Reversal study (flip_spec.md), 2026-11-02. 68 tests, 435 versions; self-test passed.

PART 1, hindsight, not evidence (whole history)
  lost after costs: 257 of 435 (59.1%); of those lost before costs too: 108; lost only to costs: 149
  once reversed, positive after costs: 37 of the 257 losers; 0 of the 149 lost only to costs; 0 of the 178 winners
  all reversed, positive after costs: 37 of 435 (8.5%; each test weighing the same 9.2%)

PART 2, the real test: chosen on development, measured on validation + holdout: 423 eligible versions in 67 tests
  Rule A: 218 versions in 52 tests. Reversed, positive later after costs: 35 = 16.1% (each test the same: 18.1%); before costs 40.8% (45.2%)
    not chosen (205), reversed, after costs: 10.7% (11.0%); before costs 27.8% (24.9%); difference chosen - not chosen 7.1 points, shuffle p 0.0316
    outside the specification: the same 218 simply kept as they are, positive later after costs: 19.7% (19.3%); losing both ways 140; the difference averages -1.2 points under the shuffle
  Rule B (primary): 110 versions in 40 tests. Reversed, positive later after costs: 27 = 24.5% (each test the same: 28.5%); before costs 46.4% (50.6%)
    not chosen (313), reversed, after costs: 9.6% (10.9%); before costs 30.4% (33.7%); difference chosen - not chosen 17.6 points, shuffle p 0.0018
    outside the specification: the same 110 simply kept as they are, positive later after costs: 24.5% (22.1%); losing both ways 56; the difference averages 4.0 points under the shuffle
  Rule C: 4 versions in 4 tests. Reversed, positive later after costs: 1 = 25.0% (each test the same: 25.0%); before costs 75.0% (75.0%)
    not chosen (419), reversed, after costs: 13.4% (13.9%); before costs 34.1% (36.9%); difference chosen - not chosen 11.1 points, shuffle p 0.7467
    outside the specification: the same 4 simply kept as they are, positive later after costs: 0.0% (0.0%); losing both ways 3; the difference averages 10.9 points under the shuffle
  prediction written in the specification: rules A and B under half positive after costs (18.1%, 28.5% with each test the same; 16.1%, 24.5% by count); rule C does best with few versions (25.0%, 25.0% by count; 4 versions)
  not eligible: 12 versions; marked with no gross column: 61, eligible 59

PART 2 again, without the versions whose log has no gross column: 364 eligible versions in 52 tests
  Rule A: 191 versions in 44 tests. Reversed, positive later after costs: 35 = 18.3% (each test the same: 21.4%); before costs 43.5% (49.1%)
    not chosen (173), reversed, after costs: 9.2% (9.5%); before costs 27.7% (24.6%); difference chosen - not chosen 11.9 points, shuffle p 0.0172
    outside the specification: the same 191 simply kept as they are, positive later after costs: 18.8% (17.6%); losing both ways 120; the difference averages 0.9 points under the shuffle
  Rule B (primary): 95 versions in 33 tests. Reversed, positive later after costs: 27 = 28.4% (each test the same: 34.5%); before costs 49.5% (56.8%)
    not chosen (269), reversed, after costs: 8.9% (10.7%); before costs 31.2% (34.3%); difference chosen - not chosen 23.8 points, shuffle p 0.0012
    outside the specification: the same 95 simply kept as they are, positive later after costs: 22.1% (18.2%); losing both ways 47; the difference averages 7.3 points under the shuffle
  Rule C: 3 versions in 3 tests. Reversed, positive later after costs: 1 = 33.3% (each test the same: 33.3%); before costs 100.0% (100.0%)
    not chosen (361), reversed, after costs: 13.9% (14.9%); before costs 35.5% (38.4%); difference chosen - not chosen 18.4 points, shuffle p 0.7498
    outside the specification: the same 3 simply kept as they are, positive later after costs: 0.0% (0.0%); losing both ways 2; the difference averages 18.8 points under the shuffle

PART 3, every reversed version against the six gates (435 more tries: a pass is a candidate, not a result)
  gates passed, reversed: 0: 23, 1: 312, 2: 77, 3: 13, 4: 8, 5: 2. Gate 4 not assessed (no readable percentile): 28. Under 30 trades, inconclusive: 8
  the originals as published, all gates passed: 3 of 435
    g1 validation PF >= 1.2: 9
    g2 50 trades per period: 399
    g3 development t >= 2.5: 0
    g4 above the control: 27
    g5 without the top 10: 14
    g6 holdout positive: 98
  all six: 0; all five assessed where gate 4 cannot be read: 0
  highest development t of any reversed version: 1.60. Passing 4 or 5 gates:
    5 bitcoin-cme-gap-fill CG-S1 [btc, meanrev, LS] fails g3; development t 0.44, validation PF 1.48, holdout 13.49 R, original percentile 0.2
    5 bitcoin-cme-gap-fill CG-S2 [btc, meanrev, LS] fails g3; development t 0.82, validation PF 1.57, holdout 3.35 R, original percentile 0.6
    4 fx-six-strategies-four-pairs USDJPY A2 [fx, meanrev, LS] fails g1,g3; development t 1.04, validation PF 0.99, holdout 544.67 p, original percentile 4.8
    4 fx-six-strategies-four-pairs USDJPY B2 [fx, meanrev, LS] fails g1,g3; development t 0.58, validation PF 1.02, holdout 592.45 p, original percentile 4.2
    4 fx-six-strategies-four-pairs USDJPY BB-L-S [fx, meanrev, L] fails g1,g3; development t 0.81, validation PF 0.98, holdout 2884.90 p, original percentile 2.2
    4 fx-six-strategies-four-pairs USDJPY BB-LS-F [fx, meanrev, LS] fails g1,g3; development t 0.23, validation PF 0.98, holdout 2003.40 p, original percentile 1.4
    4 fx-six-strategies-four-pairs USDJPY BB-LS-S [fx, meanrev, LS] fails g1,g3; development t 0.79, validation PF 1.04, holdout 5214.80 p, original percentile 0.4
    4 fx-six-strategies-four-pairs AUDUSD DC-L-S [fx, breakout, L] fails g3,g5; development t -1.81, validation PF 1.26, holdout 554.50 p, original percentile 0.4
    4 wavetrend-indicator-strategy-gold WT-LS-X [gold, meanrev, LS] fails g1,g3; development t 0.60, validation PF 0.93, holdout 92.65 R, original percentile 0.8
    4 adx-indicator-strategy-forex AUDUSD ADX-20 [fx, trend, LS] fails g3,g4; development t -1.55, validation PF 1.33, holdout 2114.10 p, original percentile 6.6

PART 4, the Part 2 shares at other costs (reversed trades only; count, then each test the same)
  Rule A: no cost 40.8% (45.2%), half 22.5% (22.9%), same 16.1% (18.1%), double 7.8% (10.1%)
  Rule B: no cost 46.4% (50.6%), half 30.9% (35.4%), same 24.5% (28.5%), double 10.0% (15.2%)
  Rule C: no cost 75.0% (75.0%), half 50.0% (50.0%), same 25.0% (25.0%), double 0.0% (0.0%)

PART 5, where (descriptive, no verdict): rule B, reversed, positive later after costs
 by market group
  index: 41 versions in 17 tests, 24.4% (each test the same 41.7%); before costs 48.8%
  gold: 16 versions in 10 tests, 18.8% (each test the same 15.0%); before costs 56.3%
  fx: 48 versions in 11 tests, 25.0% (each test the same 19.5%); before costs 41.7%
  btc: 5 versions in 2 tests, 40.0% (each test the same 33.3%); before costs 40.0%
  spy: 0 versions in 0 tests, n/a (each test the same n/a); before costs n/a
 by family
  trend: 42 versions in 16 tests, 11.9% (each test the same 22.4%); before costs 33.3%
  breakout: 19 versions in 10 tests, 47.4% (each test the same 53.3%); before costs 57.9%
  meanrev: 35 versions in 8 tests, 37.1% (each test the same 43.6%); before costs 57.1%
  pattern: 3 versions in 3 tests, 0.0% (each test the same 0.0%); before costs 33.3%
  no label: 10 versions in 4 tests, 0.0% (each test the same 0.0%); before costs 50.0%
  time: 1 versions in 1 tests, 0.0% (each test the same 0.0%); before costs 0.0%
 outside the specification, by direction of the original
  LS: 77 versions in 35 tests, 31.2% (each test the same 33.8%); before costs 49.4%
  L: 22 versions in 8 tests, 13.6% (each test the same 15.6%); before costs 36.4%
  no label: 10 versions in 4 tests, 0.0% (each test the same 0.0%); before costs 50.0%
  S: 1 versions in 1 tests, 0.0% (each test the same 0.0%); before costs 0.0%
 the 10 most reversible in development (highest t of the reversed trades, after costs), and later
  fibonacci-618-pullback-eurusd FB-LS [fx, LS]: development t 1.60, 69.13 R on 1181 trades; later -161.28 R on 2193 trades, t -2.59
  rsi-30-70-strategy-gold RS-H1-LS [gold, LS]: development t 1.45, 155.82 R on 408 trades; later 275.33 R on 666 trades, t 1.84
  bitcoin-cme-gap-fill CG-NS [btc, LS]: development t 1.19, 39.82 R on 122 trades; later -17.17 R on 128 trades, t -0.80
  fx-six-strategies-four-pairs AUDUSD BB-LS-S [fx, LS]: development t 1.15, 2928.30 p on 2143 trades; later -6901.40 p on 3288 trades, t -3.50
  inside-bar-trading-strategy-eurusd IB-1R [fx, LS]: development t 1.06, 13.85 R on 171 trades; later -6.07 R on 230 trades, t -0.40
  fx-six-strategies-four-pairs USDJPY A2 [fx, LS]: development t 1.04, 1127.82 p on 672 trades; later 476.98 p on 1167 trades, t 0.36
  5-sma-daily-strategy-eurusd SMA-TRAIL [fx, ?]: development t 1.01, 14.99 R on 328 trades; later -0.51 R on 414 trades, t -0.03
  fx-six-strategies-four-pairs GBPUSD DV-L-F [fx, L]: development t 0.94, 1441.57 p on 390 trades; later -1503.97 p on 652 trades, t -1.20
  fx-six-strategies-four-pairs GBPUSD DV-LS-F [fx, LS]: development t 0.89, 1918.08 p on 797 trades; later -2717.80 p on 1184 trades, t -1.59
  fx-six-strategies-four-pairs GBPUSD B2 [fx, LS]: development t 0.88, 998.09 p on 312 trades; later 242.11 p on 522 trades, t 0.26
  positive later: 3 of 10