Forecast quality is not one prompt or one model. It begins with a resolvable question, continues through evidence and independent challenge, and ends with an outcome that can be scored.
State what resolves YES or NO, the horizon, the authoritative resolution source, and any assumptions that condition the question.
Gather current public records, structured data, company context, and relevant base rates. Evidence quality and coverage remain part of the result.
Separate research and forecasting roles examine the question from different angles before a combined estimate is produced.
Review checks for unsupported certainty, missing alternatives, calculation errors, and evidence that should move the estimate.
A concise answer is not automatically suitable for distribution. Source readiness, traceability, and quality flags are evaluated separately.
When outcomes become knowable, probabilistic forecasts can be evaluated with proper scoring rules rather than judged by a few memorable hits.
A forecasting product should be precise about the boundary between useful structure and unjustified certainty.
A probability is an estimate under uncertainty, not a promise or prediction of certainty.
Cenva supplies research structure and probabilities; it does not replace mandate-aware judgment.
Different forecast types and engine versions require separate evaluation cohorts.
Missing, stale, or weak evidence is a limitation of the read and should be surfaced.
Research and review take time. Recency is stated with dates, not simulated activity.
Prior-engine performance remains a separately labeled archive with its cohort rules visible.
The active forecasting core will receive a versioned method note and its own evaluation cohort when integration is complete. Prior results are not pooled into that record.
Give us a decision question and we will show the assumptions, evidence plan, and forecast checks it requires.