Reversal study self-test, 2026-11-02T17:08:48.470Z

1. By hand
A. a reversed trade: gross 10, net 7 (cost 3) -> -13; gross -10, net -13 (cost 3) -> +7; half the cost +8.5, double +4; no cost +10
  [ok] as worked
B. the t-statistic and the profit factor: [1, 2, 3, 4] -> t = 2.5 / (1.29099 / 2) = 3.87298; PF of [3, -1, 1, -1] = 2; no losses = Infinity; nothing = 0
  [ok] t-statistic
  [ok] profit factor
C. the periods at their boundaries, and the market group of a version
  [ok] index futures
  [ok] spot gold
  [ok] FX
  [ok] bitcoin
  [ok] SPY
  [ok] market groups
D. a version by hand: 40 development trades (gross -2, net -3), 30 validation (gross +1, net 0), 30 holdout (gross -4, net -5)
  [ok] totals
  [ok] reversed: development 40, later 30
  [ok] half and double the cost
  [ok] chosen by rules A and B; no spread in development, so t = 0 and not rule C
  [ok] 29 development trades: not eligible
  [ok] 29 later trades: not eligible
  [ok] rule C with t -12.49
  [ok] a mild loser with t -1.48: rule B, not rule C
  [ok] a version losing only to costs: rule A, not B
E. the gates: reversed trades that win everywhere; a published percentile of 3 reads 97, of 6 reads 94, none is not assessed; the top 10 removed; the holdout
  [ok] all pass
  [ok] gate 4 is 100 minus the published percentile
  [ok] under 50 trades in a period: gate 2 fails
  [ok] positive only through its 10 best trades: gate 5 fails
  [ok] validation PF 1.10: gate 1 fails
  [ok] holdout negative: gate 6 fails
  [ok] development t 1.48: gate 3 fails
F. each test weighs the same; the shuffle stays inside each test
  [ok] test-weighted 45%, not 9 of 11
  [ok] perfect separation: difference 1, p 0
  [ok] perfect separation the other way: difference -1, p 0
  [ok] no relation: difference 0, p 0.597
  [ok] with one version per test, the labels cannot move

2. Every log against its published summary
  68 tests, 435 versions; no gross column: 61 versions in 15 tests; a published control percentile: 407
  [ok] trades, net and gross match the summaries (rounding tolerance) for 435 of 435
  [ok] trades per period against the published summaries: 265 versions checked, 265 exact, 0 off by more than 1%

3. The real mirror pair on the site: VW-TREND reversed by the formula against VW-REV, trade by trade
  [ok] 11707 trades against 11707; largest difference $0.0000 (printed to the cent)
  [ok] gate 4 as 100 minus the percentile: VW-TREND 55.4 and VW-REV 44.6 add up to 100

4. The independent reference on every version
  [ok] 435 of 435 versions identical

PASSED