This page is presented as a historical evaluation, not as a verdict under the Refusal Doctrine. It was published as Verdict-form 001. Nyalai has decided to present it as Historical evaluation 001: an evaluation of Synapse v0.3, a system built by Nyalai's founder, run on 2026-07-01 by CalibrationJudge, the calibration judge of the validator package cice. The judge's code and the run's results were first recorded together in commit cice/48910d9. The identifier 001 and the address of this page are unchanged, and so are the five measured values. NOT-CALIBRATED is the output of the judge as implemented; it is not a verdict rendered under the Refusal Doctrine. This evaluation is not counted among Nyalai's published verdicts.

Why it is not a verdict under the Refusal Doctrine.

• No text of the Refusal Doctrine dated or recorded on or before 2026-07-01 has been found in the four repositories searched: those of this site, of the doctrine and the validator, and of Synapse. The earliest doctrine file is dated 2026-07-06. The earliest mention of the Refusal Doctrine in this site's source, committed on 2026-07-08, dated two of its versions to 2026-06-30, which the doctrine files do not bear out.
• Tests III, IV and V of the judge are not the gates the doctrine specifies. Test I uses the score and limit of the doctrine's Gate 1b without the correction that gate requires; Gate 1b was added to the doctrine on 2026-07-14, after this page, then citing v0.3.2.1, had reported that score. Test II uses the calibration error and limit of the doctrine's Gate 4.
• No one-page thesis in the form the doctrine requires, dated before the run, has been found. The doctrine gives no verdict to a strategy whose thesis was not submitted (section 9.0).

Statements withdrawn or corrected.

1. "the run of 2026-07-01 was made under v0.3.2.1": withdrawn, for the reasons above.

2. "fails all five gates specified in the Refusal Doctrine v0.3.2.3", "refused across all five gates of the Refusal Doctrine v0.3.2.3" and "All five gate specifications and thresholds are defined in the Refusal Doctrine v0.3.2.3": withdrawn. The five lines are the five tests of the judge as recorded at cice/48910d9, and each failed as implemented. This table renumbers the tests of this page's historical body to follow the code: in that body, the regime groups were numbered II and the calibration error IV. The body of this page now uses the numbering of the code.

   Test of the judge	What was executed	Value
   I · Skill against the base rate	Brier Skill Score against the empirical base rate, limit 0.05, no correction	BSS = -0.079
   II · Calibration error	Expected Calibration Error on ten equal-width bins, limit 0.05	ECE = 0.141
   III · Permutation	1000 permutations of the outcomes, no correction	p = 0.773
   IV · Regime groups	Brier Skill Score and calibration error in each group of regime labels supplied with the ledger, for groups of at least eight predictions; the 43 predictions without a label, evaluated as one group, also have negative skill; one label with a single prediction was excluded	4 of 4 named groups negative
   V · Bootstrap	1000 independent resamples of the pairs, interval on the Brier Skill Score	CI = [-0.147, -0.025]

   The doctrine's gates 2, 3 and 5 call for a Sharpe ratio in each regime detected by a Hidden Markov Model, a guided-search-safe permutation test corrected for the number of configurations, and a bootstrap of the Sharpe ratio.

3. "The guided-search safety test does not reject the null hypothesis of overfitting": withdrawn. The test run was a permutation of the outcomes. It does not reject the hypothesis that predictions and outcomes are unrelated (p = 0.773).

4. "Instrument scope: Public equities, listed derivatives": withdrawn. The ledger does not record an asset class. Of its 87 instrument names, those that can be identified read as crypto-assets and blockchain networks, and about twenty cannot be identified from the name alone. One more name, SPX, is also the name of an equity index.

5. "Validator code cice/87c619a": a precision. 87c619a is a later commit, made that evening, New York time, that adds a case study and changes no existing file, so the judge's code is the same at both. The judge was first recorded in git, with the run's results, at cice/48910d9.

6. "Any independent third party can reproduce this evaluation from the same ledger" and "Any third party can request the ledger to reproduce the evaluation independently": corrected. Neither the code of the judge nor the ledger is public. Reproducing the evaluation requires both: the ledger and the code at cice/48910d9.

7. The root-cause diagnostic. Its three figures reproduce from the ledger and from Synapse's base-rate table, under conventions now stated. +0.838 is the gain in Brier Skill Score from inverting the probabilities of the 19 predictions of the sudden_trending_entry detector, measured against their own base rate. +0.035 and +0.338 come from a heuristic model of the revised writer, which did not call the language model, applied to the same corpus that produced its base rates; the weight of 0.5 was a choice, not a measurement, and the configuration retained was one of nine tried, not the most favourable. "passes 315 of 315 tests": withdrawn. The traces found, a test cache of 2026-07-01 and a recap of 2026-07-04, do not establish a dated, successful run of these 315 tests. This check covers these figures, not the whole section.

What else changes on this page. The doctrine version is removed from the heading, the metadata and the scope heading. The five lines carry the names of the judge's tests. The section "Sovereignty rule seven" is rewritten: this evaluation, of a system built by the founder, was made with Nyalai's own validator, and is disclosed as a self-evaluation; it is not a verdict-form. The signature carries the date of this notice.

What does not change. The five values, and the judge's output, NOT-CALIBRATED. The sealed record verdicts/001_synapse_v03.md is kept as it was sealed, with its SHA-256; this notice accompanies it as a separate file and does not change its bytes. The sealed record is provided unchanged as a historical document. The corrections in this notice also apply to the corresponding statements in that record. This notice addresses only the statements expressly corrected here; it does not validate the remaining contents of the historical record. Its offer to provide the ledger on request is withdrawn, as is the announced re-run scheduled four to six weeks after publication. The statement about independent third-party reproduction does not establish that the required code and ledger are available. Providing the sealed record does not include access to either. The validator's own history uses the word verdict in its own sense, an output of the validator: the run commit calls this run its fifth formal verdict, and the count rule of 2026-09-28 counts it among four formal verdicts before Verdict-form 002. That history is unchanged. The validator's historical count is under reconciliation.

Editions. The French and Japanese editions carry this notice.

Sovereignty rule seven is the core operating constraint of Nyalai. It requires structural separation between the party that constructs a probabilistic system and the party that renders a verdict on it. Without that separation, the verdict is not a verdict. It is a self-report.

The system evaluated here was built by the founder, and the evaluation was made with Nyalai's own validator. That is a self-evaluation, disclosed as such.

This evaluation is not a Verdict-form.