Where the trading system stands

Where the trading system stands SIMULATED

The goal is to find out whether this screen is worth trading. Progress is counted in independent forward readings, not in the size of a simulated gain. Model v3.2.

A Baseline -1.0%B Conviction -0.6%C Risk-managed -1.9%D Regime-aware -1.5%E Combined -2.0%F Slow -1.4%IWO Russell 2000 Growth ETF -5.5%

Every line starts at 0%, so they can be compared directly. Book A is the control — a clever book that does not finish above it has not earned its extra trading. The dashed line is the index all six are trying to beat. These are simulated results; no money is invested.

Days on the record

first mark 2026-10-01 · 6 books live

Control book

-1.0%

equal weight, no overlays — the number every clever rule has to beat

Best overlay, vs control

+0.2%

B Conviction · highest of 5, so it flatters itself · measured over the period all the books share

Days to the review

evidence is judged 2026-12-14, on whatever it says

Independent 1-week readings — non-overlapping, so a single good week cannot be counted twelve times2 of 12

Independent 4-week readings — the slower cadence, where a genuine edge should still be visible0 of 3

✓A ranked screen is published every run
✓6 paper books trading the screen
○2 of 12 independent 1-week readings
○0 of 3 independent 4-week readings
○Scheduled review of the evidence, 2026-12-14
○Verdict: trade it, change it, or abandon it

Too early to judge, by design. 2 of 12 independent 1-week readings are in. Until that bar is cleared the scoring rules stay frozen, because a model tuned while its record is being written will always look good and will always be lying. The returns above are simulated and short; the friction column in the table below is the number most likely to decide the answer. No money is invested and nothing here is a recommendation.

The December decision, decided in advance

Written 2026-09-23, to be applied 2026-12-14. This rule was fixed while the record was eight readings old and the books three days old — before anyone knew how it would come out. That is the only time such a rule can be written honestly. Without one, the review is a person looking at a number they have already seen and deciding what it means, which is how a project runs forever and never concludes anything. No outcome below authorises real money. Twelve weeks cannot tell a real edge from a lucky one, and saying so now prevents the claim being made later.

Gate	What it asks	Where it stands	Status
○ A signal at all	mean of 12 independent 1-week readings above zero, by at least 2.0 standard errors	2 of 12 readings, mean -0.44%, t = -22.00	too early to say
✓ Survives its costs	at least one book ahead of the index after all trading costs, over the period every book shares	best is book B at +3.67%	met
✓ Broad, not a few names	the leading book still ahead of the index with its 3 best holdings removed	book B would be -2.07% against the index’s -5.45%	met

The bar can no longer be met by 2026-12-14. 1w: 2 frozen, at most 11 possible by then, 12 needed; 4w: 0 frozen, at most 2 possible by then, 3 needed. The schedule had no spare, so either a reading was missed or a correction to the model restarted the record — the model version and the closed books below say which. Either way the review will return “not enough evidence” for a reason that is not about the strategy, and the pre-registered rule already covers that case: continue unchanged to 2027-03-15. The bar is NOT being lowered to fit; the shortfall is shown instead.

As things stand: Not enough evidence yet. The record did not reach the bar. No rule changes are authorised. The books continue unchanged to the second checkpoint, 2027-03-15.

One clause matters more than the gates. A disappointing result does not authorise changing the scoring rules. The only outcomes are to continue unchanged or to stop. Tuning the model after seeing its record is how a system is made to look good in hindsight, and it is the exact thing the freeze exists to prevent — so it is ruled out here, in writing, in advance. Rules may change only to fix a defect: something that does not do what it is documented to do.