01 · Attrition

Most strategies don't survive.

Around twenty candidate strategies measured against matched controls. One is still standing.

1 / 20Survived
684Validated
02 · Setup taxonomy

Each setup was measured separately.

Grouped the way they are actually traded, not split into components. FTFC does not get its own row — nobody trades timeframe continuity alone, it is a filter layered onto a sweep or a combo. Measured on its own it is −0.034R at t = −11.52 across 150,000 observations, negative in both directions — so as a filter it can only subtract. Combined forms were covered separately: a from-scratch search over 276 rules had 78 clear family-wise correction and all 78 died out-of-sample.

Measured on 56 names back to 1962, entry at the next open, scored as excess over each symbol's own drift, every rule mirrored, family-wise corrected.

64 yrLongest sample
150kObs · FTFC alone
03 · Rigor

Backtests are context, not proof.

See the live forward test →

Every number below is research. The live system earns the claim — this page shows the work behind the rules.

BACKTEST REPLAY Scroll to run · 5,284 signals
The bar

The testing is the product.

A rule has to beat a random rule admitting the same number of trades, survive its own direction mirror, hold across eras, and clear costs. Most don't get past the first gate.

~20
Strategies tested
Each against a matched control
and its own direction mirror
684
Patterns validated
Pass all quality, RVOL,
and structure gates
17%
Pass rate
One in six makes it through.
That's intentional.
Setup breakdown

Win rate and avg R — per setup.

Each Strat Combo is tested independently. Blending them into one number hides where the edge actually lives.

Setup Signals Win rate Avg R Strength
MR Core — mean reversion SURVIVED
Short-horizon oversold entry inside an uptrend; entry at the next open; strength-or-time exit; 8R catastrophe stop only
61,432 67.5% +0.091R
Beats a matched-random entry in the same names on the same rule (+0.055R signal-coupled, z = 17.9). Real throughout but roughly halved: +0.126R in 2010-15 → +0.063R in 2023-26, net of costs
LSR — liquidity sweep
sweep of the prior extreme, raw and with 25 / 35 / 50% retracement filters, plus the productised 25% entry
370,000+ −0.012 to −0.101R
fade ≡ continuation within 0.004R — no directional content. Deeper retracement is monotonically worse. The 25% entry loses to random timing at t = −6.55
Strat combos
2-1-2, 2-2 reversal, 3-1-2, 1-2-2, 3-2-2, 1-3 — both directions, intraday to weekly
160,000+ −0.028R to +0.003R
0 of 72 cells cleared correction. 2-1-2 −0.028R (t = −3.31); 2-2 reversal +0.003R (t ≈ 1, noise). Up and down arms are near-identical
ORB — opening range
15, 30 and 60-minute ranges, breakout entry, held to the close
31,478 worse than random
entering at a random time the same session beats the breakout at every range length. Negative after 2bps

Backtest · 5,284 total signals · fills and market impact not modeled · not a verified performance claim.

Signal quality scoring

Score higher. Filter tighter.

Every signal receives a composite score based on structure, momentum, and volatility conditions. The system preferentially surfaces the high-confidence end of the distribution.

All signals
53.8%
Win rate · 1,071 signals
Avg R+0.079R
Profit factor1.20
Score ≥ 7
55.4%
Win rate · 962 signals
Avg R+0.123R
Profit factor1.33
Score ≥ 9 · live threshold
56.8%
Win rate · 896 signals
Avg R+0.155R
Profit factor1.43
How backtests work here

Signal quality first. Numbers second.

Every signal in this research passed the same filters the live system uses — same quality gates, same exclusions. That's a cleaner-than-typical backtest. It's still research.

01

Signal filter

RVOL, FTFC, compression, spread, and structural quality gates — identical to the live system.

02

Data QA

Known data errors excluded. No result counted until fills, closes, and duplicates are clean.

03

Research only

No result gets promoted to a live claim until forward testing proves the setup survives real-market friction.

Signals, not auto-trades

You stay in control.

The Algomist surfaces and tracks confirmed setups — it never touches your brokerage. You decide what to take. The Mobile Trade Companion lets you follow any signal and watch the live R in real time.

See the Trade Companion →
Signal research · not an audited record
04 · The part that counts

A backtest is a hypothesis. The ledger is the test.

Everything above is in-sample — measured carefully, with matched controls and era splits, but still measured on data that already existed. This program has watched a strategy backtest well and then lose money forward: a trend sleeve tested clean and came back −0.598R over 20 live trades (t = −3.36). That is why the backtest is not the claim.

64 yr
Backtest depth · to 1962
+0.091R
Backtested per trade
2026-07-29
Forward test opened
live
Closed trades, published

Signals are written to a public ledger at the close that produces them, before the outcome exists. Results are computed mechanically afterwards and never revised. Every recorded signal is filled — losers included — because choosing which ones count is exactly how forward tests get flattered. The +0.091R above is a full-history average net of costs; the edge is smaller lately — +0.126R across 2010-15 against +0.063R across 2023-26. The recent-era figure is the honest expectation, so plan on roughly +0.06R per trade rather than +0.09R. Anything much larger early on is luck, not vindication.

R is not a floor. From 30 Jul 2026 the rule carries a catastrophe stop 8R below entry — far enough out that it binds on ~0.03% of trades. Across 74,689 backtested trades it cuts the worst trade from −18.2R to −8.0R, costs ~1.2% of expectancy, and leaves maximum drawdown unchanged. A stop is a cap with slippage, not a floor: a gap can fill through it, and a loss can still exceed 1R by a wide margin. Signals recorded before that date were logged without a stop and are filled without one — they are not rewritten.

Read the open ledger — no account needed →

Get access

See the next confirmed signal — live.

The full suite plus the entire Trading RPG. One subscription.

Get access ✦
$25/mo for the radar (20% off) · the automation tool is not on general sale yet
RESEARCH, NOT ADVICE. Every figure on this page is a backtest — a measurement of what a rule would have done on historical data, not a forecast and not a claim about the future. Backtests cannot capture real spreads, slippage or the cost of being wrong. Nothing here is investment advice, a recommendation, or an offer to trade, and no order is placed from any of it. Past behaviour of any strategy does not predict future results. The only forward-looking record we keep is the public forward test, which is reported unrevised, losers included.