What can be verified, and what is so far only claimed

Verifiable today

• The arithmetic of letters and numbers: the full formula is visible and every step can be redone by hand.
• Torah text: Westminster Leningrad Codex, CC BY 4.0, with a link to the primary source next to every citation.
• The structure of 354 pages of 861 letters: recomputed independently from the WLC; 31 of 31 control pages match the edition.
• Open data and a provenance record: every calculation carries the engine version, the data version, a date and a permanent link.
• The coincidence detector: Monte Carlo, random substitutions, search space and a multiple-testing correction.
• Published research with a negative result: equal letter sums do not mean related meaning.

Claimed, not proven

• Matrix 861 and a count over 861 variables.
• Standard 354 as an analytical system.
• Temporal Sync.
• Atbash Vector.
• DTP.
• Systemic audit and risk analysis built on the above.
• Any statement about forecasting accuracy.

For all of the above there is no published specification sufficient for independent reproduction. That is why this site quotes no accuracy figure and no paid product is built on it.

How it will be tested

1. Specification. The full list of variables and counting rules is published with a version and a date — before any test.
2. Preregistration. The hypothesis, the success metric, the sample size and the failure criterion are fixed publicly with a hash, before we see the data.
3. Blind out-of-sample test. A third party supplies inputs without answers; the answers are revealed after the forecasts are submitted.
4. Baseline. Comparison not against zero but against a simple model: frequencies, averages. Without a baseline any figure is meaningless.
5. Null models and permutations. The same thing we already do in the detector.
6. Independent replication. Code and data are open so that an outside researcher gets the same result — or does not.