Faithful replication of FMZ’s – Backtest

Hypothesis

Tradable strategy by 恐龙宝宝: https://www.fmz.com/strategy/330440. Use BTCUSDT.BINANCE perpetual on 1-minute data. Source defaults are M=20 leverage, H=50 initial-position units, n1=1 BTC per grid order, grid=200 USDT, xia=35000 USDT and shang=60000 USDT. The author’s separate 2021 backtest overrides H=30, n1=0.001 BTC, grid=300 USDT and xia=50000 USDT, with a 2500 USDT balance. On startup, restore buyp and sellp only if both saved values are truthy; otherwise set them to last price minus and plus grid and save them. Subsequent grid shifts are not saved. Preserve the source’s fixed BTC quantities when testing; its claimed performance is unverified.

Uses the cited FMZ backtest overrides. Restart handling persists grid IDs and cancels carried grid orders so a following bar can recreate them; a fill racing cancellation still matches its ID. Exposure and basket loss limits address QA safety concerns. An infeasible fixed initial position raises an explicit error instead of silently skipping entry. The 31-day profitability result remains inconclusive because its metrics were marked unreliable.

verification_loop: Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The 31-day smoke test lost money, but its metrics are marked unreliable; performance needs a reliable full-history backtest. - [critical] The 0.06 BTC position cap suppresses fixed-size grid additions. The cited FMZ strategy has no position cap, so this changes the strategy being tested. (line 126) — fix: Remove the cap for the faithful replication, or test the capped version under a separate hypothesis. - [critical] The added 20% basket-loss rule closes the position and permanently blocks re-entry. The cited strategy has no such stop or shutdown. (line 75) — fix: Remove the stop for the faithful replication, or test it as a separate risk-managed variant.

Implementation

BTC perpetual arithmetic grid using the cited 2021 fixed quantities, capped at 0.06 BTC total exposure with a 20% basket loss stop.

Verification Results

Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The 31-day smoke test lost money, but its metrics are marked unreliable; performance needs a reliable full-history backtest. - [critical] The 0.06 BTC position cap suppresses fixed-size grid additions. The cited FMZ strategy has no position cap, so this changes the strategy being tested. (line 126) — fix: Remove the cap for the faithful replication, or test the capped version under a separate hypothesis. - [critical] The added 20% basket-loss rule closes the position and permanently blocks re-entry. The cited strategy has no such stop or shutdown. (line 75) — fix: Remove the stop for the faithful replication, or test it as a separate risk-managed variant.

Iteration History

Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The 20-day smoke test liquidated after one short trade. Assess the edge only after the execution and sizing defects are corrected. - [critical] The hypothesis requires fixed H × n1 BTC initial orders. Capping that quantity by available equity changes the strategy being tested. (line 61) — fix: Use the fixed quantity with sufficient test capital, or define the capital-capped version as a separate hypothesis. - [critical] Grid……Show moreShow less

Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The 20-day smoke test liquidated after one short trade. Assess the edge only after the execution and sizing defects are corrected. - [critical] The hypothesis requires fixed H × n1 BTC initial orders. Capping that quantity by available equity changes the strategy being tested. (line 61) — fix: Use the fixed quantity with sufficient test capital, or define the capital-capped version as a separate hypothesis. - [critical] Grid orders stop outside the lower-to-upper range. The FMZ source continues its grid for shorts above the upper boundary and longs below the lower boundary. (line 45) — fix: Apply the boundary condition according to position side, as in the source. - [critical] Open GTC grid orders cause the inherited bar processor to skip should_exit(). A boundary crossing therefore cannot trigger the required full-position exit while those orders remain open. (line 53) — fix: Check boundary exits before the open-order guard, cancel the grid orders, and close the position. - [critical] The code never saves the initial grid anchors, while the framework's paper snapshots save later shifts. The source saves the initial anchors and leaves subsequent shifts unsaved. (line 35) — fix: Persist the initial anchor pair only and restore it only when both values are truthy. - [critical] Fallback sizing can commit 95% of 20x leveraged capacity without a hard loss bound. The smoke test liquidated its position; this is a sizing defect rather than evidence about profitability. (line 66) — fix: Bound position risk explicitly before submitting the initial order.

Verification failed (Layer 4 — QA review) [class=code_defect]: - [edge_concern] The 31-day sandbox lost 75.4%, but its metrics are marked unreliable. Profitability requires the full backtest. - [critical] Paper restarts carry open GTC grid orders, but the strategy excludes their IDs from persisted state. A carried order's fill then cannot match either grid ID, so the anchors do not shift and the opposite order is not canceled. (line 32) — fix: Reconcile carried orders and their IDs on restart, or cancel and……Show moreShow less

Verification failed (Layer 4 — QA review) [class=code_defect]: - [edge_concern] The 31-day sandbox lost 75.4%, but its metrics are marked unreliable. Profitability requires the full backtest. - [critical] Paper restarts carry open GTC grid orders, but the strategy excludes their IDs from persisted state. A carried order's fill then cannot match either grid ID, so the anchors do not shift and the opposite order is not canceled. (line 32) — fix: Reconcile carried orders and their IDs on restart, or cancel and recreate them. Test a restart followed by a grid fill. - [critical] The margin check covers only the initial position. Adverse grid fills keep adding 0.001 BTC without a cumulative exposure limit or basket stop; short-side required margin and loss can grow without bound. (line 117) — fix: Bound cumulative grid exposure and loss on every addition, and enforce a basket-level stop. - [critical] The added 3% initial-margin gate suppresses the source's fixed-size entry at the cited 2,500 USDT balance whenever BTC exceeds 50,000 USDT. It therefore prevents the specified short entry above 60,000 USDT. (line 90) — fix: Preserve the fixed-size entry for a faithful test, or explicitly reject an infeasible capital configuration instead of silently skipping the entry.

Outcome Summary

Test the source grid without the added limits, and evaluate a risk-managed version under a separate hypothesis.

After three iterations, Layer 4 QA found that a 0.06 BTC position cap and a 20% basket-loss shutdown changed the strategy being tested. The run was abandoned at verification; later review stages were not reached.

It tried to replicate an FMZ BTC perpetual arithmetic grid using fixed BTC quantities.

A 31-day smoke test lost money, but its metrics were marked unreliable. No reliable full-history results or trade and Sharpe figures were provided.

Abandon Reason

verification_loop: Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The 31-day smoke test lost money, but its metrics are marked unreliable; performance needs a reliable full-history backtest. - [critical] The 0.06 BTC position cap suppresses fixed-size grid additions. The cited FMZ strategy has no position cap, so this changes the strategy being tested. (line 126) — fix: Remove the cap for the faithful replication, or test the capped version under a separate hypothesis. - [critical] The added 20% basket-loss rule closes the position and permanently blocks re-entry. The cited strategy has no such stop or shutdown. (line 75) — fix: Remove the stop for the faithful replication, or test it as a separate risk-managed variant.

Analysis

简易等差合约网格