CALIBRATED

Reference Predictor R-01 is an illustrative strategy constructed from published quality and low-volatility factors on a US large-cap universe. On a documented ledger of 512 resolved predictions evaluated by CalibrationJudge on 2026-07-06, R-01 passes each of the five gates specified in the Refusal Doctrine v0.3.1.

The Brier Skill Score is positive and material. Regime segmentation identifies three distinct regimes, and R-01 remains profitable in every one. The guided-search safety test rejects the null hypothesis of overfitting after correction for the disclosed number of configurations tested. Calibration error sits well below the accepted threshold. The bootstrap confidence interval on skill excludes zero on the downside.

A verdict of CALIBRATED is not a guarantee of future returns. It is a statement that, as of 2026-07-06, the strategy has survived a documented adversarial process against the specific gates of the Refusal Doctrine v0.3.1, within the scope declared in Section 9.7. Deployment beyond that scope requires a new evaluation.

This Verdict-form is published as a reference case. Its purpose is pedagogical: to make the shape of a passing verdict visible alongside the shape of a refusal (Verdict-form 001, Synapse v0.3, NOT-CALIBRATED). The contrast between the two documents is the intended teaching material.

A CALIBRATED verdict is more informative when the source of edge is named alongside the statistics. Reference Predictor R-01 declares an edge in the behavioral / risk-premium class, specifically a combined tilt on Novy-Marx gross-profits-to-assets and low-volatility ranking, applied to the US large-cap universe on a daily horizon.

The empirical signature matches the declaration. Cross-sectional exposure to the declared factors accounts for the majority of realized excess return, with residuals concentrated in periods where the two factors historically decouple. The Brier Skill Score decomposition confirms that the strategy's probability estimates are informative rather than merely correlated with realized outcomes.

The gates further show that the reported performance does not depend on a narrow window. Regime segmentation via HMM identifies low-vol, mid-vol, and crisis regimes; R-01 is positive in each. The Aronson-Masters guided-search correction rejects overfitting at p = 0.018 given the disclosed 42 configurations tested during construction. The bootstrap confidence interval on skill excludes zero on the downside, with a lower bound of 0.024.

A CALIBRATED verdict is time-bounded. R-01's verdict applies as of 2026-07-06 and requires revalidation on the cadence declared under its Model Trust Level. Deployment beyond the sizing envelope stated in Section 9.7 above is out of scope.