{"title":"Twenty-seven pre-registered hypotheses, and what happened to each","total":27,"all_outcomes":"killed","method":"Each hypothesis was written down with its pass threshold before the data was fitted. None was revised after the result came back.","lead":{"id":"faza-0-effective-bets-vs-crisis-drawdown","hypothesis":"The number of genuinely independent bets in a portfolio explains how hard it falls in a crisis, over and above volatility, beta, CVaR and trailing drawdown.","design":"16 portfolios, 3 crises, 48 observations. Every input measured 63 trading days before each crisis began, so nothing in the model could see what was coming.","pre_registered":[{"test":"Explains crisis drawdown beyond the classic metrics","needed":"delta R-squared >= 0.05","measured":"+0.0103","detail":"R-squared rises from 0.8805 to 0.8907","verdict":"missed"}],"outcome":"killed","why":"Real, and redundant. The block measures something true about a portfolio and adds almost nothing that volatility does not already say about how it falls.","date":"2026-08-28"},"earlier":[{"id":"drawdown-predictor-v1","what":"A drawdown predictor from market-wide stress","note":"The first version, and the one that set the pattern for every version after it."},{"id":"stress-persistence","what":"Stress that persists, and a threshold to act on it"},{"id":"hmm-regime","what":"Hidden-Markov regime detection","note":"The regimes were legible after the fact and useless before it."},{"id":"conditional-distributions","what":"Conditional return distributions given the stress state"},{"id":"quantile-forecast","what":"Quantile forecasting of the next move"},{"id":"post-crash-reversion","what":"Post-crash reversion, buying the bounce systematically"},{"id":"euphoria-fade","what":"Euphoria fade, selling the top of a run"},{"id":"trend-following","what":"Trend following, twice, on two different constructions"},{"id":"phase-density","what":"A phase-density measure borrowed from physics","note":"Steel-manned deliberately before testing, so the kill could not be blamed on a weak version."},{"id":"probability-layer","what":"A probability layer over the stress signal","note":"Held up on one market in isolation and failed when pooled, which is the version that would have been shipped."},{"id":"funding-carry","what":"Funding-rate carry as a systematic edge"},{"id":"new-mechanisms","what":"Two new mechanisms, generated specifically to break the losing streak"},{"id":"wind-tunnel","what":"A crash-test service for trading bots","note":"Not on statistics, on customer discovery. The people who had the problem already had a good-enough answer for free."},{"id":"marketplace-scanners","what":"Two marketplace and opportunity-scanning products","note":"One was already being done well by an incumbent. Looking first cost a week; not looking would have cost a year."}],"corrected":{"id":"test-b-misattribution","what":"An effect size credited to the wrong cause, and corrected","note":"On the list rather than quietly deleted, because a project that only records other people's errors is keeping a brochure."},"source":"https://www.riskcopilot.app/kills","licence":"Quote freely, with attribution to riskcopilot.app."}