Windows run in order, with no extra out-of-sample passes and no retuning after a peek. Under 100 trades or under about three months is a diagnostic, not a pass.
Example data: Matt Berg Full Mech (diagnostic, below 100 trades per window)
Run order, Berg Full Mech
Order
Window
Length
Trades
Status
Net
1
Tuning (IS)
four months test
33
diagnostic below 100 trades
−$10,520
2
Out-of-sample (OOS)
three months test
29
diagnostic below 100 trades
−$18,155
3
Holdout (HO)
about six weeks test (data ends 14 Sep 2026)
19
diagnostic below 100 trades, under about three months
−$12,575
Windows ran in order with the same locked settings: no retuning between them, and one out-of-sample pass.
No window reaches 100 trades (33 / 29 / 19), so every Berg Full Mech result here is a diagnostic, not a pass.
The holdout is about six weeks, which is under about three months, so it is a diagnostic on length as well.
Full Mech failed out-of-sample, so under rule 5 its holdout run is shown for visibility only, not as a promotion attempt.
The missed-trade fixes on the rule 8 page were chosen on tuning data only, and their out-of-sample and holdout effects were reported once. They belong to the next version, not this one.
Rule 10 is about process, so this page has no charts. The trade charts are on the rule 3, 4 and 5 pages.