Lucid 25K funded — protocol grid: confluences × management × sizing
First initiation of the prop-firm account protocol on the live champion (P1 Fast reBreak M1be, flat $400, $700 day cap). 40 declared cells: 36 confluence variations (incl. candidates onbreak / actfloor / engine-gapadr), 3 management systems (M1-no-BE / M2 / M3), 1 descriptive sizing shape (taper-down; three other declared shapes were dropped pre-run because prior studies already tested and rejected them — drops printed below). Primary decision window 2025-01-02 → 2026-07-08, paired on an identical signal stream. Design was adversarially reviewed through three rounds (24 findings) before any cell ran; results independently audited.
Every inferential cell first-fails bar 1 (positive in both equal halves + leave-one-month-out stability). Ten cells reach FDR significance — all ten as significant harm (count-threshold hikes and strict removals). The only positive-Δ cell, drop-c4/count≥3 (+$1,391), independently fails bars 2, 3, 4, 5, 8 and 11, flips sign under the causal winners-capped-at-1R stress (−$2,993), retains only 46% of its delta under 1-tick slippage, and breaches the survival guardrail by +12pp. The taper-down sizing cell costs −$419 and +2.0 days to first payout while reducing blow-ups not at all on the raw path. Notable cross-study fact: drop-c4/count≥3 was ALSO the lone positive on the MFFU 150K grid — and died the same deaths there. Engine data only: nothing here touches live; any future champion passes TV parity → NT8 verification → ratification first.
The frozen base (null hypothesis) — primary window, $700-capped, $0.74/side
TV parity: base preset PASS-WITH-NOTES 94.90% on the RAW $400 uncapped book — does not transfer to any modified cell. The $700 day cap BINDS on this account (8 base blocks in-window; the blocked trades would have lost $406 net, so the cap was mildly protective this window).
Δ net vs base — all 40 cells
Kill matrix
| cell | policy | n | Δ net | EH1 | EH2 | q (BH) | first failing bar | |
|---|---|---|---|---|---|---|---|---|
| R | R_minus_c4_count3 | drop c4; take iff count over the remaining five >= 3 (slack-preserving removal) | 195 | +$1,390 | −$338 | +$1,728 | 0.814 | bar 1 — halves + LOMO |
| B | B1_M1_noBE | one-knob delta off the anchor: M1 fixed 2R with NO breakeven stage (isolates the BE stage) | 155 | −$132 | −$1,110 | +$977 | 0.968 | bar 1 — halves + LOMO |
| A | A_plus_all3_count5 | nine-bit stack (live six + onbreak + actfloor + engine-gapadr); count>=5 | 304 | −$618 | −$980 | +$362 | 0.951 | bar 1 — halves + LOMO |
| A | A_plus_onbreak_count4 | add onbreak to the live six; count>=4 over the seven (DUAL READING: algebraically identical to 'rescue live-count-3 signals with onbreak') | 203 | −$753 | −$858 | +$105 | 0.899 | bar 1 — halves + LOMO |
| R | R_minus_c5_count3 | drop c5; take iff count over the remaining five >= 3 (slack-preserving removal) | 336 | −$773 | −$1,607 | +$834 | 0.951 | bar 1 — halves + LOMO |
| S | S_minus_c6_plus_gapadr_count4 | swap c6 out for engine-gapadr; count>=4 over the six | 315 | −$2,005 | −$1,199 | −$806 | 0.859 | bar 1 — halves + LOMO |
| B | B2_M2 | mgmt M2 (half off at +1R, post-entry-swing trail); two-effect change — management REPLACES the BE stage. Read against BASE and against B1. | 165 | −$2,258 | −$3,447 | +$1,190 | 0.814 | bar 1 — halves + LOMO |
| S | S_minus_c1c6_plus_onbreak_gapadr_count4 | swap BOTH c1 and c6 out for onbreak+engine-gapadr; count>=4 over the six | 249 | −$2,517 | +$625 | −$3,141 | 0.814 | bar 1 — halves + LOMO |
| A | A_plus_gapadr_count4 | add engine-gapadr to the live six; count>=4 over the seven (DUAL READING: algebraically identical to 'rescue live-count-3 signals with engine-gapadr') | 336 | −$2,558 | −$961 | −$1,597 | 0.817 | bar 1 — halves + LOMO |
| A | A_and_actfloor | legacy AB grade rule AND actfloor==1 (candidate as a hard gate) | 116 | −$2,803 | +$393 | −$3,195 | 0.502 | bar 1 — halves + LOMO |
| A | A_plus_actfloor_count5 | add actfloor to the live six; count>=5 over the seven (raise-the-bar) | 121 | −$3,910 | −$48 | −$3,862 | 0.442 | bar 1 — halves + LOMO |
| R | R_minus_c6_count4 | drop c6; take iff count over the remaining five >= 4 (strict subset) | 139 | −$4,278 | −$555 | −$3,723 | 0.278 | bar 1 — halves + LOMO |
| S | S_minus_c6_plus_onbreak_count4 | swap c6 out for onbreak; count>=4 over the six | 185 | −$4,615 | −$1,762 | −$2,852 | 0.442 | bar 1 — halves + LOMO |
| R | R_minus_c3_count3 | drop c3; take iff count over the remaining five >= 3 (slack-preserving removal) | 244 | −$6,177 | −$2,581 | −$3,596 | 0.442 | bar 1 — halves + LOMO |
| R | R_minus_c1_count3 | drop c1; take iff count over the remaining five >= 3 (slack-preserving removal) | 318 | −$6,363 | −$5,747 | −$616 | 0.500 | bar 1 — halves + LOMO |
| A | A_plus_gapadr_count5 | add engine-gapadr to the live six; count>=5 over the seven (raise-the-bar) | 104 | −$6,407 | −$1,766 | −$4,641 | 0.305 | bar 1 — halves + LOMO |
| S | S_minus_c1c6_plus_onbreak_actfloor_count4 | swap BOTH c1 and c6 out for onbreak+actfloor; count>=4 over the six | 275 | −$6,449 | −$2,554 | −$3,895 | 0.504 | bar 1 — halves + LOMO |
| S | S_minus_c1_plus_actfloor_count4 | swap c1 out for actfloor; count>=4 over the six | 253 | −$6,694 | −$2,956 | −$3,737 | 0.500 | bar 1 — halves + LOMO |
| S | S_minus_c1_plus_gapadr_count4 | swap c1 out for engine-gapadr; count>=4 over the six | 228 | −$6,798 | −$1,908 | −$4,891 | 0.442 | bar 1 — halves + LOMO |
| R | R_minus_c2_count3 | drop c2; take iff count over the remaining five >= 3 (slack-preserving removal) | 279 | −$6,974 | +$407 | −$7,381 | 0.442 | bar 1 — halves + LOMO |
| A | A_plus_all3_count6 | nine-bit stack (live six + onbreak + actfloor + engine-gapadr); count>=6 | 120 | −$7,132 | −$1,029 | −$6,103 | 0.389 | bar 1 — halves + LOMO |
| A | A_and_gapadr | legacy AB grade rule AND engine-gapadr==1 (candidate as a hard gate) | 99 | −$7,155 | −$3,313 | −$3,842 | 0.298 | bar 1 — halves + LOMO |
| A | A_plus_actfloor_count4 | add actfloor to the live six; count>=4 over the seven (DUAL READING: algebraically identical to 'rescue live-count-3 signals with actfloor') | 353 | −$7,243 | −$6,042 | −$1,201 | 0.500 | bar 1 — halves + LOMO |
| S | S_minus_c1_plus_onbreak_count4 | swap c1 out for onbreak; count>=4 over the six | 108 | −$8,345 | −$3,622 | −$4,723 | 0.278 | bar 1 — halves + LOMO |
| R | R_minus_c6_count3 | drop c6; take iff count over the remaining five >= 3 (slack-preserving removal) | 407 | −$9,830 | −$7,453 | −$2,377 | 0.442 | bar 1 — halves + LOMO |
| B | B3_M3 | mgmt M3 (BE at +1.5R, 2xATR post-BE trail); two-effect change — management REPLACES the BE stage. Read against BASE and against B1. | 164 | −$9,905 | −$10,045 | +$139 | 0.208 | bar 1 — halves + LOMO |
| S | S_minus_c6_plus_actfloor_count4 | swap c6 out for actfloor; count>=4 over the six | 334 | −$10,024 | −$6,548 | −$3,476 | 0.430 | bar 1 — halves + LOMO |
| S | S_minus_c1c6_plus_actfloor_gapadr_count4 | swap BOTH c1 and c6 out for actfloor+engine-gapadr; count>=4 over the six | 397 | −$10,383 | −$5,767 | −$4,616 | 0.442 | bar 1 — halves + LOMO |
| K | K_live6_count3 | live six unchanged; move the count threshold to >= 3 | 427 | −$10,403 | −$6,815 | −$3,587 | 0.442 | bar 1 — halves + LOMO |
| R | R_minus_c5_count4 | drop c5; take iff count over the remaining five >= 4 (strict subset) | 79 | −$12,497 | −$3,272 | −$9,225 | 0.026 | bar 1 — halves + LOMO |
| A | A_and_onbreak | legacy AB grade rule AND onbreak==1 (candidate as a hard gate) | 30 | −$14,787 | −$4,998 | −$9,789 | 0.036 | bar 1 — halves + LOMO |
| R | R_minus_c1_count4 | drop c1; take iff count over the remaining five >= 4 (strict subset) | 75 | −$15,963 | −$5,418 | −$10,545 | 0.026 | bar 1 — halves + LOMO |
| K | K_grade_A_legacy | legacy grade path with grade_filter='A' (A-only; the HTF grade amendment binds) | 98 | −$15,980 | −$5,924 | −$10,056 | 0.007 | bar 1 — halves + LOMO |
| A | A_plus_onbreak_count5 | add onbreak to the live six; count>=5 over the seven (raise-the-bar) | 48 | −$18,172 | −$5,939 | −$12,234 | 0.004 | bar 1 — halves + LOMO |
| R | R_minus_c3_count4 | drop c3; take iff count over the remaining five >= 4 (strict subset) | 46 | −$20,025 | −$8,262 | −$11,763 | 0.002 | bar 1 — halves + LOMO |
| R | R_minus_c2_count4 | drop c2; take iff count over the remaining five >= 4 (strict subset) | 49 | −$21,230 | −$8,608 | −$12,622 | 0.004 | bar 1 — halves + LOMO |
| K | K_live6_count6 | live six unchanged; move the count threshold to >= 6 | 3 | −$21,755 | −$7,516 | −$14,239 | 0.004 | bar 1 — halves + LOMO |
| K | K_live6_count5 | live six unchanged; move the count threshold to >= 5 | 23 | −$24,655 | −$8,456 | −$16,199 | 0.002 | bar 1 — halves + LOMO |
| R | R_minus_c4_count4 | drop c4; take iff count over the remaining five >= 4 (strict subset) | 24 | −$24,683 | −$8,484 | −$16,199 | 0.002 | bar 1 — halves + LOMO |
FDR — significant harm only
Family = the 39 inferential cells; contiguous-week block bootstrap, 20,000 resamples, two-sided centered p at full precision; all ten q≤0.10 cells are NEGATIVE. The lone positive cell sits at qBH = 0.81. Week-vs-month sensitivity published in the artifacts.
| significantly harmful cell | q (BH) | q (BY) | Δ net |
|---|---|---|---|
| K_live6_count5 | 0.0019 | 0.0083 | −$24,655 |
| R_minus_c3_count4 | 0.0019 | 0.0083 | −$20,025 |
| R_minus_c4_count4 | 0.0019 | 0.0083 | −$24,683 |
| A_plus_onbreak_count5 | 0.0036 | 0.0152 | −$18,172 |
| K_live6_count6 | 0.0036 | 0.0152 | −$21,755 |
| R_minus_c2_count4 | 0.0036 | 0.0152 | −$21,230 |
| K_grade_A_legacy | 0.0067 | 0.0285 | −$15,980 |
| R_minus_c1_count4 | 0.0262 | 0.1115 | −$15,963 |
| R_minus_c5_count4 | 0.0262 | 0.1115 | −$12,497 |
| A_and_onbreak | 0.0363 | 0.1543 | −$14,787 |
Management axis — what M1/M2/M3 cost on this account
BE-at-+1R is nearly free to remove (−$132 — a wash, but LOMO-unstable at 8/19). M2 costs −$2,258 and M3 −$9,905 on the primary window. Per-exit-reason decompositions are in the artifacts. Full-window context (sealed until after the kill matrix froze): M2/M3 were positive full-window but negative in the decision era — the exact regime-rescue pattern the seal exists to catch.
The lone positive cell — and its seven failures
- Bar 1: EH1 −$338 / EH2 +$1,728 — boundary-INsensitive (identical split under all four tested half boundaries: False).
- Bars 2/3/4/5: concentration, temporal, tail and quality failures (details in kill_matrix.json).
- Bar 6: causal winners-capped-at-1R flips the sign (Δ −$2,993).
- Bar 7: slippage retention 46% (< 50% floor).
- Bar 8: +$1,390 vs the $3,634 floor.
- Survival guardrail: blow-up +11.96pp vs base — breached.
Same cell, same profile, second account: on the MFFU 150K grid this exact policy was also the only positive and died the same way. The c4-sweep bit is load-bearing at the take line on this method family.
Survival guardrails (Amendment P-1 diagnostics)
| cell | blow-up (60d, intraday, $0.50) | vs base | guardrail |
|---|---|---|---|
| A_plus_all3_count5 | 89.9% | +37.1 pp | DEAD (guardrail breached) |
| B1_M1_noBE | 79.1% | +26.4 pp | DEAD (guardrail breached) |
| R_minus_c4_count3 | 64.7% | +12.0 pp | DEAD (guardrail breached) |
Base null: blow-up 52.8% intraday / 41.7% EOD; attainment 47.2% / 58.3% (326 starts, H=60 trading days). Degraded stresses are brutal at this sizing: PnL-halved blow-up 96.3%, winners-capped 100% — at $400 risk against a $1,000 EOD trail the account survives only if the edge performs near full strength. The design-time null figures were found defective at audit and are superseded by these (full decomposition in the artifacts). Fill-model-optimistic; ~5.6 independent 60-day windows — rate resolution is coarse and stated wherever compared.
Sizing (axis C, descriptive only)
Caveats & provenance
- Fill model optimistic (touch fills, zero slippage, entry-bar SL/TP unchecked in fill minute). Sim ≠ live.
- Burned data; post-selection era; exploratory hypothesis-testing — not a forward test.
- Modified cells are NOT TV-verified; protocol Stages 3–5 (TV parity → NT8 on latest data → ratification) gate any change.
- MES only; single method family; ~18.2-month decision window.
- `gapadr` here is the ENGINE port (denominator = engine strategies_p1._ors_daily_atr20, per design Amendment 1 §3a). It is NOT bit-identical to the Courtroom-screened b_gapadr: 23/2836 (0.81%) of emitted signals grade dif…
- Nine dense axis-A cells: the $0.50 vs $0.74 cap-decision bases differ by exactly one trade (decision basis is $0.50 throughout; disclosed per cell).
- The survival walker mirrors M1/M2 paths only — bar 9 is structurally uncomputable for an M3 cell (never fired here; disclosed).
- MANIFEST's 390-day equal halves are superseded by the analysis's 385-day NYSE+traded-holiday spine (EH1→2025-10-07); per-cell primary_window_by_half blocks inherit the old cut.