How it works

Methodology

Data sources, update schedule, signal families, and known limitations.

Essays

Long-form methodology writeups on specific topics.

Why technical-only signals don't survive on their own
An honest look at what our backtests show — and what the signal workflow is actually for. Single-indicator technical signals on liquid US equities are mostly priced in, mostly by construction, mostly arbitraged away.
Read →

Coming soon

These sections are in draft. Numbers/prose not yet finalized.

  • ◌
    Data pipeline — Yahoo Finance ingest, per-exchange close schedule, 3x-daily intraday refresh windows, how we handle splits/dividends, known data gotchas (GBp sub-unit, JSE ZAc, CN same-day bar NaN).
  • ◌
    Signal families — the three-family tag (trend / mean-reversion / pattern) that every signal carries and how it drives the Explorer "By family" view + composition logic in Daily Reports.
  • ◌
    Backtest methodology — entry = open T+1, horizons 1d/5d/20d/60d/252d, per-stock regime quadrant (ADX≥25 trending / 20d realized vol≥20% high-vol), sub-period splits (2015-19 / 2020-22 / 2023+), 1/99 winsorization, HAC + permutation null.
  • ◌
    Benchmark convention — why we use single-sided long benchmarks (SPX / MSCI World USD / SPXEW) for both bullish and bearish signals. For bearish triggers, negative alpha = signal was right.
  • ◌
    EQTR index construction — equal-weighted daily rebalance, quarterly eligibility check, MIN_PRICE filter, MAX_DAILY_RETURN caps, Investable/Liquid variants. The survivorship-bias caveat and the known penny-stock arithmetic-mean bias.