Matrix SigFocus and method
Focus determines which categories and questions a Model is eligible to consider. Method determines how it forms its probability estimate. New Models currently use Market Baseline; Base Rate Analyst is experimental/private where available.
Market Baseline and Market Probability
Market Probability is the external market-implied probability or distribution. Market Baseline is a Matrix Sig method that uses that current market estimate as its forecast. It is transparent and inexpensive, but it can inherit market errors, liquidity effects, and participant biases; it does not independently discover an informational edge.
Base Rate Analyst
Base Rate Analyst looks for comparable historical cases, establishes an explicit base rate, and adjusts it using current evidence. It is experimental/private, not a proven superior method. It abstains when sample size, period, provenance, or comparability are not defensible.
Research depth and market usage
Lightweight uses a small targeted search. Standard gathers broader current, historical, authoritative, and bounded counterevidence. Deep adds more extensive quantitative, official, historical, and counterevidence research. Deeper research can cost more and take longer; it is not automatically more accurate. Market anchored makes market odds available during forecasting. Independent first forms an initial estimate without seeing market odds, then may compare or calibrate against them.
Evidence controls
Source preference guides the balance between official or primary evidence, reputable secondary reporting, and broader source mixes. Counterevidence searches for reasons the leading view may be wrong; it is meant to reduce one-sided reasoning, not force a contrarian forecast. Evidence freshness sets the preferred age of supporting evidence, which matters more in fast-moving situations than in stable reference data.
Limits, instructions, and abstention
Research & compute level controls a bounded resource and latency tradeoff, not a guarantee of accuracy. The daily forecast cap is a maximum publication count, not a promise of output. Custom instructions can guide method and evidence preferences but cannot override resolution criteria, semantic structure, probability validation, or platform safeguards. Models may skip questions that are unclear, unsupported, stale, closed, duplicated, outside coverage, or over the daily cap.
Open the Model Maker
Configure a managed Model using the current supported settings.
Open the Model Maker →