Coverage

Our 94.91% number, explained

We publish one Safe Buy Price per company because our numbers have been tested against reality — before we ever showed them to anyone. This page breaks down the three coverage statistics we stand behind, with the exact denominators and the honest caveats attached to each.

94.91%

Floor coverage — our Safe Buy Price stayed below the company's later filing-derived fair value in 94.91% of out-of-sample tests. This is the headline number and the one we quote on the homepage.

Denominator: 746 out of 786 observations across 96 issuers, under a leave-one-out and time-forward protocol.
Honest caveat: This is Safe Buy Price coverage against later filing-derived fair value. It is not a probability of stock profit and it is not exact fair-value accuracy. A safe buy price is deliberately conservative; the point is to give room to be wrong about timing.
96.88%

Qualified evidence support — for the subset of situations where the engine explicitly qualified a filing as safe to enter, the later filings confirmed it in 96.88% of first-origin issuers.

Denominator: 96 first-origin issuers evaluated under the same leave-one-out protocol; this is a different denominator from the ongoing headline production coverage.
Honest caveat: First-origin diagnostic; not the same denominator as ongoing production coverage. A high number here reflects the discipline of when the engine chooses to qualify a situation — not a promise about market behavior.
96.55%

Calendar-lockbox H1 — in the hardest test we run, we walk an entire calendar year of history out of the room, then compare the engine's calls against that walked-out year. On the qualified subset, coverage held at 96.55%.

Denominator: Qualified subset within the walked-out calendar year, on the same 96-issuer research universe.
Honest caveat: Lockbox coverage on the qualified subset — the strictest read on our discipline. It is still a statement about our historical numbers versus later filings, not a statement about future stock returns.

What these numbers are — and what they aren't

Coverage is the fraction of the time our Safe Buy Price was below the fair value the company's own later filings implied. In plain English: when we said “the safe entry is here,” the company's own later filings showed the business was worth more than that. That is a testable claim, and we test it under a strict protocol before we publish anything.

Coverage is not a probability of stock profit. The market can stay wrong about a company for years. And coverage is not the same thing as fair-value accuracy — a safe buy price is intentionally lower than fair value, because the whole point is to have a margin of safety.

Every stat above is checked against the same 786-observation, 96-issuer research universe under a leave-one-out, time-forward protocol. When a company is being scored, its own history is removed from the calibration set; the engine has to price it as if it had never seen it before.

Why we publish denominators

A number without a denominator is not a claim — it is a slogan. Any engine can be right on a hand-picked case study; the value of the three numbers above is that they come with the population they were measured on, and the protocol they were measured under. If a competitor engineer wants to challenge them, the shape of the challenge is completely defined by this page.

Not investment advice. Educational research based on public SEC filings. See methodology and full disclosures.

Read the full methodologyFull disclosures