TL;DR
Write acceptance requirements before testing and keep them fixed even when the selected candidate makes money. This NVDA version returned 22.06% on April to June 2024 dates but failed four of five checks: it lagged the 36% stored NVDA benchmark, exceeded 10% drawdown, completed no round trips and held overnight. Freeze the rules, search budget and data roles, then inspect signals and holdings against the intended task. Published dates are now reproduction evidence; choose untouched dates before developing another candidate.
Five predeclared checks rejected this profitable candidate
Show the table · 5 rows, 3 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Positive later return | 22.06% | Pass |
| Beat the stored NVDA buy-and-hold return | 22.06% | Fail |
| Maximum drawdown at most 10% | 12.73% | Fail |
| At least 10 closed round trips | 0 | Fail |
| No position held overnight | One buy, no sells; position held to test end | Fail |
No matching rows. Clear the filter to see all records.
All five requirements had to pass. The NVDA benchmark threshold was 36.00%; the drawdown ceiling was 10%; the activity minimum was 10 closed round trips.
The candidate's 22.06% later return passed one requirement and failed four. The original acceptance criteria and measured results are retained here. Keeping them fixed prevented a profitable buy-and-hold-like result from being promoted as the requested intraday strategy.
Freeze a protocol that can reject the winner
Show the table · 5 rows, 3 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Define the task | Entry family, allowed mutations, sizing, exits and intraday holding requirement | The selected exit became $378,389.83; inspect the complete rules, not the seed name |
| Declare the search budget | Population, generations, windows and every restart | This search used population 6, 3 generations and 1 training window; six final rows do not count every evaluated variant |
| Separate data roles | Training for development; validation for selection; a later frozen test for acceptance | Validation selected the inactive 0% candidate; April–June then tested the frozen configuration |
| Preserve failure criteria | Return versus same-period baseline, drawdown, closed-trade count and overnight holdings | Do not raise 10% drawdown or remove 10-round-trip/no-overnight requirements after seeing the result |
| Protect the next test | Later dates excluded from every candidate revision and corpus query | The published 2024 dates are now a reproduction window; choose untouched dates before developing a new variant |
No matching rows. Clear the filter to see all records.
Inspect the frozen later test
Portfolio value (USD)
- Genetic winner 2024-06-28, 20:00:00 EDT
- $12,205.98
- Original morning strategy 2024-06-28, 20:00:00 EDT
- $8,980.94
- NVDA buy and hold 2024-06-28, 20:00:00 EDT
- $13,600.01
- SPY buy and hold 2024-06-28, 20:00:00 EDT
- $10,371.61
Genetic winner: 12.40% largest decline in plotted observations. 2024-06-20, 04:01:00 EDT ($13,328.24) to 2024-06-25, 04:01:00 EDT ($11,675.87).
The selected candidate earned 22.06% in the later test. The stored order record contains one filled buy and no filled sells, so this positive result came with an overnight position and zero completed round trips. Compare the winner, unchanged morning seed and the stored NVDA/SPY baselines on their shared later dates.
Verify the accepted signal became the intended trading behavior
Inspect the exact historical measurements
All four recorded entry conditions were true at April 1, 13:30 UTC. The complete order query returned one buy of 66.015730 NVDA shares at $90.30 with a $2.98 fee, and no filled sells. End holdings were $8,169.45 beside $4,036.53 cash. The planned no-overnight and closed-trade checks therefore matter independently of the 22.06% return.
Scroll sideways to compare all columns. Select a heading to sort.
| Minutes after open | 0 | ≤ | 0 | True |
| Minutes after open | 0 | ≤ | 139 | True |
| NVDA price / SMA | 90.29900360107422 | = | 90.29900360107422 | True |
| Position value | 0 | ≥ | 0 | True |
No matching rows. Clear the filter to see all records.
No comments yet.