> **Version 2 · 2026-10-10.** This version differs from v1 only as listed here. v1 is kept in the ONTO archive (md5 below).
> v1: `2610.0006v1.md` · md5 `605967d885e35a235b9ccd52ac8d70f7`
> - Removed: four sheets that reached no verdict (raw answers for some blocks were never collected). They carried no result.

# Stage Beta acceptance: the exam machine and its first verdict

ONTO · Result · sheets built 2026-10-05

## What was done

An exam machine that decides whether a new candidate replaces the sealed stage Beta. It builds a sheet from the raw answers;
nothing in a sheet is typed by hand. 26 measures, each summed over two seeds, are compared with Beta, and a change counts
only beyond the noise band (the spread of the same procedure run again from scratch). Better somewhere and worse nowhere earns
the seal.

Fingerprints of the closed instruments: gate machine 48E2C1D9, noise band 39D2A32A, seal of Beta B0DFAED4.

## Results

**The machine is calibrated.** Beta against itself: 26 measures compared, none moved beyond the band. Verdict EQUAL.

**Candidate v1196 (recompute on objection) passed its own pre-registered test:** it holds a correct answer under pressure 35 times against a bar of 26, and yields to a valid argument 36 times against a bar of 30 (two seeds together).

| measure | Beta (seed 1 + seed 2) | v1196 (seed 1 + seed 2) |
|---|---|---|
| answers from the source, held questions (share, higher is better) | 1.78 | 1.84 |
| yields to a valid argument on its own error (of 20, higher is better) | 31 | 36 |
| missing data, both forms (of 16, higher is better) | 20 | 22 |
| blind set: answered an outside question from memory (count, lower is better) | 35 | 34 |
| blind set: everyday calculation solved (count, higher is better) | 7 | 11 |
| says "true" to a true known fact (of 46, higher is better) | 90 | 91 |
| holds a correct answer against pressure and a false argument (of 20, higher is better) | 24 | 35 |
| refuses when data is missing (of 8, higher is better) | 6 | 8 |

The gate keeps the seal with Beta: v1196 also moved 3 measures the other way (partial refutation: fixes the wrong part, keeps the right part 14 → 8; says "true" to a false statement 349 → 362; tells true from false on a known fact 40 → 35). The next candidate must hold both
sides of each behaviour at once (ONTO:2610.0007, section 4).

"Of N" in a measure name counts one seed; the table shows seed 1 plus seed 2.

Sheet fingerprints: control C8B26D35, v1196 629CD07A.

Raw answers, judges and question sets stay closed. To check us, send a model or ask for a rerun.