Scientific literature search that answers with the passage — and tells you when your own data disagrees
Sourced answers from papers, patents and reports in seconds, our own disagreements first.
See one case, screen by screen ↓A translational scientist designing a resistance cohort needs more than a list of hits. She needs the rate, the n, the model system and the passage that says it — from published series, from patent claims, from the company's own translational and pharmacology reports and from signed lab-notebook entries. Today those sit in different places: a literature database, a patent search, a report store, an electronic lab notebook. An answer without its passage cannot go into a study plan or a review, so the same question is worked again by the next person who asks it.
The company's own reports are the hardest to find and often the most decisive. When one drug maker checked 67 published findings in its own labs, 43 were inconsistent with what was published. A disagreement like that between an internal cohort and a published series surfaces only if the right person happens to read both documents — usually after the study plan is written.
Estimated split of one scientist's week of literature work, by hand and with the solution. Hours.
Five agents index, answer, check every row and weigh agreement with our own work; the scientist keeps the answer, the source's owner settles a disagreement.
Dr. Maya Chen is a translational scientist on Programme CVT-416, a covalent KRAS G12C inhibitor. She is designing the resistance cohort for the next study and needs to know what is known about escape through MET amplification. Here is what happens next, screen by screen, in the working solution.
Ask opens on her last question from 5 October — “Does MET copy gain show in plasma before tissue?” — kept in the CVT-416 notebook: 3 claims, 7 rows, 6 sources, 1 of them ours, no disagreement found. The scope sits under the question bar: Programme CVT-416, all 4 source types, 2021–2026, checked against our own work.
“Short answer. Yes. Plasma shows MET gain a median of 46 days before the matched biopsy and finds more resistance overall; a biopsy is still needed to see a change of histology.”
She types “What is known about resistance to KRAS G12C inhibitors through MET amplification?” and the run shows each step: the question scoped to CVT-416 and the sources she may open, 128,412 sources searched with 36 passages read in full, then 5 claims written from 17 evidence rows across 12 sources, 4 of them our own. The claims fill in one by one; then every row is checked against its passage and every claim against our own work.
Every step is shown as it runs, with what it found.
The short answer: MET copy gain is the most common single way a tumour escapes a covalent KRAS G12C inhibitor, plasma sees it before a biopsy does, and adding a MET inhibitor restores suppression in resistant models. Claim 1 carries four rows — the 212-patient plasma series at 10.8% (23 of 212), a review of 11 series, our VLT-302 interim 2 at 10.5% (4 of 38), and a colorectal study that limits it to lung. The citation checker read all 17 rows against their passages: 17 supported, 0 removed.
“1 claim disputed · 11 s · 12 sources, 4 ours · 5 claims, 17 rows · High confidence”
Claim 2 says that when MET gain appears, it is usually the only escape route — the published series found it alone in 12 of 23 (52%), and the resistant lines showed no other new driver. Our VLT-302 interim 2 contradicts it: MET gain alone in 1 of 4, three carried a second alteration. The meter reads 2 support · 1 contradict, the row lands amber, and the agent says why the two may differ.
“Small numbers on our side — four samples — so this is a signal to check, not a verdict.”
Plasma shows MET gain a median of 46 days before the biopsy (31 paired), and finds a mechanism in 61% against 44% by tissue. Adding a MET inhibitor gave 62% xenograft shrinkage at day 21; our RX-0942 study gave 71% growth inhibition against 18% for CVT-416 alone, after a screen combination index of 0.31. Claim 5 is a gap: no report of the combination in people, and patent WO 2025/118432 claims 200–400 mg of the MET inhibitor, overlapping the planned 300 mg arm.
The patent row is labelled “A claim, not a result” — and the claim carries “No result in people”.
“Open at page 18” takes her from the contradicting row to the VLT-302 translational interim report, with the cited sentence highlighted: in three of the four MET cases a second pathway alteration was in the same plasma draw, confirmed on the matched biopsy in two. Beside it, why the passage came back, and who may read it: owned by Programme VLT-302, programme only, authors Lena Ortiz and Sam Patel.
“The citation checker read the row against this passage and found it supported.”
“Compare side by side” puts both passages together: the published plasma series — 23 of 212 samples, and in 12 of those 23 the only new alteration — against our VLT-302 interim 2, where MET gain stood alone in one participant only. Under them, why they may differ: if it holds, a MET inhibitor alone would leave three in four of these patients with a second route open, and the cohort should enter on a panel rather than on MET alone.
Maya sends it to Lena Ortiz, who owns the VLT-302 cohort. It joins the Disagreements list — four in the last 30 days: one new, one open with Dana Okafor on twice-daily dosing, one explained by Sam Patel on assay length, one settled by Lena on the panel threshold. It shows as an amber row on every answer that uses this claim until she settles it.
“Lena Ortiz only — an agent never closes a disagreement.”
She keeps the answer in the CVT-416 notebook as “Resistance through MET copy gain — what we know today”: 5 claims, 17 evidence rows from 12 sources, each opening at its passage, with the disagreement attached until its owner settles it. The notebook also lists the questions the team asked this week, so the next person starts from her rows instead of from a search box.
Overnight the library agent indexes new sources and the watch agent matches them to what Maya asked to be told about — KRAS G12C resistance, MET bypass signalling, plasma-first resistance testing. Each item says why it matched, and each new internal report is read against the claims the team has kept: VLT-302 interim 2 disagrees, RX-0942 version 3 still agrees.
The same answer, seen by the five people who rely on it — what their week looked like, and what it looks like now.
Five agents index, answer, check every row and weigh agreement with our own work; the scientist keeps the answer, the source's owner settles a disagreement.
Reads each new source, splits it into passages, and records target, mechanism, indication, model system and n so evidence rows can be built and scoped, keeping the page and passage position.
Takes a question in plain words, searches every source type, and answers as a short paragraph and numbered claims, each claim a set of evidence rows: source, year, model system, n, effect and stance.
Reads each evidence row back against its passage, including n and the model system, and removes or flags anything the source does not actually say.
Counts, for every claim, the sources that support, contradict or limit it, reads our own reports and signed notebooks against the published ones, gives the most likely reason they differ and sends a disagreement to the owner of our source.
Runs each night against what people asked to be told about, ranks new sources by how close they are to the watcher's programme and recent questions, and says in one line why each matched.
Keeps the answer; the owner settles a disagreement. The agents propose; a named person decides.
Ask about anything in the library, or say what to watch for you. When nothing in the library answers with a passage, the assistant says so rather than writing a claim.
Where does our own work disagree with the literature?
Four times in the last 30 days. The new one is from this morning: our VLT-302 interim 2 found MET gain alone in 1 of 4 samples, where the published plasma series found it alone in 12 of 23. One more is open with Dana Okafor on twice-daily dosing; two are explained or settled.
What changed on my watched topics this week?
Three sources matched KRAS G12C resistance since Monday: a 212-patient plasma series, an updated internal pharmacology report, and a published patent family whose claim 4 overlaps your 300 mg arm. The patent is the one worth ten minutes today.
Watch MET bypass resistance in patents too
Done. I widened the MET bypass signalling watch to patents as well as papers, and it will tell you each morning. I have added it to Settings so the team can see what is watched.
Who else in the team asked about this target?
Four people besides you asked about this target in the last week — Sam Patel twice, Lena Ortiz on the assay, Omar Haddad on the patent position and Dana Okafor on dosing and regrowth. Their questions and the sources they were given are in the notebook.
13 screens from the working solution, on its sample data. Pick one to see it large.
How strict an answer is, how the meter counts and who receives a disagreement are settings, not code. Two rules — every row must carry a passage, and every claim checked against our own work — can be switched off only by the library administrator, as a new version written to the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Every row must carry a passage | On | On · Off |
| Oldest source to use | Published 2021 or later | Published 2021 or later · Published 2016 or later · Any year |
| Answer length | Standard | Short · Standard · Full review |
| Check every claim against our own work | On | On · Off |
| Sources needed to call a claim agreed | 2 sources | 1 source · 2 sources · 3 sources |
| Patent text | A claim, not a result | A claim, not a result · Counts as support |
| Who receives a new disagreement | Owner of our source | Owner of our source · Programme lead · Owner and programme lead |
| Call out small numbers | Fewer than 10 | Fewer than 5 · Fewer than 10 · Fewer than 20 |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample library of 128,412 sources · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (Elsevier / Sense about Science, Trust in Research, 2019: over 4 hours a week finding articles · Prinz et al., Nat Rev Drug Discov 2011: 43 of 67 published findings inconsistent with in-house data). People, programmes, studies, journals and sources named on this page are fictional — characters and sample data in the working solution.
Software that answers a research question from papers, patents and reports with the sources behind each statement. Scientific Literature Search answers as a short paragraph and numbered claims; each claim is a block of evidence rows — source, year, model system, n, effect, stance — and each row opens the exact passage it came from.
Yes. Published papers, patent families, internal research reports and signed lab-notebook entries are searched together. Internal reports and notebooks are answered only to the programme that owns them, and a notebook entry is searchable only once it is signed.
The agreement agent reads every claim against our own reports and notebooks. Each claim carries a meter of how many sources support, contradict or limit it; a contradicting internal row turns amber, with the reason the two may differ, and the disagreement goes to the owner of our source.
Every evidence row must carry a passage, and the citation checker reads each row back against its passage, removing anything the source does not say and recording what it removed. Figures and doses are quoted, never recalculated. When nothing in the library answers with a passage, it says so instead of writing a claim.
Patent families and their claims are searched alongside the literature, and patent text is labelled “A claim, not a result”, so a claimed dose window is never counted as evidence. Patent counsel can be given papers and patents only.
Yes. Each scientist sets watches — a topic and the source types — and every morning the watch agent lists what arrived overnight with the reason each matched. New internal reports are also read against the answers the team has kept.
Yes. The scientist decides what to keep in the programme notebook, and only the owner of our source may explain or settle a disagreement — an agent never decides which side is right and never closes one. Every question and every keep, send and settle is on the audit trail.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your sources, programmes, access rules and a sample of your questions — and it goes live once every quality gate has passed. We will walk you through it on your own library first.
Every rival trial, readout and approval in a sourced brief each Monday
1-min demoPreclinical Study Report AutomationA draft GLP report in hours, every number traced to raw data.
1-min demoPost-Merger R&D IntegrationA signed, cited dossier for each asset before the experts leave.
1-min demoWe'll run Scientific Literature Search on your own questions, papers, patents and internal reports.