Specimen N-01 · constructed example · Refusal Doctrine v0.3.3.1
Specimen N-01
NOT-CALIBRATED · specimen, not a verdict
What it shows
Section 2.3 of the doctrine says that the gates run in order and that “A strategy that fails one gate is not tested at subsequent gates.” Specimen N-01 follows that path: Gate 1a passes, Gate 2 fails, and Gates 3, 4 and 5 are not tested. Its NOT-CALIBRATED result rests on the gate that failed; section 6.6 says “A verdict rests only on the gates actually measured”.
| Gate | Test and threshold | Value | Result |
|---|---|---|---|
| 1a | Deflated Sharpe Ratio, six disclosed trials, 1,260 daily returns · threshold ≥ 0.95 | 0.968 | PASS |
| 2 | Sharpe ratio in each regime, with its 95% interval at the effective sample size · threshold > 0 in each regime | low-vol 3.13 [1.58, 4.68] mid-vol 2.00 [0.88, 3.16] crisis −4.57 [−7.63, −1.73] | FAIL |
| 3 | Guided-search safety, permutation test | not tested | |
| 4 | Probabilistic calibration | not tested | |
| 5 | Bootstrap confidence | not tested |
Sharpe ratios are annualized. The crisis regime is negative and its whole interval lies below zero. That is the failure mode section 9.2 names: “strategy is profitable in aggregate but negative in one or more regimes”; here the aggregate annualized Sharpe ratio is 1.15.
How it was built
A script with a fixed seed generates five years of daily returns, 1,260 days, from three regimes with different means and volatilities. The configuration evaluated is the best of the six trials disclosed for it. The regimes are constructed for the example: they stand for the Hidden Markov Model detection on market data that section 9.2 prescribes, which is not run here. The specimen therefore illustrates the decision rule on given regimes, not a full execution of the prescribed test. The intervals are moving-block bootstrap intervals of the annualized Sharpe ratio.
A second script, written by the same author, recomputed the threshold of the trials, the Deflated Sharpe Ratio and the intervals by other routes, and its results agree. It is a cross-check, not an external validation. The formula of the Deflated Sharpe Ratio reproduces the worked example of Bailey and López de Prado (2014).
Read with
Specimen R-01, also constructed, shows what a CALIBRATED form would look like. The gates and their thresholds are set out on the methodology page and in the doctrine.
Specimen: no hash, no signature. A constructed example, not a verdict.