Calibration

What a readout is worth

Seven reads of the data at two oncology meetings in mid-2026 -- each written down and dated before the result, then scored against what actually happened. The prices here are the scorecard: proof a method kept its own score.

Before each result I write down what the data would have to show for my read to count as right, and I date it. After the result I score it -- including the rows I got wrong. This is the full card from those two meetings.

The discipline, before the method

A prediction you can quietly revise after the fact is worthless. So before each readout I locked a written, dated call -- what the data would have to show to count as a hit -- into a file I cannot edit afterward. The ASCO calls were locked May 27 and May 29; the EHA calls were frozen June 2, before any company was scored. Then the data printed, and I checked the work against it. The grades below were also run through an independent model of a different family, whose job was to catch me grading myself generously. It did, twice; these are post-correction.

The seven reads, scored

Prices are close-to-close (Yahoo Finance), bracketing each readout’s public-disclosure date.

Kura OncologyKURAziftomenib · frontline AML
Locked, in advance
The high complete-remission rate would hold in the larger cohort.
What the data showed
CRc 96% (NPM1-mutant) / 90% (KMT2A-rearranged) across 99 patients; durability not yet reached in the NPM1-mutant arm.
$9.05 -> $9.54 (Jun 12->15) +5.4%Hit -- science and market agreed
A loose call that landed: the bar was directional, not numeric.
Source: clinical disclosure · price series
Cogent BiosciencesCOGTbezuclastinib · systemic mastocytosis
Locked, in advance
A clean, registration-grade print.
What the data showed
65% response, discontinuations for toxicity rare, NDA filing on deck for June 2026.
$32.62 -> $33.17 (Jun 11->15) +1.7%Science hit, price flat
Good data, already in the price -- the cleanest card moved the least.
Source: clinical disclosure · price series
ImmuneeringIMRXatebimetinib · pancreatic
Locked, in advance
Response above 35%, progression past 9 months, survival mature -- all three; survival immaturity flagged as the risk.
What the data showed
Response just cleared (36%); progression missed (8.3 mo); the OS headline (17.3 mo) rests on still-immature follow-up.
$5.77 -> $4.49 (May 29->Jun 1) −22%Miss -- the primary bar failed
The progression bar failed; the survival-maturity risk I flagged was nonetheless real.
Source: clinical disclosure · price series
Revolution MedicinesRVMDdaraxonrasib · pancreatic
Locked, in advance
Benefit would concentrate in the RAS-G12 subgroup.
What the data showed
Benefit was excellent but not G12-confined -- hazard ratio 0.40 in both the RAS-G12 and the overall ITT population.
$157.47 -> $163.67 -> $151.30 (May 29->Jun 2) +3.9%, then fully reversedMiss -- broader than the thesis
The drug worked more widely than I called; a landmark benefit, wrong shape.
Source: clinical disclosure · price series
Caribou BiosciencesCRBUallogeneic CAR-T · lymphoma
Locked, in advance
A positive durability update.
What the data showed
Headline (82% ORR, 67% CR, 17.1-mo PFS) rested on a 27-patient "optimized" subgroup of the 63 dosed.
$1.79 -> $1.66 (Jun 10->11) −7.3%Miss -- headline narrower than the cohort
The durability headline rested on a subgroup, not the full dosed cohort.
Source: clinical disclosure · price series
CellectisCLLSallogeneic CAR-T · leukemia
Locked, in advance
Deliberately two-sided -- would the full cohort confirm, or was the topline a thin slice?
What the data showed
100% response (7/7, CR/CRi 57%) -- but in the small target population of the 45-patient trial.
$3.18 -> $3.11 (Jun 10->15) −2.2% (flat)No edge, by design
The thin-slice risk I named is exactly what printed -- a correct abstention.
Source: clinical disclosure · price series
Legend BiotechLEGNin-vivo CAR-T · context, not scored
Locked, in advance
Noted it had already run; the result was pre-disclosed.
What the data showed
The late-breaker confirmed the June 2 headline -- 100% ORR (6/6) at dose level 2.
$25.51 -> $36.28 (Jun 1->2) +42%, then −1.1% on the actual printPriced -- rose on the news, not the confirmation
Context, not a scored read: the move came on the announcement, not the data.
Source: clinical disclosure · price series

What survives honest scoring

Read top to bottom, this is neither a winning streak nor a bust. It is a calibration record.

The market does not pay for good data; it pays for surprise. COGT printed the cleanest card on the board and moved 1.7%, because the quality was already expected. KURA moved 5.4% on a result that resolved genuine uncertainty. RVMD posted a landmark survival benefit and round-tripped in a day. The move lives in the gap between what was known and what was shown.
My most rigorous calls were the ones I missed. RVMD and IMRX carried specific, numeric bars -- and both came in misses. The looser calls (KURA, COGT) landed. I say this against myself on purpose: a call with a numeric bar can fail honestly, and mine did; a call without one can always be graded a hit afterward. One rule carries forward -- every bar numeric before the data, or it is not a test.
The read I am proudest of is CLLS -- and it is neither a hit nor a miss. I declined to take a side and instead named the exact risk: that a clean topline can be carried by a handful of patients. A seven-patient target cohort is what printed -- 100% in 7 of 7. A read that refuses to guess, states why, and is right about the why is doing the real work -- but it is an abstention, and I will not dress it up as a scored win.

I show the misses because a scorecard you cannot lose on is not a scorecard. What you get is a dated, auditable read of the published data -- graded in the open against what the stock actually did, wrong rows included.

Reading the numbers honestly

These seven are the complete set I locked for these two meetings -- nothing was dropped to flatter the record. The price moves are raw close-to-close changes, not adjusted for the biotech sector’s move that week; they show that the market reacted, not how much of the reaction was company-specific. A contemporaneous move is correlation, not proof of cause. LEGN is included as context, not as a scored read -- it had no locked call; it simply illustrates the "priced before the news" pattern. Every clinical figure and price links to its primary source.

Method: pre-registration locks dated May 27 / May 29 / June 2, 2026. Clinical results from company disclosures at ASCO and EHA 2026. Prices: Yahoo Finance daily closes, close-to-close across each disclosure date. Independent adversarial review: cross-family model (DeepSeek V4), June 18, 2026.