Primary sources. Explicit assumptions. Forecasts that can be scored.
RC is designed around a clear separation between evidence, inference and prediction.
Primary-source first
Company filings, government documents, program records, technical material and direct disclosures should anchor material factual claims.
Make uncertainty visible
Separate what is observed from what is inferred. State assumptions that materially drive conclusions.
Put a number on the future
Use 1–99% probabilities, defined close policies and predetermined resolution criteria. Resolved forecasts are scored using Brier scores.
Compare judgment on the same question.
RC, community, ChatGPT and Claude probabilities are kept distinct. AI estimates are timestamped assessments, not live feeds or endorsements.
Same question. Same resolution rule. Same protocol.
RC records the exact model/version, assessment time, information cutoff, probability, rationale and prompt snapshot. When a model changes its view, the new assessment is appended as a revision; the old probability is not overwritten.
Model assessments are entered and timestamped by RC using documented prompts so each probability can be audited against its model version, information cutoff and rationale.
Estimate the probability that the forecast below resolves YES according to the supplied resolution criteria.
Return a probability from 1% to 99%. Base the estimate only on information available through the stated information-cutoff date. Consider relevant base rates, technical readiness, schedule risk, management guidance, external dependencies and historical evidence. Do not simply reproduce market, RC or community consensus. State the probability first, then concise reasoning, then the three factors most likely to change the forecast.
FORECAST QUESTION:
{{QUESTION}}
RESOLUTION CRITERIA:
{{RESOLUTION_CRITERIA}}
FORECAST CLOSE POLICY:
{{CLOSE_POLICY}}
INFORMATION CUTOFF:
{{INFORMATION_CUTOFF}}