Methodology v1.0

This page is generated from the same configuration file the scoring pipeline runs, so the published rules always match the shipped rules. Any change is a versioned methodology change.

What counts as a claim

Any forward-looking statement by a company executive on an earnings call: guidance, targets, timelines, input-cost expectations, macro views, however hedged. Analyst statements and restatements of past results never count. Quotes are stored verbatim; claims are extracted by claude-sonnet-5 in two passes with a human review queue for low-confidence extractions.

Verification

Each claim carries a direction, an optional magnitude, and a deadline, verified independently against reported financials and public series (FRED). Flat means within ±4.5%. Numeric magnitudes verify within ±20% of the claimed change (±75 bps for margins); absolute targets verify as range containment. Direction achieved up to 1 quarter late counts as a partial, not a hit. Claims with no mappable series are displayed but not scored.

Scoring

Accuracy (0 to 100) is a weighted hit rate: precise and longer-dated claims earn more credit, partials earn half. Bias is the signed average error on numeric claims: positive means overpromising, negative means sandbagging; per-claim errors are capped at ±100%. No executive receives a public score below 12 resolved claims. Tiers are percentile bands within the rated universe.

What the scores are

Opinions derived from public statements via the published rules above. Every claim links to its verbatim quote, context, and the SEC filings of the relevant period. Corrections: methodology@maatindex.com.