Method, limits & privacy
What was measured, what it cannot tell you, and what this app knows about you
This app is educational. It is not financial advice, not a recommendation, and not a solicitation to trade. It is not personalised to anyone. Its author is not a licensed financial adviser or broker.
THE DATADaily open/high/low/close bars for 27 large US companies plus three index funds (SPY, QQQ, IWM), from 1 January 1998 to 30 August 2026, adjusted for splits and dividends, sourced from Yahoo Finance. Weekly and monthly bars are built from the same daily data.
SURVIVORSHIP BIASThose 30 names were chosen because they are large today. Companies that shrank, merged or went bankrupt over the period are not in the sample, so the absolute figures are optimistic — this is a set of winners. Because every pattern is compared against a baseline drawn from the same stocks over the same period, the bias largely cancels in the difference, which is the number this app actually reports. But treat it as 30 survivors, not as "the market".
MULTIPLE TESTINGTesting 28 patterns means one or two will look convincing by luck alone. So instead of the usual threshold we require z above 3.1 (a Bonferroni correction, 0.05 ÷ 28). Patterns that would pass on their own but fail this bar are labelled as such rather than hidden.
WHAT THE % MEANSFor single- and multi-bar candlestick patterns it is how often price was higher five bars later. For the larger formations it is how often the break carried on in the direction it broke — not an up-rate — measured over a window as long as the formation itself, between 10 and 120 bars, against a random position taking the same side over the same stretch.
NOT A BACKTESTThese are raw directional frequencies. No entry or exit rules, no costs, no spread, no slippage, no taxes, no position sizing, no stops, no compounding. A pattern with a positive number here would not necessarily make money, and probably would not.
WHY THE RESULTS LOOK BACKWARDSMost of these shapes form after a move has already happened, and short-horizon price changes tend to lean back the other way — an effect documented in the academic literature since around 1990. Controlling for the preceding move removes most of the apparent effect. The patterns are largely short-term reversal wearing a costume.
PRIVACYThis app stores one thing: a flag in your own browser recording that you have read the opening notice. Nothing else. There are no accounts, no sign-in, no email, no analytics, no tracking and no advertising. Your answers and scores live in the page while it is open and are gone when you close it — they are never uploaded anywhere.
The typefaces are embedded in the page rather than fetched from Google, so the app makes no third-party requests at all. No other company learns that you were here. The one thing outside this app's control is the web host: like every website, it keeps ordinary server logs, which include visitors' IP addresses. The support link opens a separate site with its own terms; nothing about your use of this app travels with you.
BUILT8 September 2026 · Results are reproducible from the underlying data pipeline. If the dataset is rebuilt, these numbers may shift slightly.
Verdictus is free, has no ads and asks nothing of you. If it taught you something and you would like to chip in towards keeping it running, there is a tip jar — entirely optional, and nothing in the app changes either way.
Support this project →