Transparency · CTS Decipher

Agent Performance

Agent performance replays CTS Decipher's exact qualification gates over 150 days across every tradable coin: every eligible trade, its odds of reaching +1/2/3/4/5% in 5% conviction bands, how far the coin really runs before the stop, and the net return and liquidation risk on your capital from 1× up to 10× leverage — plus a data-driven read on the best entry, exit and leverage to arm it with.

Replaying the ledger since May 27, 2026 · 342 signals recorded · 270 resolved · 72 live

Backtested rule-replay, not profit. The agent has almost no live fills yet, so this page shows what its rules would have done: a 150-day path-aware backtest across every tradable coin, corroborated by the live forward-recorded ledger. Win rates are first-passage to a target, not realised P&L; a liquidation is a −100% trade, and after fees the net edge is roughly break-even even at 1×. A research and configuration tool, not a promise of returns.

The 150-day agent trade study

Backtested · rolling 150-day window · data as of Jul 4, 2026

Over the last 150 days we replayed the live engine and the agent's gates across all 195 tradable coins. 1,160 distinct signals fired; 216 passed the agent's structural gates. Below is every eligible trade, its odds of reaching each target, how that varies by conviction, strategy and regime, and — the part that matters for real money — the net return and liquidation risk at 1×, 5× and 10× leverage.

How it's measured. Entry = the next daily bar's open after the signal; stop = the engine's own ATR stop (identical at every leverage). The price path is replayed bar by bar with a conservative intrabar rule — the adverse extreme is assumed to hit before the favourable one within a bar, so these are a floor, never a look-ahead best case. Max hold 15 days. Leverage only scales P&L and sets the liquidation distance (≈ 1/leverage − 0.5% maintenance); a liquidation is a −100% trade. Round-trip costs 0.15% of notional (so ×leverage on capital). Same audited engine as the live signals — reconciled against the live ledger below.
216Agent-eligible tradesSignals over the window that passed the agent's structural gates ($5M liquidity, Trap ≤ 6, first target ≥ 2R, regime aligned).
207ResolvedFinished trades used in the rate tables. The rest are still inside their hold window.
89%Reached +2%Share of resolved eligible trades whose coin price touched +2% before the stop. A price move, not profit on your capital.
85%Reached +3%Share of resolved eligible trades whose coin price touched +3% before the stop. A price move, not profit on your capital.
76%Reached +4%Share of resolved eligible trades whose coin price touched +4% before the stop. A price move, not profit on your capital.
71%Reached +5%Share of resolved eligible trades whose coin price touched +5% before the stop. A price move, not profit on your capital.
+0.06%EV/trade · 1×Expected net return per trade on your capital at a +2% exit, 1× leverage (no leverage — the honest baseline).
+0.19%EV/trade · 3×Expected net return per trade on your capital at a +2% exit, 3× leverage (full capital; liquidation ≈ −33% away).
+0.68%EV/trade · 5×Expected net return per trade on your capital at a +2% exit, 5× leverage (full capital; liquidation ≈ −20% away).
−2.10%EV/trade · 7×Expected net return per trade on your capital at a +2% exit, 7× leverage (full capital; liquidation ≈ −14% away).
−3.74%EV/trade · 10×Expected net return per trade on your capital at a +2% exit, 10× leverage (full capital; liquidation ≈ −10% away).

The odds, by conviction (5% bands)coin price move

For every eligible trade, the probability its price reached +1/+2/+3/+4/+5% before the stop, plus the typical peak and heat — in 5% conviction bands so you can see exactly where to set your gate. These are first-passage rates on the coin's price, not profit on your capital.

Reached +2% before the stop · by conviction band
[25, 30)93%n=15
[30, 35)95%n=22
[35, 40)100%n=16
[40, 45)100%n=6
[45, 50)89%n=9
[50, 55)60%n=5
[55, 60)89%n=19
[60, 65)85%n=20
[65, 70)73%n=26
[70, 75)88%n=16
[75, 80)94%n=18
[80, 85)80%n=15
[85, 90)100%n=12
[90, 95)100%n=7
[95, 100]100%n=1
ConvictionTrades+1%+2%+3%+4%+5%Med peakMed heatStop%
[25, 30)1593%93%93%73%67%+8.5%11.7%67%
[30, 35)2295%95%86%68%64%+8.9%13.0%59%
[35, 40)16100%100%94%94%81%+16.3%10.9%44%
[40, 45)6 *100%100%100%100%100%+11.8%12.9%50%
[45, 50)9 *89%89%89%89%89%+10.8%14.7%56%
[50, 55)5 *80%60%60%60%60%+7.9%13.4%100%
[55, 60)1989%89%89%84%63%+8.2%16.8%42%
[60, 65)2095%85%80%70%70%+10.9%13.1%30%
[65, 70)2685%73%69%50%42%+4.0%17.1%35%
[70, 75)1694%88%88%81%81%+12.3%16.1%50%
[75, 80)18100%94%94%94%89%+14.0%14.7%28%
[80, 85)1580%80%67%67%67%+14.2%14.5%20%
[85, 90)12 *100%100%92%83%83%+12.1%25.5%50%
[90, 95)7 *100%100%86%71%71%+12.2%17.8%29%
[95, 100]1 *100%100%100%100%100%+7.2%32.2%100%
All eligible20793%89%85%76%71%+10.3%14.4%44%

Buckets are lower-inclusive, upper-exclusive — e.g. [60, 65) holds conviction from 60% up to but not including 65%; only the top band [95, 100] is closed. * = small sample (<15), directional only. Rates are over resolved trades; “reached +X%” means the peak touched +X% before the stop.

The coin's move, then your capital & liquidation

Two different questions, kept separate. First: how far does the coin actually travel in our favour (below, leverage-independent)? Then: what does that do to your capital once leverage and the liquidation price are counted? The agent deploys full capital, so leverage is the real risk dial — same trades, same stop, but higher leverage puts the liquidation price closer and a liquidation is a total (−100%) loss.

How far the coin moves in our favour before the stopcoin price move
reached +1%93%
reached +2%89%
reached +3%85%
reached +4%76%
reached +5%71%
reached +7%63%
reached +10%51%

Median favourable peak +10.3%, top-quartile peak +20.5%, against a median heat of 14.4% (n=207). This is the coin's price move only, before any leverage — and because the agent trails its stop rather than exiting at a fixed target, realised moves frequently run well past any single take-profit (or give some back to the trail).

Now apply leverage. Pick a fixed take-profit and read the net return on your capital at each leverage. Remember the live agent trails its stop rather than exiting at a fixed target — so treat these fixed-take-profit rows as a “what-if you took profit here,” not the agent's real exit.

Return on your capital = coin move % × leverage − feesLiquidation (coin ≈ 1/leverage against you) = −100%, whatever the stop does
“Reach” = the coin's price move (leverage-independent)“EV / return” = net on your capital, after fees
Fixed take-profit target
1×liq ≈ −99.5%
EV / trade (on capital)+0.06%
Reached +2% before stop/liq88.9%
Liquidated0.0%
Median return+1.85%
3×liq ≈ −32.8%
EV / trade (on capital)+0.19%
Reached +2% before stop/liq88.9%
Liquidated0.0%
Median return+5.55%
5×liq ≈ −19.5%
EV / trade (on capital)+0.68%
Reached +2% before stop/liq88.9%
Liquidated2.4%
Median return+9.25%
7×liq ≈ −13.8%
EV / trade (on capital)−2.10%
Reached +2% before stop/liq85.5%
Liquidated11.1%
Median return+12.95%
10×liq ≈ −9.5%
EV / trade (on capital)−3.74%
Reached +2% before stop/liq80.7%
Liquidated18.4%
Median return+18.50%
Expected value per trade (on capital) at a +2% exit
1× leverage+0.06%
3× leverage+0.19%
5× leverage+0.68%
7× leverage−2.10%
10× leverage−3.74%
Leverage
ConvictionTradesReach +2%LiqEV / tradeMedian
[25, 30)1593%0%+1.1%+1.8%
[30, 35)2295%0%+1.6%+1.8%
[35, 40)16100%0%+1.8%+1.8%
[40, 45)6 *100%0%+1.8%+1.8%
[45, 50)9 *89%0%−0.4%+1.8%
[50, 55)5 *60%0%−3.6%+1.8%
[55, 60)1989%0%−0.0%+1.8%
[60, 65)2085%0%−1.3%+1.8%
[65, 70)2673%0%−3.2%+1.8%
[70, 75)1688%0%−0.4%+1.8%
[75, 80)1894%0%+1.6%+1.8%
[80, 85)1580%0%−0.5%+1.8%
[85, 90)12 *100%0%+1.8%+1.8%
[90, 95)7 *100%0%+1.8%+1.8%
[95, 100]1 *100%0%+1.8%+1.8%
All eligible20789%0%+0.1%+1.8%

Showing 1× leverage at a +2% take-profit (liquidation ≈ −99.5% away). Same fixed stop price at every leverage — only the liquidation distance changes. Reach is the coin's price hitting +2% before the stop or liquidation; EV and Median are net return on your capital after fees, and a liquidation is a −100% trade. Conservative intrabar (the adverse extreme is assumed to hit first), so these are a floor, not a look-ahead best case.

By strategy, regime & timeframe

The same eligible trades cut by the setup that fired them — with EV at each leverage, so the strongest configurations to point your filters at are obvious.

Strategy · timeframe

StrategyTradesReach +2%Med peakMed heatStop%EV 1×EV 10×
Breakout · Daily (1d)11185%+10.6%17.3%34%−0.6%−13.5%
Trend · 4-hour (4h)6197%+11.5%12.3%56%+1.6%+7.9%
Trend · Daily (1d)3391%+8.2%13.9%55%−0.3%+7.7%
Mean-reversion · Daily (1d)2 *50%+7.9%1.7%50%−0.6%−6.3%

Market regime at signal time

RegimeTradesReach +2%Med peakMed heatStop%EV 1×EV 10×
High-volatility14488%+11.4%16.1%37%−0.1%−9.8%
Trend-down25100%+8.0%8.5%32%+1.9%+18.5%
Trend-up2085%+8.4%17.2%95%−1.8%+0.7%
Range1889%+7.7%10.4%61%+0.9%+9.2%

Timeframe

TimeframeTradesReach +2%Med peakMed heatStop%EV 1×EV 10×
Daily (1d)14686%+9.9%16.1%39%−0.6%−8.6%
4-hour (4h)6197%+11.5%12.3%56%+1.6%+7.9%

What the data says

The leverage truth
At a +2% take-profit the expected value per trade is +0.06% at 1×, +0.68% at 5×, and −3.74% at 10×. Expected value turns negative at 10× — liquidation on full capital eats the edge. Higher leverage is not more profit; it is more ruin.
Liquidation risk
Deploying full capital, roughly 18% of trades hit liquidation at 10× and 2% at 5× (versus 0% at 1×, where the stop is always nearer than the liquidation price). One liquidation is a −100% trade — it takes many winners to recover.
Honest baseline
Even at 1× — no liquidation risk — the edge is close to break-even after fees (+0.06% per trade at a +2% exit). The agent's value is disciplined, manipulation-aware entries and a wide trailing stop, not a return multiplier.
Where to set conviction
Among adequately-sampled bands, conviction [35, 40) reached +2% before the stop 100% of the time (n=16). But read the whole curve — the link between conviction and win rate is uneven, so a very high gate also thins your sample and excludes lower-conviction 4-hour setups.
Strongest setup
Trend · 4-hour (4h) is the strongest adequately-sampled cut — 97% reached +2% before the stop (n=61), typical peak +11.5% against −12.3% heat. Point your strategy filter here.
Regime matters
Trades in a trend-down regime reached +2% 100% of the time versus 85% in trend-up — which is exactly why the agent requires the trade to align with the trend.

What to arm — the best insights

The essence, read straight from the study above and refreshed as the window rolls forward. Each card maps to a control on the Decipher arming consoleStrategy filter, Timeframe filter, Minimum conviction, Mode (Balanced ~15% / Aggressive ~40% risked per stop) and the Leverage override. This is guidance for configuring the agent, never a promise of profit — after fees the edge is thin, and leverage is a risk cap, not a return multiplier.

Best entry
Trend · 4-hour (4h)
Reached +2% before the stop 97% of the time (n=61), with a typical peak of +11.5% against −12.3% heat. Point your strategy / timeframe filter here.
Conviction lever
A weak standalone lever
The conviction curve is uneven, not a clean climb (the best-sampled band is [35, 40) at 100%, but low bands score similarly). Don't crank the gate — lean on strategy + regime, and note a high gate also excludes lower-conviction 4-hour setups.
Best exit
Take profit near +3%
Of the fixed exits modelled, +3% carries the best expectancy at 1× (+0.19% per trade, reached 85% of the time). But the live agent trails its stop, so realised moves often run past any fixed target — median peak is +10.3%, the top quartile +20.5%.
Best leverage
Cap near 2× — size fractionally
Per-trade EV at a +2% exit stays positive up to 5× (+0.68%) and turns negative at 7× — but that is arithmetic, not survival. On full capital, repeating any high-leverage trade compounds to ruin: one liquidation is a −100% you can’t win back. Keep the multiplier low and make fractional position sizing the real risk lever.
Suggested starting presetStrategy filter → Trend · Timeframe → 4-hour (4h) · Min conviction → a moderate floor (not the lever) · Mode → Balanced (~15% risk/stop) · Leverage override → ≤ 2×

To improve the agent & the signal engine — where the data says our own edge is weak.

Conviction ↔ follow-through
Higher conviction did track higher follow-through (80% at [80, 85) vs 93% at [25, 30)) — the conviction score is carrying real signal in this window; keep it as an entry gate.
Weakest regime
Trend-up setups only reached +2% 85% of the time (n=20) — the engine should down-weight or veto entries here, which is exactly what the agent's regime-alignment gate already starts to do.
Weakest setup
Breakout · Daily (1d) is the weakest adequately-sampled cut (85% to +2%, n=111). A candidate to tighten or retire in the next engine iteration.

Every eligible trade

Every resolved eligible trade over the window, newest first — with the fired date/time (UTC & IST), conviction, regime, setup, excursions, the coin's reach to +1/2/3/4/5%, and the net return on your capital at each leverage if exited at +2% (⚡ = liquidated). These are the study's backtested, simulated eligible trades; the agent's real on-exchange fills — and every future one, logged automatically — are in Trades the agent has actually taken below.

Fired (UTC / IST)CoinDirConvRegimeSetupStopPeakHeat+1/2/3/4/5StatusHeld1×3×5×7×10×
Jun 28, 2026, 00:00 UTCJun 28, 2026, 05:30 ISTSUIShort34%Trend↓Trend · 4h11.1%+2.0%13.9%✓ ✓ · · ·stopped6d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 17, 2026, 00:00 UTCJun 17, 2026, 05:30 ISTWLDLong86%High-volBreakout · 1d41.6%+7.1%47.4%✓ ✓ ✓ ✓ ✓stopped15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 16, 2026, 00:00 UTCJun 16, 2026, 05:30 ISTTRXShort32%High-volTrend · 4h3.2%+0.6%5.0%· · · · ·stopped8d−3.4%−10.1%−16.8%−23.5%−33.5%
Jun 16, 2026, 00:00 UTCJun 16, 2026, 05:30 ISTINJLong56%Trend↑Trend · 1d16.1%+9.5%19.2%✓ ✓ ✓ ✓ ✓stopped8d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 16, 2026, 00:00 UTCJun 16, 2026, 05:30 ISTZECLong58%Trend↑Trend · 1d19.0%+3.0%20.8%✓ ✓ ✓ · ·stopped8d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 15, 2026, 00:00 UTCJun 15, 2026, 05:30 ISTBABYLong74%Trend↑Trend · 1d18.0%+11.6%18.8%✓ ✓ ✓ ✓ ✓stopped4d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 14, 2026, 00:00 UTCJun 14, 2026, 05:30 ISTTAOLong62%Trend↑Trend · 1d12.3%+11.0%12.4%✓ ✓ ✓ ✓ ✓stopped5d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 12, 2026, 00:00 UTCJun 12, 2026, 05:30 ISTONDOLong46%Trend↑Trend · 1d17.3%+7.2%19.7%✓ ✓ ✓ ✓ ✓stopped13d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 11, 2026, 00:00 UTCJun 11, 2026, 05:30 ISTCHZShort28%High-volTrend · 4h19.6%+36.5%15.5%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%⚡−100%⚡−100%
Jun 11, 2026, 00:00 UTCJun 11, 2026, 05:30 ISTZBTShort35%High-volTrend · 4h23.4%+4.4%23.6%✓ ✓ ✓ ✓ ·stopped2d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 10, 2026, 00:00 UTCJun 10, 2026, 05:30 ISTXRPShort30%Trend↓Trend · 4h7.8%+4.3%13.7%✓ ✓ ✓ ✓ ·stopped6d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 10, 2026, 00:00 UTCJun 10, 2026, 05:30 ISTTAOShort28%High-volTrend · 4h13.9%+4.1%34.5%✓ ✓ ✓ ✓ ·stopped4d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 10, 2026, 00:00 UTCJun 10, 2026, 05:30 ISTDASHShort33%High-volTrend · 4h15.9%+8.9%14.4%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 10, 2026, 00:00 UTCJun 10, 2026, 05:30 ISTSAHARAShort83%High-volBreakout · 1d97.0%+39.7%8.6%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 9, 2026, 00:00 UTCJun 9, 2026, 05:30 ISTFETShort55%Trend↓Trend · 1d17.7%+19.3%6.5%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 8, 2026, 00:00 UTCJun 8, 2026, 05:30 ISTBABYLong64%Trend↑Trend · 1d20.3%+18.4%20.3%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTTRXShort83%High-volBreakout · 1d8.8%+3.0%2.9%✓ ✓ · · ·expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTFILShort79%High-volBreakout · 1d43.9%+8.8%13.3%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTICPShort57%Trend↓Trend · 1d19.4%+6.7%14.0%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTHBARShort55%Trend↓Trend · 1d10.1%+4.7%5.6%✓ ✓ ✓ ✓ ·expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTFETShort60%Trend↓Trend · 1d20.1%+7.7%12.8%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTWIFShort76%High-volBreakout · 1d38.1%+8.8%17.6%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTTAOShort74%High-volBreakout · 1d35.2%+6.2%42.1%✓ ✓ ✓ ✓ ✓stopped8d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTASTERShort55%Trend↓Trend · 1d11.1%+4.4%30.1%✓ ✓ ✓ ✓ ·stopped12d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTPOLShort64%High-volBreakout · 1d22.7%+8.5%7.7%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTPUMPShort89%High-volBreakout · 1d40.6%+7.7%17.4%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTPENGUShort78%High-volBreakout · 1d36.4%+6.9%20.0%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTXPLShort86%High-volBreakout · 1d57.9%+10.2%84.2%✓ ✓ ✓ ✓ ✓stopped12d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTZENShort90%High-volBreakout · 1d41.4%+7.5%19.3%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 6, 2026, 00:00 UTCJun 6, 2026, 05:30 ISTBABYLong46%Trend↑Trend · 1d17.9%+0.0%21.4%· · · · ·stopped1d−18.1%−54.3%−90.4%⚡−100%⚡−100%
Jun 5, 2026, 00:00 UTCJun 5, 2026, 05:30 ISTDOTShort60%High-volBreakout · 1d24.4%+14.8%1.0%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 5, 2026, 00:00 UTCJun 5, 2026, 05:30 ISTSEIShort43%Trend↓Trend · 1d14.9%+14.8%9.3%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 5, 2026, 00:00 UTCJun 5, 2026, 05:30 ISTCHZShort92%High-volBreakout · 1d47.2%+22.9%14.4%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 5, 2026, 00:00 UTCJun 5, 2026, 05:30 ISTONDOShort36%High-volTrend · 4h17.2%+15.5%5.9%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 5, 2026, 00:00 UTCJun 5, 2026, 05:30 ISTZECShort92%High-volBreakout · 1d51.3%+45.6%18.5%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 4, 2026, 00:00 UTCJun 4, 2026, 05:30 ISTBNBShort55%Trend↓Trend · 1d7.0%+10.3%2.0%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 4, 2026, 00:00 UTCJun 4, 2026, 05:30 ISTWLDLong80%High-volBreakout · 1d36.6%+33.5%27.7%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%⚡−100%⚡−100%
Jun 4, 2026, 00:00 UTCJun 4, 2026, 05:30 ISTGPSShort45%RangeTrend · 4h17.2%+8.5%20.5%✓ ✓ ✓ ✓ ✓stopped4d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 3, 2026, 00:00 UTCJun 3, 2026, 05:30 ISTONDOLong34%High-volTrend · 4h14.9%+11.5%17.4%✓ ✓ ✓ ✓ ✓stopped3d+1.9%+5.5%+9.3%+12.9%+18.5%
Jun 2, 2026, 00:00 UTCJun 2, 2026, 05:30 ISTWLDLong82%High-volBreakout · 1d33.6%+56.9%14.0%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%⚡−100%⚡−100%
Jun 1, 2026, 00:00 UTCJun 1, 2026, 05:30 ISTTONLong47%Trend↑Trend · 1d13.7%+20.4%13.9%✓ ✓ ✓ ✓ ✓stopped4d+1.9%+5.5%+9.3%+12.9%+18.5%
May 31, 2026, 00:00 UTCMay 31, 2026, 05:30 ISTBNBLong67%High-volBreakout · 1d10.8%+3.8%14.3%✓ ✓ ✓ · ·stopped4d+1.9%+5.5%+9.3%+12.9%+18.5%
May 31, 2026, 00:00 UTCMay 31, 2026, 05:30 ISTFETLong73%High-volBreakout · 1d26.5%+3.5%34.5%✓ ✓ ✓ · ·stopped6d+1.9%+5.5%+9.3%+12.9%+18.5%
May 31, 2026, 00:00 UTCMay 31, 2026, 05:30 ISTSTGLong49%Trend↑Trend · 1d12.7%+84.2%14.7%✓ ✓ ✓ ✓ ✓stopped7d+1.9%+5.5%+9.3%+12.9%+18.5%
May 30, 2026, 00:00 UTCMay 30, 2026, 05:30 ISTXLMLong57%High-volBreakout · 1d20.8%+14.9%23.7%✓ ✓ ✓ ✓ ✓stopped6d+1.9%+5.5%+9.3%+12.9%⚡−100%
May 30, 2026, 00:00 UTCMay 30, 2026, 05:30 ISTINJLong84%High-volBreakout · 1d29.8%+14.2%25.7%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
May 28, 2026, 00:00 UTCMay 28, 2026, 05:30 ISTPENDLEShort72%Trend↓Trend · 1d16.4%+27.0%0.7%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
May 26, 2026, 00:00 UTCMay 26, 2026, 05:30 ISTTONLong54%Trend↑Trend · 1d15.0%+17.7%15.8%✓ ✓ ✓ ✓ ✓stopped10d+1.9%+5.5%+9.3%+12.9%+18.5%
May 26, 2026, 00:00 UTCMay 26, 2026, 05:30 ISTNEARLong88%High-volBreakout · 1d22.7%+11.4%34.2%✓ ✓ ✓ ✓ ✓stopped11d+1.9%+5.5%+9.3%+12.9%+18.5%
May 26, 2026, 00:00 UTCMay 26, 2026, 05:30 ISTWLDLong57%High-volBreakout · 1d26.5%+91.1%16.8%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
May 26, 2026, 00:00 UTCMay 26, 2026, 05:30 ISTEIGENLong79%High-volBreakout · 1d29.5%+7.5%38.9%✓ ✓ ✓ ✓ ✓stopped11d+1.9%+5.5%+9.3%+12.9%+18.5%
May 24, 2026, 00:00 UTCMay 24, 2026, 05:30 ISTINJLong35%High-volTrend · 4h13.3%+37.5%10.5%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
May 24, 2026, 00:00 UTCMay 24, 2026, 05:30 ISTONDOLong41%High-volTrend · 4h14.7%+6.1%17.6%✓ ✓ ✓ ✓ ✓stopped5d+1.9%+5.5%+9.3%+12.9%+18.5%
May 24, 2026, 00:00 UTCMay 24, 2026, 05:30 ISTZECLong32%Trend↑Trend · 4h13.7%+8.9%14.4%✓ ✓ ✓ ✓ ✓stopped4d+1.9%+5.5%+9.3%+12.9%+18.5%
May 24, 2026, 00:00 UTCMay 24, 2026, 05:30 ISTDASHLong28%High-volTrend · 4h14.6%+3.1%16.5%✓ ✓ ✓ · ·stopped5d+1.9%+5.5%+9.3%+12.9%+18.5%
May 23, 2026, 00:00 UTCMay 23, 2026, 05:30 ISTSOLShort32%RangeTrend · 4h6.0%+28.8%3.7%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
May 23, 2026, 00:00 UTCMay 23, 2026, 05:30 ISTAVAXShort32%RangeTrend · 4h6.5%+32.0%4.8%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
May 23, 2026, 00:00 UTCMay 23, 2026, 05:30 ISTDOTShort27%RangeTrend · 4h8.1%+29.8%6.3%✓ ✓ ✓ ✓ ✓expired15d+1.9%+5.5%+9.3%+12.9%+18.5%
May 23, 2026, 00:00 UTCMay 23, 2026, 05:30 ISTFILShort32%RangeTrend · 4h10.5%+5.3%14.3%✓ ✓ ✓ ✓ ✓stopped5d+1.9%+5.5%+9.3%+12.9%+18.5%
May 23, 2026, 00:00 UTCMay 23, 2026, 05:30 ISTVIRTUALShort27%High-volTrend · 4h12.0%+5.1%19.8%✓ ✓ ✓ ✓ ✓stopped4d+1.9%+5.5%+9.3%+12.9%+18.5%
Showing 60 of 207 resolved eligible trades (9 more still inside their hold window, excluded).

Live forward record

A second, independent lens: our forward-recorded live ledger — every signal logged the moment it fired since May 27, 2026 (342 recorded, 270 resolved), then replayed through the agent's gates. Not backtested — this is the real, recent record, and it corroborates the study above.

Arming Explorer
Set the parameters you would arm the agent with. The funnel and every table below recompute instantly to show which signals qualify and how they historically behaved.
Minimum conviction
Strategy filter
Timeframe filter

Which signals qualify for the agent

Every recorded signal, filtered by the agent's gates in order. Only setups that clear all of them are handed to the agent — which is why so few survive. Each stage is tagged by how faithfully we can replay it from the ledger.

Qualification funnel · newest data
All recorded signalsexact342 100%
Trap Score ≤ 6exact340 99%−2
Conviction gateexact124 36%−216
Strategy filterexact124 36%
Timeframe filterexact124 36%
Sane stop existsexact124 36%
Regime alignedapprox13 4%−111
Composite quality ≥ 55approx13 4%
Re-entry guardexact10 3%−3
At these settings, 10 of 342 recorded signals would have cleared every replayable gate — 2.9% of all signals. The live agent then picks the single best of the qualifiers each cycle.

exact reconstructed from stored data · approx uses a documented proxy (regime: only the 1d timeframe was stored on older signals, a conservative lower bound; quality: the small liquidity term is treated as 0 without historical volume). New signals now store the full per-timeframe regime and volume, so the funnel becomes exact going forward.

Gates the live agent also applies

Three more gates run live but cannot be reconstructed from past signals, so they are shown here as context rather than thinning the analysed pool.

24h volume ≥ $5M
A liquidity floor so the edge is not slipped in a thin book. No volume was recorded historically — new signals now store it.
First target R ≥ 2
When a fixed take-profit exists it must be ≥ 2R. The validated edge trails a stop with usually no fixed target, so this rarely binds.
Fresh entry only
Live, the agent enters only a setup that activated after you armed it and within hours — never a stale, already-moving one.

Every eligible trade — live & auto-updating

The autonomous record: every signal that clears the agent's structural gates (Trap ≤ 6, regime-aligned, a sane stop) in the forward-recorded ledger, plus every trade the agent actually took (marked Taken— a taken trade is eligible by definition). Driven by the live tracker, so new signals and new trades land here automatically, with no manual step. This is the live counterpart to the backtested “every eligible trade” in the study above.

Eligible ≠ taken — this is a candidate pool, not a buy list. A signal appears here the moment it clears the structural bar, but the agent only actually opens a position when all of these also hold: you have it armed and running; the signal activated after you armed it (it never chases a stale, already-moving setup — the fresh-entry rule); it also clears your minimum conviction, the composite quality score ≥ 55, the $5M 24h-volume floor and the re-entry guard; and it was the single best fresh pick that cycle with capital free. So a signal can be eligible here and never be taken — most often because it fired before the agent was armed, or wasn't the freshest, highest-quality pick at that moment.

Fired (UTC)CoinDirConvRegimeSetupPeakHeat+1/2/3StatusHeldAgent
Jun 25, 2026AAVELong78%High-volBreakout · 1d+23.5%3.6%✓ ✓ ✓openTaken
Jun 22, 2026MMTLong75%High-volBreakout · 1d+28.0%10.5%✓ ✓ ✓openTaken
Jun 21, 2026EIGENLong58%High-volBreakout · 1d+25.3%23.1%✓ ✓ ✓openPaper
Jul 4, 2026IOLong57%Trend↑Trend · 1d+1.5%2.2%✓ · ·open
Jul 3, 2026ALLOLong67%Trend↑Trend · 1d+14.6%5.1%✓ ✓ ✓open
Jul 2, 2026ZAMAShort34%Trend · 4h+0.1%4.6%· · ·stopped15.0h
Jul 2, 2026DOTShort32%Trend↓Trend · 4h+0.0%3.5%· · ·stopped12.0h
Jul 2, 2026NEOShort32%Trend↓Trend · 4h+0.4%3.4%· · ·stopped13.0h
Jul 1, 2026SKLShort36%Trend↓Trend · 4h+2.6%4.6%✓ ✓ ·stopped23.0h
Jul 1, 2026CHZShort32%Trend↓Trend · 4h+3.6%5.4%✓ ✓ ✓stopped3d
Jul 1, 2026RIFLong63%Trend↑Trend · 1d+59.2%5.7%✓ ✓ ✓open
Jun 30, 2026AShort29%Trend↓Trend · 4h+0.0%4.6%· · ·stopped30m
Jun 30, 2026AShort29%Trend↓Trend · 4h+0.0%6.0%· · ·stopped40m
Jun 30, 2026AShort29%Trend↓Trend · 4h+0.0%5.1%· · ·stopped50m
Jun 30, 2026DGBShort32%Trend↓Trend · 4h+4.6%3.8%✓ ✓ ✓stopped3d
Jun 30, 2026LINEAShort28%Trend↓Trend · 4h+4.1%4.4%✓ ✓ ✓stopped34.0h
Jun 30, 2026FLOKIShort37%Trend↓Trend · 4h+3.4%4.1%✓ ✓ ✓stopped2d
Jun 30, 2026EGLDShort32%Trend↓Trend · 4h+2.8%4.3%✓ ✓ ·stopped34.0h
Jun 30, 2026ARBShort34%Trend↓Trend · 4h+2.3%4.0%✓ ✓ ·stopped20.0h
Jun 30, 2026ETHShort35%Trend↓Trend · 4h+2.3%3.3%✓ ✓ ·stopped41.0h
Jun 30, 2026AShort32%Trend↓Trend · 4h+2.0%5.1%✓ · ·stopped9.0h
Jun 30, 2026XRPShort30%Trend↓Trend · 4h+2.2%3.5%✓ ✓ ·stopped2d
Jun 30, 2026LINKShort37%Trend↓Trend · 4h+2.9%3.0%✓ ✓ ·stopped41.0h
Jun 29, 2026ANKRShort33%Trend↓Trend · 4h+5.5%3.7%✓ ✓ ✓stopped4d
Jun 29, 2026ROBOShort29%Trend · 4h+15.0%4.4%✓ ✓ ✓open
Jun 28, 2026VETShort30%Trend↓Trend · 4h+5.2%3.8%✓ ✓ ✓stopped4d
Jun 28, 2026MANTALong32%Trend↑Trend · 4h+83.0%6.4%✓ ✓ ✓stopped27.0h
Jun 28, 2026XECShort32%Trend↓Trend · 4h+0.0%26.9%· · ·stopped13.0h
Jun 28, 2026PENGUShort30%Trend↓Trend · 4h+5.0%5.8%✓ ✓ ✓stopped4d
Jun 27, 2026SUIShort34%Trend↓Trend · 4h+2.1%5.4%✓ ✓ ·stopped4d
Jun 27, 2026BLURShort32%Trend↓Trend · 4h+1.0%6.3%✓ · ·stopped17.0h
Jun 27, 2026KAITOLong55%Trend↑Trend · 1d+25.0%0.2%✓ ✓ ✓open
Jun 26, 2026BNTShort30%Trend↓Trend · 4h+9.1%2.6%✓ ✓ ✓open
Jun 26, 2026ALLOLong27%Trend↑Trend · 4h+0.0%20.7%· · ·stopped50m
Jun 26, 2026ALLOLong27%Trend↑Trend · 4h+0.0%13.0%· · ·stopped3.0h
Jun 25, 2026OPGShort35%Trend · 4h+26.8%5.3%✓ ✓ ✓open
Jun 25, 2026MANAShort62%Trend↓Trend · 1d+9.4%11.8%✓ ✓ ✓stopped8d
Jun 23, 2026POLShort34%Trend↓Trend · 4h+12.8%1.7%✓ ✓ ✓open
Jun 23, 2026MATICShort34%Trend↓Trend · 4h+12.8%1.7%✓ ✓ ✓open
Jun 23, 2026SUIShort32%Trend↓Trend · 4h+5.2%4.3%✓ ✓ ✓stopped4d
Showing 40 of 102 eligible trades · 3 the agent actually took (marked).

Shown independent of the arming filters (the full structural-eligible set). Regime on older rows uses the 1d proxy; a trade the agent took that the proxy would mark unaligned still appears — correctly — because the live agent accepted it on the full multi-timeframe regime. Signals recorded after the enrichment migration are exact.

How the qualifying setups performed

Across the 124 setups that pass your trap, conviction, strategy and timeframe gates (90 resolved), first-passage to each take-profit target and the typical excursion. These are first-passage rates, not trade P&L — the agent trails its stop and pays fees, so the net record is roughly break-even.

59%Reached +2%Share of resolved qualifying signals that reached +2% before the stop.
57%Reached +3%Share of resolved qualifying signals that reached +3% before the stop.
+4.4%Median peakTypical furthest favourable move (median MFE) among resolved qualifying signals.
12.1%Median heatTypical furthest move against before resolving (median MAE). Lower is better.
4dMedian holdHalf of qualifying signals resolve faster than this.
90ResolvedFinished qualifying signals. Win rates use these.

Which qualifying setups win

Win rate at a +2% take-profit, by the strategy and timeframe that fired the qualifying signal — the cut that maps straight to your strategy / timeframe filter.

Win rate @ +2% · by strategy · timeframe
Mean-reversion 1d27%n=48
Breakout 1d94%n=32
Trend 1d100%n=10
Strategy · timeframeSignalsWin +1%Win +2%Win +3%Median peakMedian heatRaw stop
Mean-reversion · 1d4833%27%23%+0.5%−10.3%90%
Breakout · 1d3297%94%94%+8.8%−15.4%22%
Trend · 1d10 *100%100%100%+11.2%−14.5%70%

* small sample (fewer than 15 resolved) — directional only. Raw stop assumes no take-profit; the win columns reflect how the agent actually exits.

Does raising conviction help?

Win rate at +2% across every 5% conviction band— the full range, shown independent of your minimum-conviction setting so you can see exactly what a higher or lower gate would include or exclude. If a higher gate reliably meant better trades, this would climb cleanly left-to-right. It doesn't — read the shape before you crank it up.

Win rate @ +2% · by conviction band
ConvictionSignalsWin +1%Win +2%Win +3%Median peakMedian heatRaw stop
[25, 30)3583%71%63%+4.2%−6.0%83%
[30, 35)6978%68%55%+3.6%−4.5%81%
[35, 40)3583%71%49%+3.0%−6.0%83%
[40, 45)8 *88%88%88%+10.5%−7.2%50%
[45, 50)2 *50%50%50%+3.8%−16.2%100%
[50, 55)1 *100%100%100%+84.1%−17.1%0%
[55, 60)2879%68%64%+5.7%−12.7%57%
[60, 65)2871%61%57%+4.6%−11.1%61%
[65, 70)13 *62%54%54%+4.5%−12.9%85%
[70, 75)2730%30%26%+0.0%−10.3%89%
[75, 80)11 *91%91%91%+5.5%−17.6%36%
[80, 85)3 *100%100%100%+13.3%−5.4%0%
[85, 90)6 *100%100%100%+9.0%−15.0%17%
[90, 95)2 *100%100%100%+23.6%−13.9%0%

By market regime

How qualifying setups did in each regime at signal time. This is why the agent requires regime alignment— a trend in the trade's direction historically outperforms range / high-volatility, and the live agent only takes the aligned rows.

RegimeSignalsWin +1%Win +2%Win +3%Median peakMedian heatRaw stop
Range4833%27%23%+0.5%−10.3%90%
High-volatility3297%94%94%+8.8%−15.4%22%
Trend-down6 *100%100%100%+9.0%−13.9%67%
Trend-up4 *100%100%100%+23.2%−15.3%75%

By manipulation risk (Trap Score)

Does a cleaner order book win more? Win rate and typical heat by the coin's Trap Score at signal time (the agent already caps this at 6/10).

Trap bandSignalsWin +1%Win +2%Win +3%Median peakMedian heatRaw stop
Low (0–3)8561%56%54%+4.2%−12.0%66%
Moderate (3–5)5 *100%100%100%+7.1%−14.4%20%
Elevated (5–6)0 *

What to set, from this data

Point your strategy / timeframe filter at the strong rows. Here, Breakout · 1d reached +2% before its stop 94% of the time (n=32), versus 27% for Mean-reversion · 1d.
Don't over-trust conviction alone. The conviction curve is uneven, not a clean climb — a high minimum-conviction gate is not, by itself, a reliable edge.
Match your exit to the data. Qualifying setups resolve in 4d (median) and typically run to +4.4% while sitting through −12.1% of heat — size for that drawdown, and a +2–3% take-profit realises the win columns.

Trades the agent has actually taken

Full transparency: the agent's real, on-exchange and paper positions to date — this table updates automatically as the agent opens and closes positions, so every future trade lands here with no manual step. It is a very small set today, which is exactly why the analysis above replays the rules over the full signal ledger instead.

CoinBookDirectionLeverageModeStatusBooked P&LOpened (UTC)
AAVE/USDTLiveLong10×Aggressiveclosed · killedJun 25, 00:02
MMT/USDTLiveLongAggressiveclosed · killedJun 22, 00:07
EIGENLAYERPaperLongopenJun 21, 01:15
No booked P&L yet. The live positions above were closed manually (killed) before an exit fill, and any paper position is still open — so none has a realised return to report. There is not yet a real, resolved live track record, which is why every performance figure on this page is a labelled rule-replay over the signal ledger.

Leverage & the soft-stop reality

Read this before arming with real capital — the risk boundary is not where a beginner assumes.

The agent deploys your entire declared capital as isolated-futures margin on its single pick. Leverage is derived per trade from the stop width and the mode you choose: Balanced risks ~15% of capital on a clean 1R stop-out, Aggressive ~40%. Leverage sets position size, not which signals qualify.

The agent's stop is a monitored soft-stop evaluated by the execution function, not a resting stop order on the exchange. At high leverage on full capital, a wide-ATR 1R stop can sit beyond the ~18–20% liquidation distance — meaning liquidation, not the intended stop, becomes the true risk boundary. Size accordingly, and treat the median-heat figures above as the drawdown you must be able to sit through.

How to read this page

Plain-language definitions for every term above.

Qualification funnel
Every recorded signal filtered by the agent’s gates in order. Each stage shows how many survive it and all prior gates, so you see exactly where and why signals drop out before reaching the agent.
Gate fidelity (exact / approx / live-only)
How faithfully a gate can be replayed from history. Exact = reconstructed from stored data. Approx = uses a documented proxy (older rows). Live-only = applied by the live agent but not reconstructable from past signals (shown as context).
Qualifying setup
A signal that passes the gates for the parameters you set. The performance tables slice the setups passing your trap, conviction, strategy and timeframe gates; the funnel’s endpoint additionally requires regime alignment, the quality floor and the re-entry guard.
Win rate (per target)
The share of RESOLVED qualifying signals whose price reached that take-profit target (+1/2/3%) before the stop. First-passage, not trade P&L — read the column that matches how you would exit.
Median peak (MFE) / heat (MAE)
The typical furthest move in the signal’s favour (peak) and against it (heat) before it resolved — medians, so a few outliers don’t distort them.
Raw stop
Worst case only: the share that touched the raw initial stop assuming NO take-profit and no trailing. The agent trails its stop, so this is context, not the headline.
Conviction / Regime / Trap Score
The engine’s confidence (0–100%), the market state at signal time (trend-up/down, range, high-volatility), and the 0–10 manipulation-risk gauge. All three feed the agent’s gates.
Balanced / Aggressive mode
The leverage dial. Balanced risks ~15% of capital on a clean stop-out, Aggressive ~40%. Leverage is derived per trade from the stop width — it changes position size, not which signals qualify.
Eligible trade
A signal that passed the agent's always-on structural gates — $5M+ 24h liquidity, Trap Score ≤ 6, a first target ≥ 2R, and regime alignment. The 150-day study shows every eligible trade; you then choose your conviction / strategy / timeframe on top.
First-passage (reached +X%)
Whether the price touched +X% in the trade's favour before the stop was hit — measured bar by bar, adverse move checked first (conservative). It is a probability of touching a target, not a realised profit.
Leverage (1× / 5× / 10×)
The agent deploys your full capital as margin, so leverage sets the position notional (capital × leverage) and therefore how far price can move against you before liquidation. Same stop price at every leverage — only the liquidation distance changes.
Liquidation
When price moves ≈ 1/leverage against the position, the margin is wiped and the trade closes at −100%. At 10× that is only ~9.5% adverse; at 5×, ~19.5%; at 1× it effectively never happens. A wide ATR stop can sit beyond liquidation, so liquidation — not the stop — becomes the real risk boundary.
EV per trade (on capital)
Expected value: the average net return on your capital per trade, after fees, across the whole population. It is the honest bottom line — a high hit-rate can still carry a negative EV once occasional liquidations (−100%) and the wide stop are counted.
Backtest vs live forward record
The 150-day study re-runs the audited engine over historical price bars (a backtest). The live forward record is every signal logged the moment it fired, then tracked — not backfitted. The two use the same engine and corroborate each other.

About this data

Why replay the agent’s rules instead of showing its real trades?

Because the agent has almost no live fills yet — a page built on real orders would be blank. Instead we replay its exact qualification gates over every signal the engine has recorded, so you see honestly what its rules would have selected and how those setups behaved. The tiny set of positions it has actually opened is shown separately.

What does it mean for a trade to “qualify for the agent”?

The agent only acts on a signal that passes every gate: Trap Score ≤ 6, your conviction minimum, your strategy and timeframe filters, a sane stop, regime alignment, a composite quality score ≥ 55, a $5M liquidity floor, and a re-entry guard. Most signals fail at least one gate, which is why only a small fraction qualify — the funnel shows exactly where they drop out.

Are these win rates the agent’s profit?

No. They are first-passage rates — the share of qualifying signals whose price reached +1/2/3/4/5% before the stop. They are not realised trade P&L. The agent trails its stop and pays fees, so its net track record is roughly break-even. This page is a research and configuration tool, not a promise of returns.

Which gates are exact and which are approximate?

Trap, conviction, strategy/timeframe, the sane-stop and the re-entry guard are reconstructed exactly from stored data. Regime alignment is approximate on older signals (only the 1d regime was stored — a conservative lower bound) and the quality score treats the small liquidity term as 0 without historical volume. New signals now store the full per-timeframe regime and 24h volume, so the funnel becomes exact going forward.

Why do so few signals reach the agent?

Deliberate selectivity. The engine surfaces many signals for research, but the agent is built to take one high-quality setup at a time, so the gates are strict. Requiring regime alignment and a quality floor removes most signals — the point of the agent is discipline, not activity.

Does a higher conviction setting guarantee better results?

Not by itself. The conviction curve on this page is uneven, not a clean climb, so a very high minimum-conviction gate is not a reliable edge on its own — and it quietly excludes the engine’s 4-hour signals, which carry lower conviction. Strategy and regime are usually the stronger levers.

Why model leverage from 1× to 10× — and why can a high hit-rate still lose money?

Because the agent deploys full capital, leverage is the real risk dial. The stop price is the same at every leverage, but higher leverage puts the liquidation price closer (≈ 1/leverage away). A trade can reach its target most of the time yet still have a negative expected value once the occasional liquidation (a −100% trade) and the wide stop are counted. That is exactly what the leverage table shows: EV per trade typically goes from near break-even at 1× to negative at the top of the range. And positive per-trade EV is not survival — deploying full capital repeatedly above ~2× compounds to ruin, so the real lever is fractional position sizing.

How far do the agent’s trades actually run, and what should I arm?

The study shows the full distribution of how far the coin moves in our favour before the stop — the median and top-quartile favourable peak and the share reaching +1/2/3/4/5/7/10% — because the agent trails its stop rather than exiting at a fixed target, so realised moves often run past any single take-profit. The “what to arm” section then reads the strongest entry (strategy × regime), the fixed exit with the best expectancy, and an honest leverage ceiling straight from that data, refreshed as the window rolls forward. It is configuration guidance, never a promise of profit.

How is the 150-day study calculated, and is it accurate?

It replays the same audited engine that produces the live signals over 150 days of real price bars across every tradable coin: next-bar-open entry, the engine’s own ATR stop, and a conservative intrabar rule (the adverse extreme is assumed to hit before the favourable one within a bar), so every figure is a floor, not a look-ahead best case. Leverage adds a liquidation check at ≈ 1/leverage. It is reconciled against the live forward-recorded ledger, which uses the same logic.

Backtest or live record — which should I trust?

Both, for different reasons. The 150-day backtest gives depth and sample (hundreds of trades, leverage modelling); the live forward record proves the engine behaves the same in the wild (nothing back-fitted). They use one engine and agree, which is the point. Neither is a promise of future results — crypto trading carries significant risk.