Phone Bot Lab · Sample report · Standard tier
Backtest report: KLV/USDT minute bot, "v2" update
Prepared for: the author's own bot · Report date: 9 October 2026 · Format: Standard tier, plus extra fix testing
Pair: KLV/USDT · Data: public KuCoin 1-minute candles · Sleeve modelled: $100 (and $10)
1. Summary verdict
On this evidence, the v2 update doesn't fix the bot's real problem: trading costs. It loses about as much as the original version over the one-year window and makes 2.5 times as many trades. It also lost money in every calendar year tested from 2021 to 2026. A round trip on this pool costs about 3.6% at $100, and most of v2's exits take profits of as little as 0.5%. That makes them losing trades before they start.
−68.4%
v2, train window ($100 sleeve)
−11.2%
v2, test window (out of sample)
$82.20
fees paid on a $100 sleeve (train)
3.59%
round-trip cost per $100 trade
- The new −5% hard stop fires on ordinary minute-by-minute noise: 39 stops in train for a net −$42.
- 282 exits in train were "RSI exits" at gains below the cost of the trade, for a net −$31.50.
- None of the targeted fixes we tested turned the result positive in both windows. Only removing the cost leaks (Section 6) brought it close to flat, and that still lost money.
2. What changed in v2
Compared with the original code (67 changed lines). The code header lists only some of these changes.
| Setting | Original | v2 | In the header? |
|---|---|---|---|
| RSI buy threshold | 42 | 37 | yes |
| Take-profit | +11% | +12.5% | yes |
| Trailing stop / activation | 6.5% / 5% | 5.5% / 4% | yes |
| Drawdown halt | 18% | 15% | yes |
| Cooldown after a sell | 12 min | 18 min | yes |
| Entry score minimum | 0.62 | 0.62 | header says 0.58; code unchanged |
| Hard stop | none | −5% | no |
| Buys per position (including the first) | up to 5 | 1 below the 7-day average, up to 3 above | no |
| No extra buy when already down | none | more than 1.5% | no |
Swipe sideways to see every column.
Not changed: entry scoring, position sizing, the self-adjusting learner, the RSI exit at +0.5%, and the 72-hour limit that only sells small losses. Note that the new hard stop also applies to KLV the bot "adopts" from the wallet, so it can sell the owner's own coins on a 5% dip.
3. Cost model
Every simulated trade pays these costs. They were measured on the live pool on 7–10 October 2026.
| Cost | Per swap | Source |
|---|---|---|
| Network fee | 26.55 KLV ≈ $0.018 | checked on-chain |
| Pool fee | 0.3% of the trade | matches a DEX router's live quotes for this pool |
| Price impact | ≈ trade size ÷ $6,600 | pool's USDT reserve |
| Pool vs exchange price gap | 0.2% (sensitivity: 1.7%) | modelled; 1.7% is the measured median |
Swipe sideways to see every column.
A $100 buy-then-sell costs 3.59% before the price gap (10 Oct 2026 reserves), and the price has to rise 3.75% just to break even. At $10 the fixed network fee dominates instead. You can check these numbers yourself in the free True Cost Calculator.
Fee updated 10 Oct 2026 from 0.1% to 0.3% after checking against live router quotes. The results in sections 4 to 6 were simulated at the earlier 0.1% fee. Re-run at 0.3% on the same data, train / test: original −77.6% / −19.3%, v2 −75.6% / −22.1%, recommended fixes (R1) −10.7% / −4.3% (were −72.6% / −9.5%, −68.4% / −11.2% and −9.8% / −1.9%). Every conclusion in this report still holds; the losses are larger.
4. Train/test results ($100 sleeve)
Train: 7 Oct 2025 – 8 Jun 2026. Test: 8 Jun – 7 Oct 2026, data the settings were never tuned on. Each window starts with a fresh $100.
| Version | Train | Test | Trades (train / test) | Winning trades | Max drawdown | Fees paid | Train / test with a 1.7% price gap |
|---|---|---|---|---|---|---|---|
| Original | −72.6% | −9.5% | 130 / 69 | 8% / 26% | 76% / 33% | $86.80 / $45.50 | −97.8% / −60.7% |
| v2 as coded | −68.4% | −11.2% | 323 / 138 | 6% / 25% | 70% / 20% | $82.20 / $41.70 | −103.3% / −66.0% |
| Recommended fixes (R1) | −9.8% | −1.9% | 36 / 36 | 28% / 50% | 13% / 7% | $6.60 / $7.90 | −18.0% / −17.6% |
| Reference: hold KLV | −81.0% | +21.8% | 1 / 1 | – | – | – | – |
| Reference: hold USDT (do nothing) | 0.0% | 0.0% | 0 | – | 0% | $0 | 0.0% |
Swipe sideways to see every column.
Losses above 100% happen because the network fee is paid from wallet KLV outside the sleeve. With a $10 sleeve v2 lost 122.8% in train and 44.8% in test. Holding KLV is shown for reference only; it isn't a recommendation.
5. Drawdowns
How far each version fell from its previous high point, day by day, on the same −80% scale for both windows. The curves are drawn from the simulator's hourly equity samples, so the deepest points can be a little shallower than the tick-level maximums in Section 4.
On a narrow screen, swipe sideways to see the whole chart.
Tap or drag across a panel to read the drawdown on a given day.
- Original bot
- Recommended fixes (R1)
- v2 as coded
6. Why v2 loses, and the fixes tested
Where the money went (v2, $100 sleeve)
| Exit reason | Train: exits, net | Test: exits, net |
|---|---|---|
| RSI exit (as little as +0.5%) | 282, −$31.51 | 117, −$3.31 |
| Hard stop (−5%) | 39, −$41.98 | 16, −$18.02 |
| Time limit | 2, −$1.10 | 2, −$0.98 |
| Take-profit and trailing stop | 0 | 3, +$9.89 |
Swipe sideways to see every column.
Fixes tested on v2
| Change | Train | Test | Verdict |
|---|---|---|---|
| v2 as coded | −68.4% | −11.2% | baseline |
| Entry score 0.58 (as the header says) | −68.4% | −11.2% | no change |
| Time limit always sells | −68.3% | −11.0% | no real change |
| One buy per position (no averaging down) | −49.2% | −12.8% | train better, test worse |
| Hard stop at −8% instead of −5% | −57.7% | −10.1% | still loses |
| Remove the hard stop | −55.2% | +0.6% | biggest single change; still loses heavily in train |
| Hard stop + time limit always sells + one buy per position | −49.1% | −12.3% | not a rescue |
| …plus skip high-fee trades, RSI exit off, learner off | −10.6% | −7.3% | cost leaks removed; still loses |
| v2 settings with the R1 fixes, no hard stop | −9.6% | −4.9% | still loses |
Swipe sideways to see every column.
Every calendar year, 2021–2026 (fresh $100 each year)
| Version | 2021* | 2022* | 2023 | 2024 | 2025 | 2026† |
|---|---|---|---|---|---|---|
| Original | −57.8% | −95.4% | −50.6% | −34.9% | −62.4% | −25.8% |
| v2 as coded | −66.7% | −93.7% | −72.2% | −68.1% | −77.7% | −43.9% |
| v2 without hard stop | −56.9% | −34.1% | −45.9% | −13.4% | −12.5% | −21.7% |
| Recommended fixes (R1) | −48.6% | −24.9% | +6.1% | −3.5% | −17.4% | −4.0% |
Swipe sideways to see every column.
* From March 2021. The Klever pool didn't exist for part of 2021–22, so today's pool depth and fees are applied as a hypothetical. † To 7 October. The network fee is held at today's $0.018; charging it at 2021 KLV prices makes every row worse.
7. Caveats
- These are backtests: hypothetical results on past data, not a reliable indicator of future results.
- Prices and fills both come from KuCoin minute candles, not the Klever pool itself. The bot's 20-second checks are reconstructed from 1-minute candles.
- The live bot's drawdown halt uses the whole wallet's KLV value, which can't be reproduced here. Run with a halt on the $100 sleeve instead, v2 is capped at about −13% (train) and −8% (test), and it still loses.
- The "accumulation" changes don't help. The 1% left after each sell is smaller than the KLV spent on network fees in train, and changing the refill floor moved results by at most 2 percentage points.
- Results are based on a single pair and one cost snapshot. Costs on this pool change with its liquidity.
8. What this report is, and isn't
This report assesses method and costs: what the rules would have done under realistic costs, where the money went, and which changes make a difference. It doesn't tell you what to trade, and it isn't a recommendation to run any strategy with real money. No keys, wallets or live systems were used to prepare it. The code was read and simulated offline.
Get a report like this for your own rules
Educational content and software only. Not financial advice or a recommendation to buy, sell or hold any cryptoasset. Backtested results are hypothetical and not a reliable indicator of future results. Cryptoassets are high risk; you can lose all the money you put in. Not affiliated with Klever, bitcoin.me or any exchange.