Three rules feed the live-test Daily Picks today. Here is exactly what each does, how it was tested, and every idea that did not make it.
Everything below runs as a live test with limits written down before it started — Live tests shows how far each one is.
TF1, PQ1 and LQ1 feed one list; at most two picks a UTC day are sent to Telegram with a size for 0.5 % account risk. Judged after 100 picks and 3 months. TF1 wins about 36 % of trades; its best 5 % made more than all trades together, so it pays only if every signal is taken.
Signals →
Spot long + perpetual short while funding is high: +7.9 % a year after costs on 8 coins it was never built on, positive in 6 of 7 years. Live test since 2026-10-07; judged after 6 months.
Carry →
The same candidates taken by risk limits (2.5 % open in total, 1.5 % in coins that move together) instead of the first two a day.
Portfolio →
Would a market-timing filter (BTC trend, breadth, funding, long-term trend) have picked better trades? Recorded on every candidate, never acted on until 100 forward trades judge it.
Decisions →
More frequent than TF1 and with more winners (about 53 %), but a thin edge: about +0.03R a trade over 1,865 past cases.
Rare (a few a year per coin group) and unproven: its good results came mostly from one year. It is in Daily Picks as a live-test rule and also feeds the crash alert.
Expect few signals (a few per month over all coins) and a win rate below 50 %: most trades are small losses or break-even, and a few trends pay for them.
Three rules feed one list: TF1 (daily trend), PQ1 (tall 4-hour chart patterns) and LQ1 (crash bounce). At most 2 picks per UTC day are sent, first come first served; a day with no pick is a normal result. Each pick shows a size for 0.5 % account risk. In backtests (all live coins, 2020–2026) this averaged +0.06R per pick with flat results since 2025, so it runs as a live test first. TF1 keeps its own record (TF1-v1); signals over the daily limit are tracked and shown as “withheld”.
The rule loses in bear and choppy markets, so publishing has a brake fixed in advance: new signals pause when the live record falls 10R below its peak or after 8 losses in a row, and resume automatically after 30 days. The rule itself never changes; signals found during a pause are still tracked and shown as “withheld”, so no result is hidden. Open trades are not affected.
| Period | Dates | Trades | Average per trade | Note |
|---|---|---|---|---|
| Development (TRAIN) | Jan 2020 – Jun 2023 | 129 | +0.25R | long trades of the rule as first tested |
| Validation | Jul 2023 – Dec 2024 | 81 | +0.54R | judged period; long side |
| Holdout (one look) | Jan 2025 – Jul 2025 | 21 | +0.12R | profit factor 1.33; fewer than the 30 trades required, so not yet validated |
| Recent (after the rules were frozen) | Jul 2025 – Sep 2026 | 31 | −0.22R | 11 winners; losses clustered when many coins broke out and failed together (Jan and May 2026) |
R = result in multiples of the amount risked per trade. Backtests are simulations: they show how the rule behaved, not what it will do. Across both periods after development (holdout and recent, 52 trades) the rule is slightly negative so far (about −0.09R per trade) — which is why every signal stays a live test until the live record proves it. The live record is on the signals page.
The frozen rule on three separate sets of 20 other liquid coins, 2020 – Aug 2026. Set A: 355 trades, +0.14R per trade. Set B: 325 trades, +0.15R per trade. Set C: 318 trades, only +0.02R per trade. Positive on all three, but thin — and look at the years:
| Year | Set A trades | Set A per trade | Set B trades | Set B per trade | Set C trades | Set C per trade |
|---|---|---|---|---|---|---|
| 2020–2021 | 57 | +0.12R | 54 | +0.30R | 46 | +0.29R |
| 2022 (bear market) | 19 | −0.47R | 21 | −0.44R | 28 | −0.51R |
| 2023 | 76 | +0.18R | 79 | +0.46R | 79 | +0.31R |
| 2024 (bull market) | 96 | +0.66R | 95 | +0.32R | 89 | +0.24R |
| 2025 | 62 | −0.11R | 51 | −0.39R | 53 | −0.47R |
| 2026 (to Aug) | 45 | −0.37R | 25 | −0.26R | 23 | −0.62R |
What this means: on 60 coins it was never built on, the rule made money in rising markets (2021, 2023–2024) and lost in every bear or choppy year (2022, 2025–2026) — including the current market. Three filters were tested to avoid those years (market breadth, a cap on open trades, ADX trend strength); none did (see below). Treat signals as unproven, size for a long run of losses, and watch the risk switch and the radar's breadth.
Did the code find real patterns? Judge for yourself: 24 trades of each rule, spaced evenly through 2020–2024 (not picked), drawn on their charts with the swing points the rule knew, the entry, stop, targets and the result.
Patterns, candles, scalps, breakouts, momentum, mean reversion, rotations, an SMC setup and filters for TF1 were tested the same way, each written down before its test. Almost all ended near a coin flip after costs; only PQ1 passed (narrowly) and LQ1 came close — both run as live tests in Daily Picks.
BR14-hour breakout and retest
Validation: −0.08R per trade over 550 validation trades
FR1Crowded-trade reversal (funding, open interest)
Validation: +0.04R over 72 trades — too small and too few
SQ1Volatility squeeze breakout
Validation: −0.11R over 142 trades
TF1 shortDaily trend following, short side
Validation: −0.33R over 46 trades
PL14-hour pullback in an uptrend
Validation: −0.01R over 449 trades
TF24-hour breakout in an uptrend
Validation: +0.13R over 261 trades, but flat in the earlier period — not consistent
SM1Smart Money Concepts: liquidity sweep, break of structure, fair-value-gap entry
Validation: +0.02R over 256 trades on 20 unseen coins — no edge after costs
TF1 v2Daily trend following only when most coins are above their trend line
Validation: +0.10R over 248 trades on 20 unseen coins — below the plain rule; the filter missed the 2025 losses
TF1 v3aDaily trend following with at most 4 trades at risk at once
Validation: +0.02R over 196 trades on 20 unseen coins — half the drawdown, but no better result
TF1 v3bDaily trend following only when the coin's ADX trend strength is rising above 20
Validation: +0.11R over 211 trades on 20 unseen coins — better than the plain rule there, but still lost 2022 and 2025–2026
SC115-minute trend pullback scalp (enter now, long and short)
Validation: Win rate 47 %, −0.14R per trade over 2,137 trades on 40 unseen coins — fees eat the small wins
SC215-minute momentum burst scalp (enter now, long and short)
Validation: Win rate 42 %, −0.14R per trade over 30,797 trades on 40 unseen coins — negative on every coin
PT14-hour chart patterns: double top/bottom, head and shoulders (neckline break)
Validation: Win rate 50 %, −0.01R per trade over 2,480 trades on 20 unseen coins — break-even before costs, loses since 2025
CP1Daily candlestick reversals (engulfing, hammer, shooting star) at the 20-day average in the trend
Validation: Win rate 52 %, +0.02R per trade over 899 trades on 20 unseen coins — too small to beat costs, loses without its best year
CS1Combined confirmations: 4-hour pullback (PL1) only while 58–92 % of coins are above their 100-day average
Validation: Win rate 39 %, 0.00R per trade over 504 trades on 20 unseen coins — positive before 2025 (+0.09R), −0.45R since
MC11-hour momentum continuation on smaller coins (buy a coin already running on heavy volume)
Validation: Win rate 47 %, −0.01R per trade over 136 trades on 20 unseen newer coins — break-even before costs, positive on 7 of 20
MR1Daily mean reversion: buy sharp dips in uptrends, short sharp spikes in downtrends (RSI 2)
Validation: Win rate 49 %, −0.11R per trade over 342 trades on 20 unseen coins (2025–2026) — a clear loss, positive on 4 of 20
VB1Daytime volatility breakout: enter when price moves half of yesterday's range from today's open, with the trend
Validation: Win rate 48 %, −0.02R per trade over 4,701 trades on 20 unseen coins — no edge before or after 2025; costs make it a loss
LX1Exit rules to cut TF1's losses early (exit on a failed breakout, on no progress, or at break-even)
Validation: None beat TF1 itself on its development trades (+32.0R): 16 of 47 big winners looked like failed breakouts in their first days
XS1Weekly relative-strength rotation: hold the 3 coins with the best 28-day return (if rising)
Validation: Win rate 40 %, −0.07R per trade over 476 trades on 19 unseen coins — promising on the development coins (+0.11R in 2023–24), a clear loss on new ones
LQ1Buy liquidation crashes: a fall of 1.5 daily ranges in 4 hours while open interest drops 8 %
Validation: Win rate 65 %, +0.25R per trade over 52 trades on 20 new coins (2021–26) — but 2023 alone made the profit; without it −3.8R. Did not pass; since 2026-10-03 it runs in the live-test Daily Picks as an unproven rule, and as the crash alert
PT2Chart patterns with breakout confirmation or trade management (retest entry, neckline exit, break-even, partial, trailing)
Validation: Retest entry +0.02R a trade better but misses the best breakouts; break-even, trailing and partial exits add nothing — taking half at +0.5R lifts the win rate to 67 % yet loses more
PQ1Chart patterns, tall ones only (height at least 2.8 × the 4-hour ATR)
Validation: Passed narrowly: +0.15R in 2023–24, then +0.04R over 197 trades in 2025–26 (56 % winners) — but 2026 is negative so far. Runs in the live-test Daily Picks; not yet proven live
PQ1-1hThe same tall chart patterns on 1-hour bars instead of 4-hour
Validation: −0.03R per trade over 6,730 trades on all live coins (2021–26), profit factor 0.93 — more trades, smaller moves, and costs win. Shorter timeframes are not tested further
MTF1Multi-timeframe chain: daily bias → 4-hour pullback → 1-hour confirmation → 30-minute refinement → 15-minute trigger (four variants)
Validation: All four variants lost: −0.04R to −0.06R per trade over 388–595 trades in 2023–24 (48 % winners). Extra timeframes trimmed losses but added no edge — stopped before the unseen-coin test, as registered
MTF2The daily trend rule (TF1) with its entry timed on 1-hour and 15-minute bars (wait for a dip, then a break)
Validation: +0.11R per trade over 632 trades on 40 unseen coins — but plain TF1 made +0.14R on the same signals. Waiting for a better price misses the best breakouts; TF1 keeps its simple entry
BX1Channel-style range breakout: quiet 2-day range, volume breakout, wait for the hold, stop below the range
Validation: Win rate 44 %, −0.03R per trade over 829 trades on 60 small coins — a coin flip; 2025 −0.15R, 2026 +0.07R so far
RH1BTC and ETH held on spot while above their 200-day average, cash below
Validation: +1,598 % for the 50/50 pair since July 2020 (buy-and-hold +852 %), but a 53 % drawdown against a 46 % limit fixed in advance — it exits only after a fast crash is mostly done and whipsaws around the average. Failed