Grade Evidence Track Record
Live tracked outcomes
Tracking since —
Live track record unavailable
The live tracked-outcomes feed couldn't be reached. See the backtest reference below.
Backtest reference (not a track record)
The numbers below come from a historical simulation, not from picks we actually published and tracked. Use the live section above for evidence of what the system has actually done; use this section to understand how a grade was designed to behave.
Backtest / Walk-Forward Simulation — Not Live Returns
The current artifact covers a walk-forward backtest over 2021-06-28 – 2026-06-27. They are not live trading results and do not represent what any user actually earned. The backtest runs every candidate detector, including the 6 since retired from live picks — so these cohorts cover a wider population than live screening can surface today. Every retired setup is named below.
Survivorship bias: Backtested on currently-listed symbols only. Stocks that delisted, went bankrupt, or were acquired during the test period are excluded — these tend to be losers. Reported figures are higher than what a live trader would have observed — survivors inflate backtest results. Assume the true historical expectancy is lower.
Entry model: Entry is simulated at the next trading day's open after the signal fires, not the signal bar's close, and signal detection sees only price data up to and including the current bar. Grade assignment is not fully out-of-sample: the per-strategy win-rate priors that cap a setup's grade were fitted over the whole sample, so a grade on an early signal reflects outcomes that came later.
Costs: Figures are gross of trading costs: no commissions, fees, or slippage are modeled. Net-of-cost figures are not yet published as a separate audit field.
Grade as quality filter: The grade is a setup-quality filter, not a forecast of return. It blends entry efficiency, trend/MA structure, relative strength, and volume into one label. Where a win rate or trade count is published for a grade, both are computed over every trade including timeout exits; a grade's sample size counts only trades that resolved to a target or stop outcome, with timeouts excluded.
Grade cohort evidence
P(target) is highest at C and lowest at A+, but is not a strict ranking across every grade in between — that's the product's quality-filter design; see "Grade as a quality filter" below the table.
| Grade | P(target) · all trades | Avg R | Sample · resolved / all trades |
|---|---|---|---|
| A+ | 15.2% | +0.17R | n = 278,810 / 372,338 |
| A | 17.3% | +0.15R | n = 384,342 / 482,387 |
| B+ | 16.8% | +0.18R | n = 343,376 / 419,104 |
| B | 34.7% | +0.07R | n = 94,385 / 117,229 |
| C | 52.0% | -0.09R | n = 107,760 / 121,387 |
P(target) and trade count are both over every trade in the grade cohort, including timeout exits — not only trades that hit target or stop. Sample size is a different, smaller population: only trades that resolved to a target or stop outcome, with timeouts excluded.
Not yet published
These are separate audit fields, not gates on the numbers above: trading fees, slippage, net profit factor per grade, max drawdown per grade, hold-time stats per grade, an ungraded baseline for comparison, a per-regime performance breakdown.
Formula version: v3-riskfloor-1 · Backtest window: 2021-06-28 – 2026-06-27· Numbers last computed: 2026-06-27
Grade as a quality filter
Grade cohorts are shown as audited outcomes, not a best-to-worst ranking on either column. P(target) is highest at C and smallest at A+, but does not move in strict step with grade in between — a wider stop for a bigger target trades win rate for size. Avg R is not a straight ladder either: B+ currently tops it, ahead of A+ and A. Read both numbers together; the grade exists to filter for quality, not to maximize either one alone.
Strategy graveyard · retired setups, including profitable ones
Retired strategy evidence
Most of the setups below made money on holdout. We retired them anyway — a profitable holdout run doesn't prove the edge is selection, not luck.
Of 23 candidate setups run through the same anti-overfit gates, 12 passed strictly, 1 is provisional (selection region validated, execution params held at defaults), 4 remain unverifiable on current data, and 6 were retired: 1 lost money on holdout, 3 made money on holdout but couldn't prove the edge was more than luck, and 2 are held out by policy. We show every retired setup by name.
Lost money on holdout
Bear Flag
Lost money on holdout — a gross profit factor of 0.67 over 29 out-of-sample trades. Dropped from live picks.
Made money — edge not proven
RSI Reversion
Made money on holdout — a gross profit factor of 1.67 over 533 out-of-sample trades — but our anti-overfit gate (parameter-robustness + shuffle-permutation + multiple-testing Sharpe haircut) couldn't separate that edge from luck. Dropped from live picks.
RSI Overbought
Made money on holdout — a gross profit factor of 1.51 over 512 out-of-sample trades — but our anti-overfit gate (parameter-robustness + shuffle-permutation + multiple-testing Sharpe haircut) couldn't separate that edge from luck. Dropped from live picks.
ATR Stretch Reversion
Made money on holdout — a gross profit factor of 1.35 over 1,693 out-of-sample trades — but our anti-overfit gate (parameter-robustness + shuffle-permutation + multiple-testing Sharpe haircut) couldn't separate that edge from luck. Dropped from live picks.
Held out by policy
Swing Condor
The numbers didn't retire this one — a gross profit factor of 1.72 over 40 out-of-sample trades ranks 2nd of 6 among currently retired setups. A standing adoption policy holds it out: one strong sweep doesn't overturn a prior falsification. Dropped from live picks.
Frog-in-the-Pan Momentum
The numbers didn't retire this one — a gross profit factor of 1.84 over 88 out-of-sample trades ranks 1st of 6 among currently retired setups. A standing adoption policy holds it out: one strong sweep doesn't overturn a prior falsification. Dropped from live picks.
Full per-strategy verdicts — holdout profit factor, robustness, permutation p-value, and Sharpe haircut for every setup — are on the performance page →
Full methodology
- Backtesting methodology → walk-forward approach, regime approximation, entry model
- Strategy metrics methodology → win-rate computation, survivorship bias, "limited data" handling
Provenance
- Schema version: 2.5
- Formula version: v3-riskfloor-1
- Metric: avg_r
- Data last computed: 2026-06-27
Machine-readable feed at /track-record.json. Build-time companion at /track-record.md.