← MFFU report · extended combo study · post-hoc exploratory · MES · engine data

MFFU Rapid 150K funded — six requested confluence combos, full detail

Arman's six picks (three ★'d on the old M2 Master grid) evaluated on the P1 Fast reBreak MFFU funded method over the full engine window 2023-08-01 → 2026-07-08, with year-by-year splits, the pre-registered-study context for the four cells that were in that grid, and an M2-provenance column answering "did the old potential transfer?". Post-hoc selection on burned data — labeled as such throughout.

All six combos lose to the live stack over the full window. None survive.

Four of six were cells in the pre-registered 36-cell study and failed its kill bars (all first-fail at bar 1). The two newly-run cells are −$11,812 (8-bit take≥5) and −$20,640 (actfloor→c5) full-window. The single positive number anywhere is actfloor→c5's +$1,385 on the 2025+ window — carried by one regime flip (three straight losing years, then +$11,030 in 2026 H1) and far below the $10,901 materiality floor. The live 6-confluence stack with take-AB stands. Engine data only: any future champion still passes TV parity → NT8 verification → ratification before live (prop-firm account protocol).

Master table — full window 2023-08 → 2026-07 (exact dollars, engine $0.74/side)

combonWRexpR PFnetΔ fullΔ 2025+ max EOD DDex-top-5worst day M2 grid net (Δ)status
BASE — the live 6-gate stack
P1 Fast reBreak, MFFU Rapid 150K funded, $1,200 flat risk; legacy grade path (grade_filter='AB' over the live six ≡ count ≥ 4). The null hypothesis.
31233.3%+0.1881.43$68,784$16,204$57,040−$2,700LIVE BASELINE
9-bit stack, take ≥ 5
live six + onbreak + actfloor + engine-gapadr; take iff agreement count ≥ 5
65627.7%+0.0211.04$15,960−$52,823−$11,267$37,838$3,713−$3,608$14,295 (+$3,896)FAILED pre-registered kill bars (first fail: bar 1) — Δ vs live −$11,266.61 over the primary window
8-bit stack, take ≥ 5
live six + onbreak + actfloor (no gapadr); take iff agreement count ≥ 5
33531.9%+0.1461.31$56,971−$11,812−$7,872$19,530$45,222−$2,659$12,865 (+$2,466)post-hoc exploratory; Δ vs live −$7,872.29 over the primary window (−$11,812.46 full window) — no pre-registered kill bar, no FDR q
7-bit stack, take ≥ 4
live six + onbreak; take iff agreement count ≥ 4 (pure superset of the live book)
41330.8%+0.1051.22$51,108−$17,676−$3,244$20,977$39,336−$2,700$13,122 (+$2,723)FAILED pre-registered kill bars (first fail: bar 1) — Δ vs live −$3,243.62 over the primary window
swap: actfloor replaces c6
live six with c6_room swapped out for actfloor; take rule unchanged (count ≥ 4)
69227.2%+0.0071.01$4,938−$63,845−$34,456$41,198−$6,861−$3,609$15,129 (+$4,730)FAILED pre-registered kill bars (first fail: bar 1) — Δ vs live −$34,456.41 over the primary window
swap: onbreak replaces c1
live six with c1_trend swapped out for onbreak; take rule unchanged (count ≥ 4)
22431.2%+0.1451.31$37,221−$31,563−$25,506$11,948$25,481−$2,587$11,236 (+$837)FAILED pre-registered kill bars (first fail: bar 1) — Δ vs live −$25,505.86 over the primary window
swap: actfloor replaces c5
live six with c5_htf swapped out for actfloor; take rule unchanged (count ≥ 4)
53529.5%+0.0781.16$48,143−$20,640+$1,385$30,233$36,344−$3,608$11,669 (+$1,269)post-hoc exploratory; Δ vs live +$1,384.57 over the primary window (−$20,640.49 full window) — no pre-registered kill bar, no FDR q

All M2-grid cells were RULED DEAD by the 2026-08-08 two-judge tribunal (winners-capped-at-1R flips, ex-top-5 concentration, 1-tick slippage); the M2 column is a different method ($250-nominal replay) and is context, not evidence.

Year-by-year PnL (net · n; hover a cell for WR/expR)

combo2023 Aug–Dec202420252026 Jan–Jul
BASE — the live 6-gate stack$3,23834$328112$36,859109$28,35957
9-bit stack, take ≥ 5−$14,67186−$23,319252$24,491199$29,460119
8-bit stack, take ≥ 5−$1,45939$1,086132$37,449106$19,89658
7-bit stack, take ≥ 4−$3,42646−$7,440156$31,889138$30,08473
swap: actfloor replaces c6−$11,92580−$13,898262$9,010224$21,751126
swap: onbreak replaces c1−$8,92226$6,43287$29,63675$10,07636
swap: actfloor replaces c5−$8,58966−$9,869196$27,213172$39,388101

Why the M2 "potential" did not transfer

Nothing about the confluence bits changed between the two studies. What changed is the machine they were bolted onto, in three ways that all push the same direction. (1) THE TAIL. On M2 Master the runner is trailed, so a good added trade can pay 6.04R - and that is literally where the old winners came from: cap winners at 1R and CONFIG_7_k4's +$2,723 becomes -$5,937, actfloor->c6's +$4,730 becomes -$29,882. P1 Fast books a fixed 2R and stops; the highest r in any cell of the new grid is 1.9863, and 0 of the added trades anywhere exceed 2R. The old edge was a lottery ticket the new method does not sell. (2) THE BAR. The added trades were never good in either study - marginal expR 0.0703 / 0.026 / 0.029 R on M2, and -0.0398 / -0.0393 / -0.0831 R on Fast. But M2's base book only earned 0.079 R/trade, so junk volume still added dollars; Fast's base earns 0.333 R/trade, so the same junk volume subtracts. Every one of these cells is a volume trade, and volume only pays when the base is weak. (3) THE PATH. M2's grid was a take-rule replay over a stored ledger with no daily stop - a base trade could only leave by failing the new predicate. The Fast grid re-runs the engine with one position, a 2R daily stop and a $9,500 day cap, so the added trades EVICT good base trades: 8 evictions worth $8,945 on the 9-bit cell (79% of its loss). The swaps are worse still, because dropping a live gate now costs real booked money - actfloor->c6 dropped 17 base rows worth -$548 on M2 (a gain) and 17 base rows worth +$14,705 on Fast (a loss), and onbreak->c1 forfeits 65 base trades worth $28,602 at 0.372 expR. Regime is NOT the explanation for the three big reversals: restricted to the same 2025+ era, M2 still showed +$3,282 / +$1,481 / +$1,321 where Fast shows -$11,267 / -$3,244 / -$34,456. The one place regime does bite is actfloor->c5, which is actually POSITIVE on the matched window (+$1,385) and only negative full-window (-$20,640) on a textbook regime flip: 2023 -$11,827, 2024 -$10,198, 2025 -$9,645, then 2026H1 +$11,030. And the one thing that did NOT change is the count hump: M2's count-4-good / count-5-bad shape (+0.4372 / -0.5794 R on 2025+ candidates) reappears almost exactly in the Fast base book (+0.4368 R / +$74,266 on n=144 vs -0.4304 R / -$10,180 on n=20). Every k>=5 policy keeps those toxic count-5 rows on both methods.

comboM2 ΔFast Δ (2025+) M2 Δ capped@1Rmarginal expR M2 → Fast Fast cascade $ (n)
live6 + onbreak, count>=4n/p¹— → —— (—)
live6 + onbreak + actfloor, count>=5n/p¹— → —— (—)
live6 + onbreak + actfloor + gapadr, count>=5n/p¹— → —— (—)
actfloor replaces c5_htf, count>=4n/p¹— → —— (—)
actfloor replaces c6_room, count>=4n/p¹— → —— (—)
onbreak replaces c1_trend, count>=4n/p¹— → —— (—)

¹ not published in the stored M2 artifacts (cap stress existed only for CONFIG_7_k4 and actfloor→c6). Count-hump check: count-4 hump and count-5 hole are method-invariant — the count-4-good / count-5-bad shape reappears in the Fast base book almost exactly, so the bits behave consistently; the METHOD (fixed 2R vs trailed tails, strong vs weak base, path cascades) is what changed. Premise correction: actfloor→c5 does NOT reverse on the matched 2025+ window (+$1,385 Fast vs +$1,269 M2); its full-window negative is 2023–24 regime, not transfer failure.

Base survival context (side finding; BASE only, H=60 trading days)

variantblow-uppayout attainmentmedian to 1st payout

Fill-model-optimistic sim of the $1,200-risk preset against MFFU's $4,500 intraday trail. This is sizing geometry, not confluences — none of the six combos changes it materially enough to matter, and no combo-vs-base survival comparison was run. Full analysis in the study artifacts.

Caveats