The Research Ledger

Every pattern we tested, one page each

Each of these was written down in full — the exact rule, the pass and fail thresholds, the data window — and cryptographically fingerprinted before the test was run, so the goalposts could not move afterwards. Every one of them then failed. The individual verdicts matter less than what they have in common.

Chart patterns and candlestick setups

0 for 14

Fourteen named chart patterns, each pre-registered and tested on its own. None survived. Two findings recur often enough to be worth more than the individual verdicts: the patterns usually measured the market rather than the stock, and the specific clause a pattern is named for was frequently the part subtracting value.

This is a fraction of what has been tested

192 experiments have been through this process and 105 were rejected outright. The ones with their own page here are the patterns people ask about most. The research ledger covers the rest — exit rules, momentum variants, seasonality, sector rotation, short interest — and the eight ways a backtest can lie to you, which turned out to be more reusable than any edge the programme ever found.