pre-register-triallisted

# Pre-register a Trial Every hypothesis enters as a registry trial with its decision rule committed **before** data accrues. The git commit timestamp is the tamper evidence. No metric substitution, window cherry-picking, or post-hoc regime adjustment — ever. ## Why this exists The registry + deflation guards are the project's defense against its own hindsight. DSR/PBO deflate against the CUMULATIVE trial count — so every trial, including ones that will obviously fail, must be counted. A registry showing only adoptions is lying to itself; rejected trials are published (NEGATIVE_RESULTS.md). ## What a registration contains (all required) 1. **Hypothesis** — one sentence, falsifiable, with the honest prior stated (e.g. TRIAL-EXIT: "shallower maxDD at flat-to-slightly-lower Sharpe, NOT a Sharpe increase"). 2. **Primary metric** — the ONE deciding number (net Sharpe vs named control / forward rank-IC with CI / forward Brier vs climatology). Everything else is "reported, never deciding." 3. **Decision rule** — adopt threshold, reject threshold, minimum window, evaluation cadence, earliest decision date. Include the crash-event override where relevant (SPY −20% defers decisions to ≥6mo past trough) and a contamination clause for data defects. 4. **Frozen parameters** — whatever must not be tuned mid-trial, named. 5. **Hard constraints** — descriptive-only / never-arms / no buy-sell language until passed, where applicable. ## Procedure 1. Write the cano