Independently reproduced · BTC & ETH futures
We paid an independent analyst, Ranveer Verma, to rebuild our BTC and ETH futures backtests without our engine's source code — from our written specification, the parameters we supplied and our answers to the reviewer's questions — and rerun them on the same Binance trade data. All 14,018 BTC baseline cycles matched, and ETH was reproduced over 2.96 billion trades. The review also caught an error of ours and listed what the results do not prove. The reports and code are public on GitHub.
Commissioned · Independent · Reproduced. Historical simulations — past results do not guarantee future returns.
Each rebuild used the same Binance aggTrades (trade-by-trade) data, the same window (Jan 1, 2021 – Aug 24, 2026) and all published cost scenarios. The method, results and conclusions are the reviewer's.
Jan 1, 2021 – Aug 24, 2026
All 4 cost scenarios matched our published figures — final wallet, ROI, closed cycles, liquidations, capital resets and max drawdown. The largest money difference in these totals was under one billionth of a dollar. The published risk metrics matched 64 of 64 once we supplied their definitions and the relevant code excerpts.
Jan 1, 2021 – Aug 24, 2026
The first full run did not match our totals: our specification had given trigger checks a small tolerance, but our run compares trade prices with trigger prices exactly, so we supplied that rule and a one-day list of our cycles. After that, all 4 cost scenarios were reproduced over 2,963,456,806 trades: cycle counts, capital resets, fees, slippage, drawdown, liquidation room and completed-cycle holding statistics matched, and the baseline realized wallet matched to the last digit.
Preset unchanged · 2026-08-25 → 2026-09-30. Exit-net values the final open position at the mark close, less exit costs.
+1.53% exit-net · max drawdown −2.27% · 0 reported liquidations · end: Open position.
BTC OOS results and hashes+1.37% exit-net · max drawdown −1.05% · 0 reported liquidations · end: Open position.
ETH full window and the window not used in any prior researchETH disclosure: Aug 25–Sep 1 had been used by rejected research before preset selection on Sep 7–8 (data through Aug 24, apart from the Aug 25, 00:00 UTC minute corrected on Sep 30). The Sep 2–Sep 30 window, not used in any prior research, was also run from a fresh wallet; the ETH hash was published Sep 8, 2026.
Next out-of-sample result: Q4 2026 (Oct 1–Dec 31), published in early January 2027 — same presets, unchanged, with the execution plan frozen before the run, published whatever the result.
Checked to the minute
Our published ETH result had been valued at the Aug 25, 2026 00:00 mark close — one minute inside the out-of-sample window we had sealed. We moved it to the Aug 24, 23:59 close. The baseline ending capital rose from $1,946,288.58 to $1,946,324.41 — $35.83 higher — and the report fingerprints (SHA-256) from before and after the correction are both published.
See the correction recordReturn and drawdown, side by side
Both public backtests show the max drawdown next to the return: BTC +1,550.89% with a −81.31% max drawdown, ETH +1,359.75% with a −50.24% max drawdown. Trading fees are deducted, and funding and slippage variants are published separately. Both results include reinvestment through resets of the investment base after each 10% wallet growth; the live bot does not reinvest automatically.
What the rebuild found against us
On the baseline cost scenario, the reviewer changed one execution rule at a time. The volume and delay tests below lowered the ending balance by 24.9–59.0%, and none of the reviewer's tests produced a modelled liquidation. The reports list every test.
Fills need traded volume of 1–2× the order size at or through the order price
BTCEnding wallet 24.9–26.4% lower · max drawdown −91.9% to −92.8%
ETHEnding capital 41.7–45.5% lower · max drawdown −48.0% to −51.6%
1-second delay before re-entry and before placing or re-pricing ladder and take-profit orders
BTCEnding wallet 40.5% lower
ETHEnding capital 59.0% lower
For reference, not a test — closest approach to the modelled liquidation price in the baseline
BTCPrice came within 6.04% of the modelled liquidation price
ETHModelled liquidation price 13.49% below the mark price at that moment
None of these tests produced a liquidation. The volume tests use traded volume as a proxy with an assumed queue and all-or-nothing fills, so they bracket queue effects rather than measure them, and the tests were not combined. The model's liquidation rules are not the exchange's, so 0 modelled liquidations do not prove an account would have survived on the exchange.
Yes — they were reproduced, which is narrower than an audit. In September 2026 Blitz commissioned and paid Ranveer Verma, an independent analyst, to rebuild its BTC and ETH futures backtests from a written specification, the parameters Blitz supplied and written answers to the reviewer's questions, without the engine's source code, and rerun them on the same Binance aggTrades data. For BTC, all 4 cost scenarios and all 14,018 baseline cycles matched. For ETH, the first full run did not match; after Blitz supplied its exact price-comparison rule and a one-day list of its cycles, all 4 scenarios were reproduced over 2,963,456,806 trades. The reports and code are public on GitHub.
That the published numbers follow from the stated rules and the market data: someone who rebuilt the model from our specification, parameters and written answers, without our engine's source code, got the same results — on ETH once we had supplied our exact price-comparison rule and a one-day cycle list. It does not prove the strategy will be profitable, and it does not prove live fills — the model assumes every order fills in full the moment a trade prints at or through its price, with no delay.
Our published ETH result had been valued at the Aug 25, 2026 00:00 mark close, one minute inside the out-of-sample window we had sealed. We corrected it to the Aug 24, 23:59 close (baseline ending capital $1,946,288.58 → $1,946,324.41) and kept the report fingerprints from before and after the correction public. From indirect evidence, the report also concludes that the three funding scenarios had charged that minute's funding payment.
Ranveer Verma, an independent analyst. Blitz commissioned and paid for the work. The method, results and conclusions are the reviewer's, including the negative findings, and the reports and code are published in the reviewer's own GitHub repositories.
The published BTC and ETH runs report 0 liquidations, and none of the reviewer's stress tests produced one. At the closest point, the BTC price came within 6.04% of the modelled liquidation price, and in the ETH baseline the modelled liquidation price was 13.49% below the mark price at that moment. The model's liquidation rules differ from the exchange's — a flat maintenance rate instead of the exchange's tiers, trade prices instead of the mark price on BTC, and on ETH each minute's closing mark applied from the start of that minute — so 0 modelled liquidations do not prove an account would have survived on the exchange.
Both public backtests show the max drawdown next to the return: BTC +1,550.89% with a −81.31% max drawdown and ETH +1,359.75% with a −50.24% max drawdown, over Jan 1, 2021 – Aug 24, 2026. Trading fees are deducted, and funding and slippage variants are published separately. Both results include reinvestment through resets of the investment base after each 10% wallet growth; the live bot does not reinvest automatically.
Not from GitHub alone. The reviewer's reports and code are public, and the BTC report JSON can be downloaded with its SHA-256 fingerprint and a DOI archive copy, but the exact (unrounded) ladder values the reproductions used were given to the reviewer privately and, at our request, the public repositories omit them. The same ladders are visible in the Blitz app to anyone with a free account, as rounded values, which are not enough to match the published figures exactly.
Yes. After logging in, you can backtest your own DCA settings (long setups) free in the strategy editor, on up to a year (366 days) of 1-minute candles for curated symbols, depending on available data. There is no plan-gated lookback and no monthly run quota. The reproduced backtests apply only to the exact published presets.
The reproduced results cover two published presets. Backtest yours free on up to 366 days of 1-minute candles — no plan gate, no monthly run quota.
Product
© 2026 BlitzTrade. All rights reserved.