# Decision: Prime Academy strategy parameters (candidate 2)

Bracketed numbers are read from a grounding file. "Chain" and "wallet_mc" are mine, from `modules/prime.py` at the deployed stop's mean of 34.8 pips. Cost share of risk: 1.44% commission only, 2.87% with the 0.5-pip spread, 6.7% with the news asides (item 8).

## 1. The chosen settings

1. **Ratio, both stages: RR = 1.** RR 1 at 1.5% is the funded ridge, 52.4% chance of any payout [paper Fig 5]. Above 1 the pass rate gains under 2 points and the 35% cap binds sooner.
2. **Assessment risk: 3.0% per trade.** Chain, tick floor, target clipped, 2.87% cost: pass 45.4%, 9.7 trades mean, median 7, p90 19. At 4%: 46.2%, 5.9 trades. The 0.8 points buy compliance with the 35% cap under its strict reading: a 3% win is 33% of the 9% target, a 4% win is 44% [rules.json; paper §14.9 hatches 4%]. Broker at this cell: 47.6% against 47.7% model, z −0.03, 24 blocks, resolvable [broker_map.json].
3. **Funded risk: 1.5% per trade.** Chain, $3,000 minimum request 4.76%: P(any payout) 52.4%, 1.10 payouts per account, gross withdrawn 6.66%, median first payout 18 trades (p90 45), mean life 54 trades. Neighbours: 1.25% 52.1% at 23 trades, 1.75% 49.6%, 2% 46.3%, 4% 29.4%. Broker: 52.5% ± 3.3, first payout 46 days, life 123 days, 4 blocks, unresolvable [broker_map.json]. A 1.5% win is 31.5% of the minimum request, inside the 35% rule.
4. **Tier: $3,000, fee $38** (3.8% of the $1,000 wallet). wallet_mc, 500 trades, 6,000 paths, P(paid)=1: ruin 0.1%, median wallet drawdown $276 (p90 $532), first cash median 66 trades (about 134 days at the broker's 2.03 days a trade), log growth +0.167. The $10,000 tier: ruin 12.1%, drawdown $693, outside the 5% cap.
5. **Accounts: 1** until the first payout is honoured, **then 2**, the second entered 16:00 to 19:00 UTC, same side. Paper Table 20 at $1,000: one account ruin 1.1%, growth +0.79; two staggered 2.7%, +1.21. The second fee waits on P(paid), the unknown that moves everything, and two accounts in one name at payout are unpriced [paper §16].
6. **Side short, entry 07:00 to 10:00 UTC, stop max(0.75 σ, 30 pips), target the same distance.** Short is credited, +$1.76 against −$6.24 per lot per night [paper §9.2]. The window has the tightest spread [vault 2026-09-13]. The stop won the walk-forward on both halves [paper §12.3].
7. **Wire the target clip before the first fee.** `set_account` exists (strategy line 261) and the map uses it, but `run_live.py` freezes risk 0.04 and never calls it. Unclipped, the chain reads 42.0% at 2.5%, **40.2% at 3%** (three wins net of cost fall short of 9%), 40.7% at 4%. Clip not wired: assess at **2.5%**, never 3%.
8. **News rule: step aside, all-currency calendar**, the only policy with zero breaches [paper Table 24]. Each aside costs one commission plus one spread, (0.5 + s)/34.8 of risk, and the cohort paid 1.33 a trade. The map ran with the rule off. On, at s = 0.5, the cost share is 6.7%: pass 39.3%, P(any) 41.8%, expected cash $36.8 against the $38 fee, wallet log growth −0.29 with every payout honoured (chain, wallet_mc). At 2.87% the same cell returns $63.5, 1.67 per fee, breakeven P(paid) 60%.
9. **Purchase gate.** Buy only when expected cash per fee at P(paid)=1 is at least 1.33, which is a cost share of at most 4.5% of risk: (0.5 + s)(1 + A)/34.8 ≤ 0.045. At s = 0.5 that is A ≤ 0.55 asides a trade, at s = 0.2 it is A ≤ 1.2.

Daily rule: one loss is r(1+γ), so r ≤ 4.42% at 2.87% cost and 4.26% at 6.7%. Both stages fit.

## 2. The rule

Maximise the funded account's chance of ever paying, P(any payout), on the exact chain with the floor on the tick, the target clipped, the tier's own minimum request, and the cost share at the mean deployed stop including the measured asides. Hard constraints: every rule under its strict reading, so one loss r(1+γ) at most 5% of 91% of the balance, one win RR·r at most 35% of the target in the assessment and 35% of the minimum request when funded, one trade a day, every account on the same side. The assessment cell is the highest pass rate among cells meeting those constraints, ties within one point going to fewer trades. The tier is the largest whose fee is at most 6% of the wallet with simulated ruin at most 5% at P(paid)=1 over 500 trades, held once until a payout is honoured and then as many times as the staggered ruin cap allows. A purchase is made only if expected cash per fee at P(paid)=1 is at least 1.33 at the measured cost. A broker cell with fewer than ten blocks of twice its account span takes the chain's value. A resolvable cell more than two block-clustered standard errors below its model under the broker's protocol is dropped. An unresolvable cell beats a resolvable one only when its chain value leads by more than the largest gap seen among resolvable cells of that stage, 3 points funded and 2 assessment.

In one sentence: pass as fast as the rulebook allows, be paid as often as the rulebook allows, and buy only what the wallet can lose one time in twenty and the firm need honour no more than three requests in four.

## 3. What would change the choice

- **35% cap confirmed funded-only** (unverified in rules.json): assessment to 4%. +0.8 points, 5.9 trades not 9.7, 12 days to pass not 21 [broker_map.json]. Under the strict reading 4% forfeits every pass. Assumed: strict.
- **Aside rate A and spread s**: the gate. At s = 0.5, A = 0.5 gives 1.37 per fee (breakeven 73%), A = 1.0 gives 1.12 (90%), A = 1.33 gives 0.97. A EUR+USD-only calendar cuts A by an unmeasured amount, likely under half, since 33 of 46 cohort breaches were USD releases [paper §13.4].
- **Clip not wired**: assessment 2.5% (42.0%), not 3% (40.2%).
- **Floor at review, not on the tick**: funded extraction at 4% rises toward 10% [paper §16], the ridge stays at 1.5%, the pass rate rises about 1 point. Choice unchanged. Assumed: tick, as the operator states.
- **P(paid)**: wallet_mc at 2.87%, 3%/1.5%: log growth +0.167 at 1.0, +0.006 at 0.8, −0.18 at 0.6. Expected value breaks even at 0.60. A prior below 0.8 shrinks the wallet in log.
- **Both swap sides debited over 0.5 pip a night**: no credit on either side, cost share up about 1% of risk, pass rate down about 1 point. Unchanged.
- **14-day clock in calendar days**: about 7 trades at this pace, looser than the chain's 14 trades, so 52.4% is a floor. Unchanged.
- **Wallet above $2,500**: the tier rule picks $10,000 [paper Table 17]. **Firm objects to two accounts**: stay at one.

## 4. The falsification test, declared before any number is looked at

**Run.** `HistoricalBarBroker` over the vendor hourly EURUSD bars 2023-11-24 to 2026-09-10, `PrimeAcademyStrategy` with `set_account` on (tick floor, clipped target, $3,000 tier), news rule **on**, stepping aside on the all-currency calendar (`broker_map.replay` passes `news_on=False`). Two assessment cells, RR 1 at 3% and at 4%, 20 coin seeds each plus one short replay, accounts cycled as in the map.

**Statistic.** P3, the pooled pass rate at 3% with its block-clustered standard error, and D = P3 − P4. Read alongside: A, asides per closed trade.

**Null.** The chain at the broker's own cost, 1.44% × (1 + A), A from the same run. At A = 1.33: P3 44.6%, P4 45.7%, D −1.0 points (at A = 0: 47.7%, 48.1%, −0.4).

**Thresholds.** Block SE is 1.7 points at 3% and 1.2 at 4% [broker_map.json], so SE(D) is about 2.1. Downward: P3 more than 2 SE below its null (below 41.2% at A = 1.33) or D below −5.2 points rejects the 3% choice, and the assessment moves to RR 0.75 at 4% (chain 45.2%, 7.5 trades, win 3%, inside the 35% cap), not to RR 1 at 4%. Upward: P3 more than 2 SE above its null (above 48.0%) rejects the aside cost model, and the gate is recomputed from the measured pass rate. For A: A ≤ 0.55 at a 0.5-pip spread passes the gate, above it the purchase waits for a narrowed calendar and a re-measurement.

**Power.** At SE 1.7 the downward test rejects a true shortfall of 3.4 points half the time, 5 points 83%, 6 points 94%. A 2-point shortfall is invisible (power 21%): that is the resolution of 2.8 years, about 24 blocks at 3%. No funded test is declared: the 1.5% cell has 4 blocks and cannot be resolved [paper §14.10].

Unseen by any of this: the broker charges no spread or swap, so the aside's spread half is arithmetic. Every payout is assumed honoured. Whether $100 payouts every few weeks read as trading is a judgement no simulation makes.