Five Investing Engine backtests over one window — Dec 2019 – Aug 2026 — every figure transcribed from a captured table that renders its own settings header. Two beat, one matched, two trailed. The runs that trailed are published as prominently as the ones that beat.
Two of five published runs trailed, one matched, two beat. This page is the window, the configuration rules, the fill conventions, and those results — including the ones that trailed. How the models compute stays private: the mechanism is the product, and it stays private because you can verify without it — three of the five models ship a backtest engine you can run on your own assets.
| beat | return strictly above buy-and-hold |
|---|---|
| matched | return within 5% relative below buy-and-hold |
| trailed | return more than 5% relative below buy-and-hold |
Tightness Loose · bar open · 50% entry · LIFO · $100k · no fees
The rule, stated before the result: Tightness set to the swing end of its range, per the model's own guidance that Loose suits swing and position trading. Weekly bars leave no shorter horizon to trade, so the timeframe alone lands this run at that end. All other model settings shipped default.
Tightness Loose · bar open · 100% entry · LIFO · $100k · no fees
The rule, stated before the result: Same tightness rule — weekly chart, position-trading horizon, so Loose rather than Tight. All other model settings shipped default.
Bias Bullish · bar close · 30% entry · LIFO · $100k · no fees
The rule, stated before the result: Bias is chosen from what the asset structurally is, not from anything the chart has already done: a whole-market equity index drifts upward over long horizons, so this run leans toward buys rather than sells. All other model settings shipped default.
Style Aggressive · bar close · 30% entry · LIFO · $100k · no fees
The rule, stated before the result: Style set to surface structure rather than suppress it: a 24/7 asset generates far more tradeable structure than a session-bound one. All other model settings, including noise suppression, shipped default.
Bias Bullish · bar close · 30% entry · LIFO · $100k · no fees
The rule, stated before the result: Same bias rule as the index run. The position is long-only in a growth name, so bias is set to the direction of intended participation. All other model settings shipped default.
Derived from the runs above, never invented: on assets in sustained parabolic advance, the engine underperforms holding. Every exit is a chance to miss the next leg, and the engine takes exits. NVDA on the 4-hour returned 22% of buy-and-hold; BTC daily returned 43%. Both were configured correctly under the rules above — neither is a strawman.
Longest losing streak, worst single trade, longest time in drawdown, and regime slices (2020 / 2022 / 2023–25) are not yet derived from the runs. They will be read off the artifacts or re-run — never invented. Until then, this is the only failure mode we publish, because it is the only one we have derived. Others are not yet derived — not absent.
Each model is built for a market condition, and each has one it handles badly. Both halves are printed below for all 5 models, at the same weight.
This is stated design intent, not a measured result. The runs above cover the 3 models that ship the backtest engine; the other 2 have no published run at all. An envelope tells you when a model is the wrong instrument for what the market is doing — it does not tell you what any of them returned.
Every figure on this page is transcribed from a captured backtest table, last verified 2026-08-07. The screenshots render on the sales page. Subscribe and run the same backtests yourself — change the asset, the timeframe, the parameters. Or break it.
Three of the five models ship the backtest engine — change the asset, the timeframe, the parameters, and see what comes out. You do not need the mechanism to verify the result.