Public sample — deliverable format

H001 — Backtest Evidence Review (sample)

Pre-registered daily BTC/USDT trend hypothesis (Donchian + ATR trail, long-only, EMA filter). Archived lab report: H001_20260810T082405Z (Aug 2026). This page shows the report structure a buyer receives — applied here to my own failed experiment, not a client strategy.

I assess whether the tests and evidence you provide support the result. I do not rerun the strategy, generate new statistical tests, optimize it, or certify profitability.

This document is methodological risk screening on materials provided (here: an archived lab report). It is not investment advice, a recommendation to buy or sell any instrument, or permission to trade. It does not certify absence of leakage or reproduce a full backtest. Past backtest results do not predict future performance.

1. Executive summary

H001 tested a pre-registered daily BTC/USDT long-only trend rule with walk-forward out-of-sample segments and a documented null test (100 permutations; pass if combined OOS PnL exceeds the 95th percentile of the null distribution). Combined OOS metrics look acceptable on headline PnL and expectancy, but pre-registered gates failed on null test and minimum trade count.

Evidence verdict: Insufficient

2. Materials reviewed

ItemReceivedNotes
Trades / results exportYesFrom archived lab report H001_20260810T082405Z
Rules / assumptionsYesPre-registered hypothesis summary (Donchian + ATR trail, long-only, EMA filter)
Code / notebook (static)Not providedNot included in this public sample
OOS definition statedYesWalk-forward: 730 train / 180 test bars, step 180
Null / permutation resultsYesDocumented in report: 100 permutations; 95th percentile threshold 455.5

3. Methodology scorecard

CheckRatingComment
Data timing vs signal (leakage)Not assessedPublic OHLCV only in sample; no separate leakage audit documented
Costs & slippage realismPartialPre-registered 10 bps fee + 5 bps slippage per side; not validated against live fills
IS/OOS or walk-forward integrityPartialWalk-forward defined; fold 7 had zero OOS trades
Parameter selection disciplinePartialParameters locked before OOS per lab protocol; not re-audited here
Multiple testing awarenessNot assessedNot provided in public sample
Null / permutation / placeboNoFAIL — observed 113.9 < threshold 455.5; null mean 148.8
Sample size / trade countNo13 combined OOS trades < 15 minimum gate
Period robustnessPartialfold_1 negative expectancy (−6.51/trade) while other folds positive
Backtest vs plausible executionPartialDaily Binance spot; fixed bps only — execution gap remains

4. Five primary evidence risks (ranked)

  1. Null test failure — Combined OOS PnL (113.9) does not exceed the pre-registered 95th percentile null threshold (455.5); path/luck not ruled out.
  2. Under-powered OOS sample — 13 trades vs 15 minimum; fold 7 had zero trades.
  3. Fold instability — fold_1 negative expectancy (−6.51/trade) while other folds positive — fragile period dependence.
  4. Interpretation risk — Headline OOS PnL (+113.92 on 1,000 start) invites live deployment without null/sample gates.
  5. Execution gap — Daily spot backtest ≠ guaranteed fills/slippage in live; modeled only as fixed bps.

5. Findings detail

5.1 Data & leakage

Symbol: BTC/USDT · 1d · Binance spot (public OHLCV). Leakage risk not separately assessed in this public sample.

5.2 Costs & execution assumptions

Pre-registered cost model: 10 bps fee + 5 bps slippage per side. No live execution data provided.

5.3 OOS / walk-forward

Walk-forward: 730 train / 180 test bars, step 180. Combined OOS: 13 trades; expectancy 8.76 per trade (currency units on 1,000 start); total OOS PnL 113.92; win rate 46.15%; profit factor 2.48; max drawdown −1.95% on trade equity path. Gates oos_expectancy_gt_0 and oos_total_pnl_gt_0: PASS.

5.4 Statistics & null tests

Null test (as documented in report): 100 permutations; pass if combined OOS PnL > 95th percentile of null. Observed 113.9 < threshold 455.5; null mean 148.8 — FAIL. Backtest Evidence assessed this documented test; it did not generate or rerun permutations for this sample.

5.5 Robustness & limitations

Sample gate: 13 < 15 required trades — FAIL. Period mix is unstable (fold_1 negative). This review does not certify profitability or absence of overfitting.

6. Recommended next experiment (one only)

Hypothesis (pre-register before running): New hypothesis ID — do not tune H001 parameters on this OOS.

Change vs current backtest: Either archive H001 or open a new pre-registered variant (e.g. longer history requirement before any paper trading).

Success / fail gates (measurable): Same pre-registered null bar and ≥ 15 OOS trades before any paper trading — still no live capital until all gates pass.

Explicitly out of scope: Parameter optimization on H001 OOS; live deployment; signal service.

7. Clarifications

Not applicable — public educational sample. Paid reviews include one async clarification round by email.

8. Data handling & deletion

For paid client reviews: working copies of your files are deleted within 14 days of final report delivery unless you request earlier deletion. See FAQ and Privacy.

This public sample uses only metrics copied from an archived lab report — no client files were uploaded.

Educational sample based on a failed personal lab experiment. Numbers are copied from the archived report only. Not a recommendation to trade any strategy.

Get the review — €79

Legal · Privacy · Terms