MACROGURU

◎ Reality Check

← Front page · Live Monitor · Scenario Lab
The only score that matters: how close to reality are we? Two numbers per scenario — will it happen, and how far each market moves — both checked against what actually happened. Wins and losses, in public. Nothing cherry-picked.

This week's calls · what reality decided

Loading the latest outcomes…
Loading the public ledger…
Loading the track record…
Scoring method, model weaknesses, and full audit detail

Calibration — when we say 70%, does it happen ~70% of the time?

Every resolved call adds evidence. We cannot quietly edit the history: publications, resolutions, and evidence corrections are hash-chained.

Bars = the share that actually came true (accent) vs the average probability we assigned (gold tick) in each band. On the line = perfectly calibrated. Small bands are noisy — counts shown at right. Recipe: methodology.

Cascade backtest · our starting baseline — not a track record

Loading cascade results…
Across every scenario × asset cell with both a forward direction and a measured historical abnormal return: how often they point the same way. ~50% is a coin flip — we publish it anyway. This is the baseline the learning loop has to beat.

The fix-log · what our own scoring caught

We grade ourselves before anyone else can. The full history stays public; the newest outcome and custody fixes appear first.
  • 2026-08-02 · A calendar was being treated as proof — FIXED. Scheduled dates may establish when an event should occur, never that it occurred. Composite results now require explicit source review and can bind several sources into one immutable resolution receipt.
  • 2026-08-02 · Evidence-reference correction — APPEND-ONLY. A malformed price-data query was not rewritten. A chained amendment preserves the original receipt, records the reason, and publishes the corrected evidence bundle.
  • 2026-07-03 · Daily refresh wasn't reaching production — FIXED. Deployment moved to the sanctioned release path, and freshness is now tested as a product contract.
  • 2026-07-02 · Circular base rate — FIXED. The probability anchor now comes from measured, dated events rather than an average of authored priors.
  • 2026-07-02 · Interval bands too narrow. Impact bands are re-derived from analogue dispersion and checked against observed coverage.
  • 2026-07-02 · Confidence was anti-predictive. Confidence is being replaced by an empirically fitted reliability table; until then, it is not a portfolio weight.
Method: one-sided Brier score in [0,1] (0 = perfect, 0.25 = no-skill coin-flip, 1.0 = confident-and-wrong), with the Murphy reliability/resolution decomposition and a Brier-skill-score vs the base rate and vs the market. Scheduled-certain calls (a report lands, a market is closed) are shown but excluded from the skill score. Grounded in Tetlock's Good Judgment Project, Murphy (1973), and Metaculus track-record practice.