Prescience

Avg conviction —

Prescience measures how strongly the model disagrees with a coin-flip baseline (50/50). The probability shown is the served base rate for the class and horizon; the model's deviation from it is an experimental tilt that has not beaten the base rate. Distance from 50/50 therefore mostly reflects the class base rate, not model conviction. Signals are ranked by that distance, descending; it is a sort, not a skill claim.

Honesty: prescience_score = |prob_up − 0.5| × 200. A high score reflects distance from 50/50, not accuracy or skill (the served P(up) is the base rate, not the model) — calibration performance is tracked separately on the Calibration Scoreboard. Treat it as a sort order, not a prediction of outcome.

Loading…

Strongly Bullish

—

prob_up > 60%

Neutral

—

40% ≤ prob_up ≤ 60%

Strongly Bearish

—

prob_up < 40%

Conviction spectrum — distribution of prob_up across all signals

Loading spectrum…

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Bearish (<40%) Neutral (40–60%) Bullish (>60%)

Signals — ranked by prescience score

#	Ticker	Prob Up	Prescience	Tier	Horizon	Conviction
Loading signals…

Formula: prescience_score = |prob_up − 0.5| × 200 (0 = random walk, 100 = certainty). Conviction label: Strongly Bullish >70%, Bullish >60%, Neutral 40–60%, Bearish <40%, Strongly Bearish <30%.

Source: /api/daily-edge?limit=200 — horizon h=1 predictions from the enriched forecast ledger. Methodology hash: prescience-2026-10-05-v1. Auto-refresh every 120 seconds.

Composes with: Daily Edge · Calibration · Brier by cohort