Adopted is not deployed. The order of work: measure realised per-leg slippage on the running stack — the reading is now two-sided (under 0.68 the seven were fine anyway; 0.68–0.99 this adoption is the only measured fix; above 0.99 shrink the account). eu_balance_giveback needs a live warm-up mechanism that does not exist yet (250 complete sessions of causal state; a cold restart replays 2 GiB of bars for ~5 minutes, and an unconfigured warm-up silently never arms). settle_cross_europe needs the CL roll schedule shipped in the deploy bundle — today the bundle excludes it and the module cannot even be constructed on the VPS — plus every CL blocker the crude release book already carries. Both books trade 06:00–08:00 ET, an hour the paper stack has never traded: paper-first is the staging. And one unrelated live defect found on the way: the NQ catalog has zero bars for 2026-08-24/25, an interior hole the live warm-up cannot fill.

If the account ever grows, re-shape before adding anything. The same books re-weighted for a 150k account are worth $2,704 a month (+70%) with no new book at all; at 300k, $3,914. The deployed shape is optimal only at exactly 100k — the integer floor wastes nothing there and a great deal everywhere else.

Adopted is not deployed — and the slippage cliff now gates the standing account, not just the new book. Measured on the ladder: the six-book vector goes infeasible above ~0.63 extra ticks per slipping leg, this one above ~0.67. Measuring realised per-leg slippage on the running NQ stack is job one and costs nothing; above ~0.6 the correct action is to shrink the account, not add books. The print_break bump is deployable today — one config line plus the mandatory guards re-derivation to 1,300/1,300, because at the old 1,100 the disaster stop sits inside the 7-lot book’s own worst excursion.

What the crude book is, and what it still has to clear. It survived all five gates on a genuine holdout and a 2–1 skeptic panel — the one refutation was release-second execution realism: 79% of its entries fill in the four seconds after 10:30:00, its exit legs are pinned to the level rather than the fill, so entry slippage is unrecoverable one-for-one. The honest live band is 4.74–6.18 net ticks (fills through the target carry 27.7% of PnL, and CL’s monthly roll holds 11.6% of PnL on election dates a live guard would flatten). Before it trades: the exec_client order-list defect (entry straddles are wrongly marked reduce-only — a live-only bug no backtest can see) must be fixed; NYMEX data entitlement and MCL in NT8 confirmed; the installers taught per-book instruments; and an EIA holiday-calendar rule built — 46 of 340 Wednesdays are reschedule weeks where the release moves to Thursday and the module correctly stands down.

Where the money is if the account grows: at 150k–300k the same crude book is worth +$384 to +$522 a month, because those accounts waste a quarter to three-quarters of a fractional unit that its cheap $97-per-unit drawdown soaks up.

Adopted is not deployed. The running stack — paper and the NT8 VPS — is still the four NQ books at 6,6,6,6 until, in order: realised per-leg slippage is measured on the running stack (it decides between the full vector, the fallback, and no gold leg); two yes/no questions are answered — COMEX entitlement on the Databento subscription, and MGC 12-26 resolving in NT8; run_live.py learns a second instrument (today it hard-refuses anything that is not MNQ, and --root GC steers only the feed — used alone it would price gold risk at $2 a point on a $10-a-point contract); and stall_fade’s paper book is restarted — adopted 2026-08-16, never yet run live, its 02:00–06:00 ET thin-book fills unverified. The one change deployable today is dwm 6→7.

What the gold book is, and what it is not. It survived a genuine 3.5-year holdout (+2.96 net ticks against gold’s own 1.06 floor on 350 untouched trades), a 3–0 skeptic panel, and the anchor falsification — moved one or two hours off the pit open the same rule loses decisively in every direction. It is also the most concentrated book ever admitted: 22% of its PnL is the single 08:30:00–08:30:59 minute, 96% rides trades held across the macro print, and it breaks even if release-minute fills cost ~8 extra ticks. It is a second bet on the same print print_break trades, hedged by being on gold — and it is the only book of the six that can be suspended unilaterally: dropping it changes account drawdown by $0 to the dollar.

One tool verdict from the same run: portfolio_search.py’s ranking is unusable for this mixed-contract set — its top vectors collapse 33–50% on the 1-second path because the maximiser parks on zero-headroom cells. The frontier behind this adoption was built directly on equity_path.py, 59 true-path rows.

Error one: every candidate was measured at the wrong size. The screen adds one contract of a candidate against one of each incumbent, and the allocation search that should have caught it capped total weight at 16 — which excludes 6,6,6,1 before enumeration. Held at a third of the incumbents' size, a book adds a third of the drawdown. dryup_resume, rejected on drawdown, became the fourth book at 2 MNQ.

Error two: slippage was charged twice. The backtest already fills through a model that slips a stop entry one tick half the time, and the trade log is net of it. The account was then charged another 0.5–0.75 ticks per leg on top. That double count was costing ~31% of reported income — $647 a month against the true $847 on four books. Every figure on this page is now on engine fills only; the overlay remains as a sensitivity, which is a different question from what the account earns.

Why twenty-two rejections need re-reading. Every candidate ever screened here was measured at equal weight with the incumbents — one contract of the candidate against one of each deployed book. Held at a third of that size, the same book adds a third of the drawdown. portfolio_search.py was built to ask exactly this and could not: its --max-total 16 cap drops 6,6,6,1 (total 19) before enumeration, and the two strongest candidates ever measured post-date the last allocation run entirely.

So the “$616 of free drawdown” below is the budget for a book held at the incumbents’ own size. That was never the only size on offer.

A second correction moves every number on this page. The marked equity path was charging the $4.50 NQ round turn already netted into each trade, while the account trades micros — ten of which cost $11.20. Charged properly, the trio’s drawdown is $5,927, not $4,814: 5.06 contracts at half a tick of slippage, four at three quarters. The +23.1% that moves it by matches the figure CLAUDE.md recorded for this bias, exactly.

What is still open. Corrected 2026-08-16: an earlier version of this page said print_break earns only +1.95 net ticks on the quietest fifth of days. That figure came from the proposal's own quintile definition and was repeated without re-deriving it on the account's terms. Measured the way every other number here is measured — CME session day, deployed fill convention — it is +5.8 in the quietest quintile and +11.4 in the second, contributing $21,172, the most of any of the five books. The dead zone is real but smaller than stated: the account earns 15% of its PnL on 40% of its days, and it earns 8x more per day in the widest quintile than the quietest. That gap is what a sixth book is for. It also carries a real execution cost — in one month of sample the entry stop filled into the print, the protective leg was rejected because price was already through it, and the position was flattened 22 points past the intended stop. It trades the fastest second of the day; that will recur.

dryup_resume fails the concentration gate — its top ten trades are 86% of its PnL. Sizing dilutes the consequence to roughly $133 a month of exposure; it does not fix the property. If that tail does not recur, the drawdown reduction goes with the income, because the reduction is earned by making money during the trio’s declines rather than by sitting them out.

Stacking it with turn_of_month measures $815 a month and wins the holdout as well — deliberately not recommended yet, since two books each rejected on their own double the fitting exposure. The charts below now plot dryup_resume alongside the deployed three, each at one contract.