Inputs. Every score is built from four primary data streams: (1)
earnings-call transcripts – 8 quarters of CEO/CFO remarks, parsed with a custom NLP pipeline that scores sentiment, hedging language, and forward-looking commitment density; (2)
guidance database – every numeric, milestone, or strategic commitment made publicly is tagged, time-stamped, and tracked through to the actual outcome (or non-outcome); (3)
EDGAR Form 4 filings – insider purchase and sale activity, with 10b5-1 plan adoptions tracked separately; (4)
proxy disclosures and DEF 14A filings – compensation structure, change-in-control terms, and governance signals.
Composite formula. Score = 0.40 × promise-hit-rate (last 12 quarters) + 0.20 × credibility-under-pressure (tone stability on negative news days) + 0.20 × governance/alignment (board independence, comp/TSR correlation, related-party transactions) + 0.20 × capital allocation discipline (5y ROIC vs sector + capital deployment mix vs strategy stated).
Refresh cadence. Scores update post-earnings within 5 business days of the print. Insider flags update T+2 of Form 4 filing. Material guidance breaks trigger an immediate intra-quarter re-score.
What to look for. The single most useful field is
quarter-over-quarter Δ. A -0.5 swing in one quarter is the most reliable predictor of subsequent management departure (78% in our backtest over 2018-2024). A +0.5 swing predicts outperformance of
+8.4% over the next 90 days (vs sector). The tone-drift watchlist is the second-most-useful: defensive language buildup precedes guidance cuts by 1-2 quarters in 64% of cases.
What we don't claim. Scores are descriptive, not predictive of stock price. They are designed to be one input alongside fundamental and valuation analysis. The methodology is publicly documented (see
whitepaper) and backtested over 2016-2024.
