Unified test table — Topstep · MFFU · Lucid · FundedNext

Every campaign-level test cell from the four conveyor studies in one filterable table, in the Topstep stress-table format: each row shows the base result beside the winners_half result (every winning trade halved — the standard “is this an edge or a tail?” shock) for the same cell, plus the 1-tick-slippage median where that stress arm exists. Numbers are copied verbatim from the study artifacts — nothing is recomputed here. Lucid rows are the 2026-08-01 growth-conveyor rerun (start 1 eval, grow on terminal events, Lucid household caps 10E/5F) so all three firms now read in the same fleet-total units; the earlier per-slot/per-pair Lucid frameworks remain at /trading/research/lucid. FundedNext rows were added 2026-08-05 and cover a single 7-month 2026 window — pick 7m in the Arm filter to see them (the default 12m basis does not exist for that study). Full write-ups: Topstep · MFFU · Lucid.

Read this before comparing across firms

  1. Units are now comparable but protocols still differ where the firms differ. All rows are fleet-total bankNet (net of every fee) over the same window (2025-01→2026-07) and cohort grid. Topstep = 4 evals from day one, never retire; MFFU = start 1, grow on terminal events, 10-account pool; Lucid = start 1, grow on terminal events, 10-eval/5-funded household caps, one-time eval fees instead of monthly subscriptions (a real structural difference, not a modeling choice). The Lucid conveyor also parks unlimited dormant passed evals (a median cohort ends with dozens queued) — per Lucid's rules, but a large share of fee spend buys accounts that never get a funded seat.
  2. Lucid Daily 150K tops the base table by ruleset, not edge: it trades the byte-identical book as PRO 150K r1200 — the lead is uncapped daily sweeps at hot sizing (72% funded breach rate) with next-day seat refill. It loses 68% to one tick of slippage and the source study ruled the comparison NOT IDENTIFIED. Read its row notes before acting on it.
  3. Almost nothing here is TV-verified. The exceptions are the one Topstep 50K path cell marked in notes and the FundedNext signal book (P1-Fast|M1be+RB), which passed an independent TradingView Pine referee on 2026-08-05 (PASS-WITH-NOTES, 96.2%/96.5% entries matched, median per-trade PnL difference ≈ $0). That covers the BOOK only — not the campaign ladder, not the eval taper, not the DLL day-stop, and the same run measured the engine as ~5% optimistic at the eval stage in this window. Optimistic fills (touch fills, zero net slippage in base), no holdout, MES-only, overlapping cohorts (12m medians ≈ 1–3 independent observations).
  4. winners_half is a sign-flip test, not a haircut — it kills every arm at every firm. A cell whose wh column is deep red and whose base is large is a bet on full edge persistence.
  5. FundedNext rows are a different shape and must not be ranked beside the rest. They cover one 7-month window (2026-01→2026-07, 51 book days, 61 trades) that was seen during strategy selection and contains the favourable Nov-2025→Apr-2026 regime — there is no 12m box to compare on. The six 50K speed rows are cross-start medians over 21 overlapping January-2026 starts; the three size-anchor rows (25K/50K/100K) are single campaigns, n=1, so their “median” columns are headline values — each says so in its own note. Their primary metric is days-to-first-payout, not net. winners_half pays zero on every FundedNext cell.
  6. Lottery cells (funded breach rate ≥ 0.95 per stress) are draws over start days, not measurements — hidden by default.
  7. Horizon labels are normalized across firms (Topstep h6/h12 and MFFU h6/h12/m18 shown as 6m/12m/18m; Topstep's "full" ≈ 17.5-month Jan-2025 starts, n=21, stays its own label). 12m is the ranking basis everywhere and is the default filter; 18m boxes are tiny-n (MFFU n=5, Lucid 1 window) and the Lucid Daily 18m cells are exactly the "not identified" tail the daily study rejected — don't rank on them. 6m is not half of 12m.