Congress Intelligence · Case Study
A congress-intelligence hub that grades its own confidence — and ships RED when the evidence isn't there
ASCO is where oncology theses are made and unmade in a weekend. ASCO 2026 carried eight readouts an investor would have to navigate through a fog of sell-side notes. We built a hub where every claim traces to a primary source — and a gate that refuses to print a verdict the data cannot support.
Why Congress Weeks Decide Theses
Once or twice a year, an entire therapeutic field empties its desk drawers in public. At ASCO it is cancer; at EHA, blood cancers; at AHA, the heart. For a few days, clinical results that have been secret for years go on stage at once — and the market reprices in real time. A single pivotal readout can add or erase billions of dollars before the session ends. If you hold the stock, advise the board, or compete in the indication, those few days are the most information-dense — and most dangerous — of your year.
The trouble is how the information arrives: as a firehose. Dozens of abstracts drop at once, every company narrates its own data in the most flattering light, the sell-side notes that follow are written to a house view, and a general-purpose AI will hand you a confident summary it cannot source and sometimes invented outright. The reader who most needs a clear, trustworthy read is the one drowning in spin.
That read is what we build. Fact-gated congress intelligenceis one document that maps a meeting's readouts onto the biology connecting them, traces every claim to a primary source, grades how much weight each claim can bear, and — the part nobody else does — tells you plainly where the evidence runs out. What follows is one worked example: ASCO 2026. The same method is now pointed at EHA 2026.
The Challenge
For ASCO 2026, the interesting structure was hiding in plain sight. Eight separate company readouts were not eight independent bets — they were eight probes of the same biological network: the RAS pathway, the most-mutated oncogene family in human cancer. A win or a failure at one node carries information about all the others.
That is the cross-read a generalist summary flattens and a single-name analyst never sees. So the question we set out to answer was not “what did each company report” — it was “what does the network say, and how much of it can we actually stand behind?”
What We Built
One red thread through eight readouts.
The hub is a single argument told in five chapters — a “red thread” that starts at the RAS mechanism and runs through all eight ASCO 2026 readouts, a table of contents, a branch matrix that shows how the bets relate, and a bibliography of primary sources. It reads like a Nature Reviews piece and behaves like a Bloomberg terminal: every figure is two clicks from the paper, filing, or trial record it came from.
Every claim graded, every grade traced.
No assertion stands on its own authority. Each is graded on a four-tier evidence scale — T1 (primary, independent, accountable) down to T4 — and anchored to the exact source that earns the grade. The grading is not decoration; it is enforced. A structural gate walks the rendered page and fails the build if a single claim is missing its anchor or its tier. On the published build, twenty-two claims cleared with twenty-two verified anchors. Nothing reaches the reader on the strength of “the model said so.”
Design that disappears into the data.
Tufte governs the surface: a warm paper canvas, hairline rules, small multiples, and word-number-image composition with no chartjunk. The Economist sets the register — assumed-intelligent reader, dry wit, the understated implication. The effect is a document that respects the reader's time and never raises its voice to sell.
The Distinctive Part
The gate ships RED by design.
Most intelligence products launder confidence. They round a maybe up to a call, because a confident page sells better than an honest one. This hub does the opposite. A second gate forces every catalyst to carry a falsifiable call — and where the readout has not yet happened, the call is withheld and the catalyst renders RED rather than green. The page would rather show you a blank than a guess. The status is not a bug to be cleared before launch; it is the product telling you, precisely, where the evidence stops.
For a client, that is the whole value. A tool that is green everywhere tells you nothing. A tool that is honest about its own RED is one you can act on — because when it finally turns green, it has earned it.
Why This Exists
You can ask a general-purpose AI about the RAS readouts at ASCO 2026. It will give you a fluent summary. It will not refuse to answer where the trial has not read out, because it has no concept of a falsifiable call it is not yet entitled to make. It will not grade its own confidence against a primary source and fail itself when the anchor is missing, because nothing checks it. Fluent and unaccountable is the default. This hub is the opposite of the default on purpose.
The method is a pipeline, not a prompt: facts are collected from primary sources, verified against the literature, graded on a four-tier evidence scale, and only then composed into prose — with a structural gate that blocks publication if any claim loses its anchor, and a calibration gate that withholds a call the data cannot support. The author reasons over canonical facts; the author never invents one. The same pipeline is repeatable for any congress. It is now pointed at EHA 2026.
The other half is judgment. A system can gather and grade evidence; it cannot decide which silence matters. That read — which RED is the one to watch, which clean network signal is real — comes from 25 years of evaluating drug programs at Novartis NIBR, through Shire's $32B acquisition diligence, and across 500+ compounds. The pipeline keeps the facts honest. The judgment is mine.
See the Live Hub
This case study describes the product. The product itself is live — read the full fact-gated ASCO 2026 RAS hub, follow any claim to its source, and watch the gate hold the line.
The Product
ASCO 2026 RAS · Eight readouts, one biological network →
Five-chapter red-thread narrative · eight catalysts · T1–T4 grading on every claim · full bibliography · calibration gate
Have a congress, a pipeline, or a thesis to map?
I build fact-gated intelligence for biotech investors and operators — congress readouts, competitive landscapes, claims verification, and adversarial review. The deliverable grades its own confidence and traces every claim to a source. No black boxes, and no confidence it has not earned.
Background: 25 years in biotech — drug discovery at Novartis, $32B M&A diligence at Shire, AI platform development as a founder. I read the biology, not just the pitch deck.
chris@bigbio.ai