CORTIST / FORECAST CALIBRATION WORKPAPERS / OCT 2026Where our estimates
come from.
Three internally audited assumption models for AI Agents, AI in Medicine and AI & Energy. None has passed independent review or is an official forecast.
INTERNAL PROBABILITY WORKPAPERS · These 86%, 46% and 54% figures are sensitivity-based working calculations only, not three additional approved forecast probabilities. All three remain under
methodological HOLD.
Editorial responsibility ↗ INTERNAL AUDIT RESULT: HOLD / 0 EXTERNAL REVIEWERS. All three illustrative working models remain not statistically calibrated, not independently approved and not eligible for scoring. The previous 85% and 45% displays were corrected to 86% and 46% after an arithmetic rounding check; earlier versioned workpapers are retained.
Read the full methodology audit and external review blockers ↗ 01 / FIRST-PASS CALIBRATION
Make the math
inspectable.
None of these values is independently verified. Before any official probability is adopted, each model must survive a preregistered definition, reference-class audit and outside reviewer approval.
These are exploratory editorial estimates, not approved Cortist forecasts. Each model shows how its assumptions affect the result. The ranges reflect alternative inputs, not statistical confidence intervals.
A-001 / RESEARCH ESTIMATEAI Agents
Eight-hour technical tasks by 2028
86%
Editorial model / Review pending64-96%Alternative assumptions, not a confidence interval
An estimate based on task capability and the chance that independent tests confirm it.
Explore the calculation ↗M-001 / RESEARCH ESTIMATEAI in Medicine
An AI-designed FDA-approved drug by 2032
46%
Editorial model / Review pending20-72%Alternative assumptions, not a confidence interval
Two open questions: approval of a qualifying medicine and verifiable evidence of AI molecular design.
Explore the calculation ↗E-001 / RESEARCH ESTIMATEAI & Energy
Data centers reaching 12% of US electricity in 2030
54%
Editorial model / Review pending39-69%Alternative assumptions, not a confidence interval
A scenario-based calculation using an assumed probability model, not a distribution supplied by LBNL.
Explore the calculation ↗ 02 / CALIBRATION AUDIT STANDARD
What could
change our minds?
Each published model is a versioned workpaper, not a retrospective score or public claim of predictive accuracy.
REQUIRED BEFORE A PROVISIONAL FORECASTQuantification and ownership
- Lock the binary question, eligible sources, geographic scope and deadline.
- Audit scenario parameter choices against empirical reference-class data or independent elicitation.
- Record model date, editor, source snapshot and every later revision.
REQUIRED BEFORE INDEPENDENT SCORINGValidation and prospective lock
- Two independent reviewers sign off on the unchanged resolution contract.
- Externally timestamp an immutable forecast snapshot.
- Only a future resolved event can receive a score; do not count pre-lock examples.
A-001 and M-001 multiply judgmental conditional factors. E-001 transforms nonprobabilistic LBNL scenarios into a hypothetical triangular probability distribution. None is a fully validated probabilistic forecast. The site continues to show only two provisional numerical estimates (F-001 and F-002), and five exploratory questions.
