions you may face: a page per cancer laying out each treatment setting's options, evidence and trade-offs; expert centres as a sortable table of what can be measured, with a near-you marker; your remembered cancer follows you through the header, home, hubs, trials and search; a what-changed page per cancer; red cards for warnings on cancer pages
• Machines wave 2: BNCT accelerators, FLASH, electron, hyperthermia, PDT lasers, LITT, focused ultrasound, ECT, HIPEC, endoscopy, skin and cervical imaging, RT guidance and QA, CSTDs
• Prostate cancer by state: eight subtype pages from low risk to neuroendocrine, linked from 45 trials and drugs; Tregzi added from the FDA notice and the remaining update bot proposals checked against the registry; Hansoh merged into one record with a redirect
• Patient roadmap 102, 104, 106: remembered cancer, per-cancer what changed, red cards
• Decision pages per cancer and outcome-based centre tables (roadmap items 107 and 108)
• Prostate risk-group and disease-state pages, update bot leftovers, Hansoh merge
• Chinese quality pass: 82 TL;DRs corrected (standard NMPA drug names, basket trial term, methotrexate spelling, idioms), account strings in the formal register
• Radiation wave 5: LET, RBE and when protons beat IMRT
• Chinese quality pass on today's TL;DRs and interface strings: 篮式试验 for basket trial, 甲氨蝶呤 typo, 磁共振直线加速器, established drug names in 61 templated trial lines, natural wording in 30 machine-flavoured sentences, 您 throughout the account strings
• First sixty days after diagnosis: a week-by-week guide for every cancer built from its records, and a printable appointment sheet per cancer with questions, words you may hear and treatments you may be offered
• Model review panel: schema, batch script and page for AI model commentary on records, with two example panels; diagnostics wave two: 24 tests from RAD51 foci and CTC counts to pharmacogenomics, cardiac markers and low-resource screening, with five companies; institution networks: Parker Institute members and leaders, CRUK, NCI, UNICANCER and DKTK links, and ten weakly linked centres strengthened
• Add the first 60 days guide and per-cancer appointment sheets
• Mine institution networks (rows 110, 111): PICI, CRUK, NCI, UNICANCER, DKTK members and leaders; strengthen weak institution pages
• Diagnostics wave 2: 24 tests around sequencing, 5 companies (roadmap row 119)
• Animations pause off screen, when the tab is hidden and on slow phones, with a shared budget of six spinning molecules or drawings at once; the record page sidebar stays inside its column and never reaches the footer
• Model panel: AI models comment on records beside the human review path
• Search results page as an index: kind chips, did-you-mean, pages strip, Ask OnCo card, related strip; 29 trial design terms and a design picker page; startup requests ranked by urgency and commerciality with every input cited; 51 law and regulation terms by jurisdiction with a law page; translation hygiene: language tags, no-translate on names and doses, a browser translation offer
• Animation budget for slower phones: one hook (use-animation-budget) pauses molecules, wireframe schematics, the docking scene, the garden sway, the body map, the resistance map and the steppers when off screen or the tab is hidden, draws a still frame under reduced motion, caps thumbnails at six spinning per kind, animates on tap on Save-Data or four-core devices, defers structure fetches until on screen, lazy-loads the gallery 3D viewer; record page sidebar confined to its grid column and not sticky when taller than the viewport; roadmap rows 127 and 138
• Add laws around oncology: 51 regulatory terms and /law/ page (roadmap 137)
• Manufacturing in oncology: seven supply chains, 18 technologies and 14 makers; theories of cancer: a hub map and sixteen theory records with status; platform trials at depth: 21 master protocol trials and a hub term; technology dependency map with 119 edges and a viewer; Chinese TL;DRs folded in for every new record
• Browser translation and language attributes (roadmap row 128, part three): summaries, field values, data tables, bullets and table bodies carry lang="en" while TL;DRs keep their own language; drug, gene, company, trial and person names, aka lines, NCT ids, gene symbols, database ids, molecule labels, formulas and doses carry translate="no" with the notranslate class; TranslateOffer shows one dismissible line at the top of the main column that switches to a carried language or explains the browser's translator in nine languages; wireframe canvas labels repeated as hidden real text; scripts/translation-coverage.ts measures the share of a page in the chosen language per kind
• Rank startup requests by urgency and commerciality (row 112)
• Trial designs at depth: 29 new terms, 40 enriched, design picker page (row 136)
• Search results page as a proper index (roadmap row 123)
• Manufacturing wave: remove the Hitachi remnant
• Platform and adaptive trials at depth (roadmap row 124): master-protocol hub term, 21 platform, basket and umbrella trials, STAMPEDE enriched
• Theories of cancer: hub map and sixteen theory records in the mechanics section
• One Hitachi record (machines wave)
• Machines of oncology: 21 machine technologies, 17 vendors and 66 centres with their proton, carbon, MR-linac, Gamma Knife and other rare machines, with a machines page; Free in oncology page with 62 services; tumour sequencing tests compared across 13 laboratories; useful 404 page with nearest matches; section tabs, aligned header icons and unsqueezed tables; machine-readable links, Wikidata sameAs and short titles for search engines and agents; footer credit; navigation starts at the top
• Technology dependency DAG viewer (roadmap 134)
• Manufacturing wave: seven supply chains, 18 technologies, 14 makers
• Merge fixes: single Strata Oncology record, Chinese dictionary without duplicate keys
• Machines wave: imaging, radiotherapy, particle, ablation, surgical and laboratory hardware with vendors and centres
• Free in oncology and tumour sequencing test pages (roadmap 117, 118)
• SEO and agent surface on record pages: titles within 60 chars, Wikidata sameAs, hidden machine-readable landmark
• Compact section tabs, table column widths and header icon alignment
• Make the 404 page useful: nearest matches, ways in, local miss log
• Tables: one shared stand-in for rows with no picture, so no visual slot is ever an empty gap
• Header star count stays put on reload; counts grid fills its rows; region menu closes after the tap so it cannot fall through to the profile link; drawer close button without a border; pills never break inside a word; page titles capitalised; landing cards show a corner arrow; off-site links open in a new tab; intoDNA STRIDE and Strata Oncology with sequencing links for testing companies; roadmap rows 115 to 121
• Accounts through WorkOS: browser-only sign-in behind the same account functions, recorded in WorkOS, on when the client id is set; README retitled Total information dominance on cancer with an open-repository note and a screenshot gallery; site title to match; pivot modality headers use one classifier so every header links
• Search index is fresh on every visit (service worker network-first for the API, cache rule narrowed to snapshots and images); site pages ranked higher in search; models table shows each maker's logo or an initials tile; orphan detector script and ratchet test; hub page linked from digital twins
• Apply the update bot's proposals (PR 16): 55 EU approval rows from the EMA register and 28 FDA regulatory events; pancreatic cancer wave: four failed phase 3 programmes, nimotuzumab's NOTABLE approval, elraglusib, their sponsors and cross-links; PDAC and Cloudflare rows on the roadmap
• Merge PRs 25, 28 and 29 (CheckMate split, topic cards, WebMCP); mobile region menu stays on screen; one heading per cancer group; bevacizumab NICE note; glioblastoma MGMT and tovorafenib wording; mathematical models hub and site pages in search; long lists load in pages as you scroll
• Add progressive WebMCP tools for public oncology search and lookup
• fix: filter cancer topic cards by relevance
• fix(glioblastoma): split CheckMate 143/498/548, add extent of resection and CCTG CE.6
• Pivot headers all link to pages: modality, front, status and phase; new small molecule and cancer vaccine pages; day-long cache headers for snapshots; Cloudflare caching on the roadmap
• Changelog sync keeps the pending entry through the build
• Changelog kept current from the commit log; linked from the README
• Link 51 new technologies back from their drugs, targets and base drawings
• Point 27 trials, drugs, targets, pathways and terms at their 44 landmark papers
• Link 28 cancer subtype pages from the drugs and trials that treat them
• Refresh provenance for this week's records
• Cancer links for 20 unlinked trials; Chinese version complete for trials
• Watch button confirms each click
• Country pages: cancer site labels link to their OnCo pages in both tables
• Chinese TL;DRs for the last 87 curated trials and drugs
• Mathematical and biophysical models: 22 model technologies, a third type on the models page
• Types strip as pills, navigator visuals, account-style signup, linked grid headers
• chore: refresh pulse, abstract and citation snapshots (#20)
• chore: refresh institution research output from OpenAlex (#17)
• chore: draft change proposals from feed snapshots

16 September 2026

• Add Anocca; biophysics sources as data-source candidates
• Prevalence rows for 20 more targets
• Prevalence rows for 36 targets
• Molecules for 26 more drugs and tracers
• Molecules for 29 more small-molecule drugs
• Pipelines and missing rows for 29 new cancer pages
• Keep the corpus out of the tool pages' JavaScript
• Keep the corpus out of the guideline pages' JavaScript
• Load translation tables on demand; Chinese TL;DRs for 2,678 registry records
• Chinese TL;DRs for the remaining 131 people
• Subtype pages: leiomyosarcoma, liposarcoma, synovial sarcoma, marginal zone lymphoma, cutaneous T-cell lymphoma
• Subtype pages: kidney, testicular, oesophageal, bladder
• Fix accessibility regressions in shared UI components
• Subtype pages: thyroid, head and neck, ovarian
• Chinese TL;DRs for every company, institution, term and idea
• Radiation wave 4 (21 landmark trials) and Chinese TL;DRs for every cancer, target, technology and curated drug
• Parent pages for breast, lung, leukaemia and biliary cancer; radiation drugs; audit link report; Chinese TL;DRs
• check-links: separate blocked from broken in the report (#15)
• Cancer subtype pages (pleural and peritoneal mesothelioma), radiation wave 2, roadmap 112
• Mobile search opens into the box; radiation wave 1
• Add SciRouter, a prevention-first longevity start-up
• Molecules for 31 small-molecule drugs; gumarontinib folded into glumetinib
• Essential thrombocythaemia page, aspirin record, vorolanib
• Polycythaemia vera: dedicated page, seven pivotal trials, eight terms, two ideas
• China deep dive: fifteen trial sponsors as company records
• Offline page only on real failures, China wave 1, favicon set
• Registry trials wave 6 and products wave 7
• Clickable breadcrumbs, pipeline products wave 6, checklist additions
• Header signup, trial wave 5 and Ask weighting
• trials: wave 4 from ClinicalTrials.gov, 309 recruiting or active industry phase 3 oncology trials each linked to a recorded drug and cancer (295 to their sponsor); wave 3 trials gain sponsor links; search lifts the kind named at the start of a question so a 'drug for' query ranks drugs above trials and people
• data: pipeline wave 2 from ClinicalTrials.gov, curated: 27 products missing from the corpus (ten China-approved kinase and checkpoint drugs, Japan's amrubicin and pirarubicin, tabelecleucel, HER2 and GDF-15 antibodies, PSMA and CAIX PET tracers, CD45 radioimmunotherapy and others) with 32 phase 3 trial records and China regional rows; launch checklist gains the agentic and AI accessibility to-do
• cancers: the four thin pages filled from sources (epithelioid sarcoma, extragonadal germ cell tumour, metastatic cancer, NUT carcinoma) with state of the art, history and development pipelines; six trials (TIGER, SABR-COMET, the tazemetostat run-in, three NUT carcinoma trials) and the BET inhibitor ZEN-3694; gaps page counts only developers as missing products
• site: Google Analytics tag with an honest disclosure on the about page; refresh workflows merge their own pull requests once validate and tests pass (the proposals workflow stays a human gate); deploys the FCCT directory and the nine merged weekly snapshots
• chore: draft change proposals from feed snapshots (#1)
• chore: refresh pulse, abstract and citation snapshots (#2)
• chore: refresh trial counts from ClinicalTrials.gov (#3)
• chore: refresh public/votes.json from Discussions reactions (#5)
• chore: weekly citation link check (#7)
• chore: refresh FDA approvals feed (OCE notifications, openFDA drugsfda) (#8)
• chore: refresh preprint tracker from Europe PMC (#9)
• chore: check regional approvals against the EMA register (#10)
• chore: refresh universe lists (NCI drugs and types, NCI centres, OECI, NHS alliances, NLM journals, OpenAlex, KEGG, ChEMBL, ClinicalTrials.gov, FDA OCE) (#11)
• data: the Sijbrandij Foundation and Future of Cancer Care Today as entities with their leads, Sid Sijbrandij linked, and the FCCT directory brought in: 39 suppliers, tools and services and 10 clinics and centres for tissue preservation, tumour profiling, functional drug testing, liquid biopsy, theranostics, vaccines, cell therapy, trial finding and navigation
• owner requests: aldoxorubicin and Gemini Therapeutics; anetumab ravtansine, its two mesothelioma trials, the UCL proton therapy trial and an idea on why ADCs have failed in mesothelioma; cancer pages replace the curated Pipeline list with an In development tab built from the whole corpus; Even One Ventures and fourteen portfolio companies with a weekly-refreshed universe list
• data: forty-six approved cancer drugs missing from the corpus, curated from every ChEMBL molecule with an oncology indication (historic hormonal and cytotoxic agents, photodynamic and diagnostic agents, growth factors, antiemetics, drugs approved in Japan, China, India or Australia); approval pill no longer wraps
• header: profile icon on every page opens an email capture dialog (magic-link sign-in when accounts are on, list endpoint otherwise); weekly-issue signup boxes removed from footer and newsletter page; strings in nine languages

[1.0.0] - 2026-09-16

Public launch. The site is live at https://onco.cc and the repository is the source of record for the corpus.

Added

• Glossary: twenty canonical categories (mapped at load from the labels written across the data files), one regular category grid with an animation for every category, and a specific picture on every term that has one (its target's schematic, its molecule, its technology's schematic or the organ it concerns) in the table and at the top of the term page.
• Email signup: a SignupBox in the footer and on /newsletter/, driven by NEXT_PUBLIC_SIGNUP_ACTION (Buttondown or Listmonk form endpoint); until it is set the box offers the Atom feed and the per-page Watch button.
• Initials placeholder (RowAvatar) for startups, organisations and people without a usable logo or portrait; hotlinked favicons that fail or come back as the generic 16 px globe fall back to it.
• Journal wordmarks for Nature, the New England Journal of Medicine, Science and The Lancet from Wikidata via their Wikipedia titles, shown on the journals index and journal pages.
• Seventeen ChEMBL drug targets (SRC-family kinases, GM-CSF and IL-11 receptors, xanthine oxidase, DNA polymerase alpha, steroid, serotonin, dopamine, opioid and cannabinoid targets) with HGNC, UniProt and ChEMBL links.
• Six NCI-listed supportive-care drugs (eltrombopag, romiplostim, emapalumab, fostamatinib, ravulizumab, propranolol as Hemangeol) from FDA labels; the chlorambucil-prednisone regimen from the ECOG trial; six earlier Cancer Statistics reports as key papers.
• Papers for sixty-one leaders and OECI representatives via Europe PMC author and affiliation matching, each set checked against the person's role.

Changed

• Provenance tags (gap-fill, chembl-gap, ctgov-ingest, nci-list and similar) are hidden from reader-facing pills and excluded from similarity scoring; they remain in the JSON API.
• People pages show one plain-English provenance line under the papers table instead of a fetcher note on every row.
• Table cells wrap long tokens site-wide; the models table's paper column uses short labels with the identifier as hover text.
• Nine dead organisation websites replaced with verified current addresses; four Wikipedia links repointed to existing articles; every Wikipedia and organisation link in the corpus checked.
• NCI drug list credits two-letter regimen acronyms; the completeness matcher no longer drops "AC".

Also in 1.0.0: work unreleased since 0.4.0

#### Added

• Molecule3D: ball-and-stick models of small molecules (CPK colours, shaded spheres, depth-sorted bond cylinders with double and triple bonds, drag to rotate, element legend) and smooth backbone ribbons for proteins (Catmull-Rom through the C-alpha trace, per-chain colours, wider on helices and strands, bound drug in ball-and-stick), with a wireframe toggle. Used on product pages, in every molecule thumbnail, and in a new "Solved structures with a drug bound" panel on target pages.
• PDB snapshots now keep secondary structure from the HELIX and SHEET records (ss); npm run fetch:structures -- --refresh-pdb 5A9U,5DK3 refreshes chosen entries without touching the rest.
• Startups and investors: company records gain optional stage, ycBatch, investors, funding (sourced rounds, amounts only where the source states them) and acquiredBy; companyType: "investor" for venture funds, corporate venture arms, an accelerator and disease foundations, whose portfolios are derived from backlinks.
• 62 Y Combinator companies attacking cancer, drawn from the open YC directory dataset and checked one by one (src/data/companies-yc.ts, snapshot with every hit and its decision in src/data/universe-lists/yc-oncology.json); 187 further venture-backed oncology companies across therapeutics, diagnostics, AI, digital care, radiotherapy hardware, surgery and tools (src/data/companies-startups.ts); 69 investors (src/data/investors.ts).
• /startups/ (stage strip, YC batch chips, filters by stage, modality, cancer, batch, investor and country, most active investors, recently funded) and /investors/; Funding panel on company pages, Portfolio panel on investor pages; stage and investor facets on /companies/; Startups and Investors in the Institutions & people group.

[0.4.0] - 2026-09-08

Added

• Animated wireframe schematics for all 18 fronts (front pages, Fronts index, home) and for all 19 glossary term categories (/terms/ and every term page).
• Molecule thumbnails wherever a product is mentioned: cards in pipelines, connected lists, and product tabs; hover popover with the rotating molecule on inline drug chips.
• Self-hosted organisation logos under public/logos/ (Wikimedia Commons via Wikidata with domain verification, favicon fallback) for companies, institutions, and collections.
• person kind (clinicians and scientists: role, specialisms, profiles, papers) with a People index and a People tab on institution pages.
• Site favicon and Apple touch icon; GitHub link in the header.
• Public-readiness: CI workflow, issue and pull-request templates, Code of Conduct, security policy, consolidated contributing guide.

Changed

• Graph explorer rebuilt as a clean radial view (fronts inner ring, cancers outer ring) with a focus mode that groups neighbours by kind; no physics.
• Header language and theme controls are single-line and compact.
• Footer and About page carry the work-in-progress disclaimer: verify every fact at its primary source; not medical advice.

[0.3.0] - 2026-09-07

Added

• Deep dives for all 31 cancers (standard of care by setting with NCCN/ESMO guideline mapping, state of the art, history, pipeline, open problems), taking the corpus from 875 to 1574 objects.
• Structured trial outcomes on every trial, out-of-100 pictograms, replication notes, and a disclosed evidence score (/evidence/).
• Product depth: dosing, label-sourced toxicity tables, cost and access, dated regulatory timelines (/regulatory/), class-wise toxicity comparison (/toxicity/), and animated mechanism cards.
• Sourced biomarker prevalence by cancer for targets, with a matrix at /prevalence/.
• Browser-only profile with Navigator (line of therapy, cautions, next options), caregiver mode, and a trial finder with country and distance filters.
• Reading-level and language layers: plain and simple TL;DRs, translations into Spanish, Chinese, Portuguese, and Hindi (machine-assisted, unreviewed), and glossary hovers.
• Power tools: multi-item Compare with differences, Pivot tables, Timeline scrubber, Query builder.
• Trust layer: /audit/ (staleness and contradictions), provenance line per record from the commit history, weekly fact checks against openFDA and ClinicalTrials.gov, confidence chips for ideas and speculative roadmap steps, /corrections/ from CORRECTIONS.md, and expert and patient-advocate review tracks with mandatory conflict-of-interest statements.
• Visuals: pathway diagrams that light up per product, Body map, theme toggle with high contrast and skip link, Story mode on roadmaps, interactive molecule viewer with nine drug–target PDB complexes, animated process schematics (ADC internalisation, CAR-T, radioligand, and more).
• Suggest-an-edit on every page with an organisation self-service track; open evaluation benchmark of 100 questions (/eval/).
• Tabbed object pages with the section kept in the URL; grouped navigation (Find, Map, Intelligence, Who, Learn & contribute) with landing pages.
• Research output ranking from OpenAlex on /universities/, trial leadership index at /leadership/, cooperative trial groups as institution records, funding flows at /funding/, gaps and bounties at /gaps/, annual report at /report/2026/, and an MCP server (npm run mcp).

Changed

• Duplicate entity ids across per-cancer files are merged at load time (first record's scalars win, arrays appended).

[0.2.0] - 2026-09-06

Added

• Slowly rotating 3D wireframes on every product page: small molecules from PubChem 3D conformers, the payload for ADCs, and an IgG C-alpha backbone (PDB 1IGT) for antibodies. Structures are self-hosted under public/structures/.
• Organisation logos on company, institution, and collection pages, hotlinked from the organisation's own site.
• Filterable, sortable products browser at /drugs/ with phase/status, modality, ADC payload class, front, target, cancer, and company facets, in card or table view.
• Explore power view at /explore/: pick a cancer type, switch entity kind, and get a ranked list with a disclosed relevance score.

Changed

• "Sections" renamed to "Fronts" (the fronts of the war on cancer); routes moved from /sections/ to /fronts/.
• Hero line changed to "Total information dominance on cancer."
• The visible "as of" stamp was removed from pages; the internal last-checked date remains in the data for maintainers.

[0.1.0] - 2026-09-04

Added

• Initial release: 642 objects across 14 kinds (cancers, fronts, technologies, targets, products, companies, institutions, pathways, terms, trials, pairings, roadmaps, ideas, collections), one page per object with plain-English TL;DR, derived backlinks, and connected-objects panels.
• Triple-negative breast cancer deep dive (the first fully built example): standard of care by setting, state of the art, history timeline, pipeline, open problems.
• Nine roadmaps including ADC generations, TROP2 ADCs, TNBC, radiopharmaceuticals, cell therapy, molecular imaging, early detection, immunotherapy, and KRAS.
• "For me" cancer picker, institution world map with a disclosed ranking, pathway diagrams, glossary, 50 hub ideas, and a static JSON API at /api/v1/.
• Build-time validation of every cross-reference; MIT code licence, CC BY 4.0 data licence.