Every signal on this site is the visible end of a research pipeline most sites this size don't build at all. This page names the real pieces — what each one checks, and honestly, where each one currently stands (some are live and enforced, some are advisory-only, some are still building sample size).
Compliance screening is explained on Methodology, and hypothesis results are published on Research Lab — but the engines that decide evidence strength, watch for data-quality regressions, condition results on market regime, and check portfolio risk before a trade fires have never been visible anywhere on the public site, even though they're real, shipped, and running today.
Every win rate and every hypothesis result on this site is graded, not just reported. A shared statistics module classifies each result as INSUFFICIENT, WEAK, MODERATE, or STRONG using real thresholds — a two-sample statistical test with an exact p-value, Wilson confidence intervals for win rates, and a minimum sample floor before any conclusion is displayed at all. This is the same engine behind every evidence label shown on the research registry and the homepage's win-rate chart — a number never gets to look more certain than the sample size actually earns.
A separate module runs 7 real integrity checks against the live trading data on a schedule — resolved signals missing an exit price, stale cached data, scoring factors silently falling back to a neutral default more often than they should. On its very first live run, it caught a real historical data bug (58 stock signals and 11 crypto signals missing their exit price, all from before a fix already shipped) — exactly the kind of regression this exists to catch automatically going forward, not just once.
A single blended win rate can hide the real story — a strategy that works well in a trending market can lose money in a choppy one. Every scored signal is tagged with the market regime active at the time (trend direction, volatility level), so performance can be broken out by regime instead of averaged away. Still building real sample size per regime bucket — shown honestly with a small-sample warning until there's enough data in a given regime to say something meaningful.
Every signal is scored against 9 real factors (liquidity structure, VWAP, momentum, volume, smart-money-concepts structure, trend, sentiment, fundamentals, risk/liquidity). A separate research module checks whether those factors actually correlate with real outcomes — pairwise correlation between factors, bucketed win rates by factor value, and a redundancy report that flags when two factors are effectively measuring the same thing. Recommendation-only by design: this module can flag a redundant or non-predictive factor, but nothing here ever auto-adjusts a live scoring weight.
Before a new stock position is considered, three checks run against the current open book: sector concentration (already heavy in one sector?), correlation exposure (does this move with what's already held?), and beta exposure. Currently advisory-only — every check runs and is logged on every real signal, but by default nothing here can block a trade; that's a config flag the auto-trader owner controls, not something silently enabled. Being honest about this rather than implying every trade is risk-screened when the enforcement switch is actually off.
A fully separate decision engine from the one that generates the signals published on Signals — its own config, its own paper ledger, long-only. It reads real order-flow structure (VWAP, initial balance, liquidity imbalances, volume profile, cumulative volume delta, large trades, market structure) across 8 stages and only approves a candidate when several stages agree. Currently running in paper mode, strongest on crypto where it has full real order-flow data from Binance; on stocks it works from structure alone and is held to a higher bar. Live, read-only results are on the Signals page's Order Flow Intelligence table.
Every symbol above is only ever a candidate for a signal after it passes compliance screening first — seven published ratio methodologies plus a business-activity gate, cross-checked against two independent third-party providers. Full breakdown of exactly how this works: Methodology.
See also: Research Lab · Screening Methodology · Performance · FAQ. None of this is financial advice — see the disclaimer in the footer.
