Placebo

Inputs

Five public sources, each a link in the forecast JSON.

01
Registry record
phase, enrolment, sponsor class, the endpoint's class
02
Open Targets
the drug target's association with Alzheimer's
03
Sponsor record
its posted readouts before tonight, and how many hit
04
Class base rate
the pipeline reviews, PMIDs 25024750 and 42095064
05
PubMed
papers on the drug and Alzheimer's in 24 months

Class priors

Chance that one named endpoint clears alpha 0.05, before any factor.

Endpoint	P2	P2/3	P3
cognitive-functional	18%	16%	12%
neuropsychiatric	32%	30%	30%
biomarker	50%	50%	50%
other	22%	20%	20%

0.4% of Alzheimer's drugs reached approval over 2002-2012 (Cummings 2014). One endpoint is a lower bar than approval, so the priors sit above it.

The bond

2,000 $PLCB per open market, held by the contract.

2,000

$PLCB on each market

≤.25
Brier at or under 0.25
Returned in full by releaseBonds.
>.25
Above 0.25
Half is sent to 0x…dEaD, half returned.

The bond is posted from the first night after $PLCB is set on the contract.

The model placebo-1

log-odds = logit(prior[endpoint][phase])
         + 1.10 × (G − 0.30)
         + 0.30 × clamp(ln(n / 300), −1.5, 1.5)
         + 0.20 × industry
         + 1.20 × ((hits + 1) / (readouts + 4) − 0.25)
         + 0.12 × ln(1 + papers / 20)
p = clamp(1 / (1 + e^−log-odds), 0.03, 0.92), rounded to 0.01

Term	Definition
Class prior	logit(prior[endpoint][phase])
Genetic support	1.10 × (G − 0.30), G = highest Open Targets association of the drug's targets with Alzheimer's
Size	0.30 × clamp(ln(n / 300), −1.5, 1.5), n = enrolment
Sponsor class	0.20 if the lead sponsor is industry
Sponsor record	1.20 × ((hits + 1) / (readouts + 4) − 0.25), the sponsor's posted readouts before tonight
Literature	0.12 × ln(1 + papers / 20), PubMed papers on the drug and Alzheimer's in the 24 months before

No language model is called. The three sentences of reasoning are written from these numbers by a fixed template, so the same inputs always give the same JSON and the same sha256.

On the replay

All 81 →

Mean forecast per decile against the share that hit. The dashed line is perfect calibration.

Brier 0.1895 on 81 readouts, against 0.1859 for always saying the base rate of 25%. The model is close to the base rate and not yet better than it; the live record decides whether its factors add anything.

Tonight

2026-10-13