Cost study across our backtests, 2026-10-22T13:02:40.944Z
Inventory: 54 tests, 387 variants, 0 excluded

Check 1. Trade logs against the published summaries (tolerance: the rounding of the printed CSV values, n x half a unit of the last printed digit)
  rows with an unquoted comma in a text cell, read from the right: 2 (2026-04-09,2026-10-20 (end of data, clos; 2026-05-01,2026-10-20 (end of data, clos)
  [ok] trade log found for 387 of 387 variants
  trade count matches: 387; net matches: 387; gross matches: 350; logs without a gross column (gross checked through check 2 only): 37

Check 2. Is the cost a per-trade amount (N(k) = G - kC exact)?
  golden-cross-50-200: per-trade cost in percent from 0.0070 to 0.0720 (4 variants with both columns)
  intraday-trend-pullback-bitcoin: per-trade cost in percent from 0.0926 to 0.1077 (2 variants with both columns)
  noise-area-breakout-bitcoin: per-trade cost in percent from 0.0804 to 0.1084 (4 variants with both columns)
  accumulation-distribution-indicator-spy: per-trade cost in percent from 0.0400 to 0.0400 (3 variants with both columns)
  accumulation-distribution-indicator-bitcoin: per-trade cost in percent from 0.0848 to 0.1535 (3 variants with both columns)
  heikin-ashi-strategy-bitcoin: per-trade cost in percent from 0.0853 to 0.1185 (3 variants with both columns)
  best-time-to-trade-gold: per-trade cost in percent from 0.0054 to 0.0784 (3 variants with both columns)
  200-day-moving-average-strategy-spy: per-trade cost in percent from 0.0400 to 0.0400 (2 variants with both columns)
  fee formulas read in the libraries (ha, ad, tp, na_btc, adspy, sma200, ma, gn): every cost is a fixed amount or fee x (entry + exit), so it is linear in the cost level and N(k) = G - kC holds exactly; the spread above comes from the exit price, not from compounding
  [ok] no trade with a negative cost (gross below net) in the logs
  [ok] total cost C = G - N is never negative in the summaries

Check 3. Gold variants against the gold hub (gs_data.js)
  [ok] hub 97 versions; matched in the inventory 97
  [ok] matched hub versions: 42 lost to costs (hub 42), 13 lost before costs (hub 13)
  gold variants outside the hub: 0

Check 4. Cause labels in the articles
  [ok] 70 variants labelled "Costs" in cause tables, all in the class "lost to costs"

All checks passed.

Question 1. How often did the cost model decide the verdict?
  variants: 387; lost before costs 95 (25%); lost to costs 131 (34%); survived costs 161 (42%); passed all six gates 3
  tests (majority of variants): 54; before 12, costs 14, survived 23, ties 5
  Index futures (ES, NQ): 117 variants; before 32, costs 27, survived 58
  Gold (spot and GC futures): 97 variants; before 13, costs 42, survived 42
  FX: 149 variants; before 46, costs 56, survived 47
  Bitcoin: 19 variants; before 4, costs 6, survived 9
  SPY: 5 variants; before 0, costs 0, survived 5

Question 2. Share of the gross profit taken by costs (variants with G > 0)
  all: n 292, median 87%, 25th 35%, 75th 228%
  Index futures (ES, NQ): n 85, median 47%, 25th 21%, 75th 226%
  Gold (spot and GC futures): n 84, median 99%, 25th 38%, 75th 183%
  FX: n 103, median 113%, 25th 71%, 75th 263%
  Bitcoin: n 15, median 59%, 25th 14%, 75th 314%
  SPY: n 5, median 19%, 25th 1%, 75th 23%

Question 3. Trades per year against cost share, outside the gold hub (prediction: positive)
  n 208, Spearman rho 0.460, one-sided permutation p 0.0000; with the hub included: n 292, rho 0.539
  0.3 to 84.6 trades a year: n 69, median cost share 46%, lost to costs 19
  84.7 to 230.6 trades a year: n 70, median cost share 80%, lost to costs 30
  231.6 to 1819.5 trades a year: n 69, median cost share 126%, lost to costs 40

Question 4. Variants with a positive net at k times our cost
  k=0: 292; k=0.5: 212; k=1: 161; k=1.5: 111; k=2: 93
  Index futures (ES, NQ): 0: 85, 0.5: 63, 1: 58, 1.5: 48, 2: 44
  Gold (spot and GC futures): 0: 84, 0.5: 64, 1: 42, 1.5: 28, 2: 26
  FX: 0: 103, 0.5: 70, 1: 47, 1.5: 21, 2: 11
  Bitcoin: 0: 15, 0.5: 10, 1: 9, 1.5: 9, 2: 7
  SPY: 0: 5, 0.5: 5, 1: 5, 1.5: 5, 2: 5
Cost study, independent second computation
  [ok] variants: 387 / 387
  [ok] variants before: 95 / 95
  [ok] variants costs: 131 / 131
  [ok] variants survived: 161 / 161
  [ok] variants passing all six gates: 3 / 3
  [ok] index before: 32 / 32
  [ok] index costs: 27 / 27
  [ok] index survived: 58 / 58
  [ok] gold before: 13 / 13
  [ok] gold costs: 42 / 42
  [ok] gold survived: 42 / 42
  [ok] fx before: 46 / 46
  [ok] fx costs: 56 / 56
  [ok] fx survived: 47 / 47
  [ok] btc before: 4 / 4
  [ok] btc costs: 6 / 6
  [ok] btc survived: 9 / 9
  [ok] spy before: 0 / 0
  [ok] spy costs: 0 / 0
  [ok] spy survived: 5 / 5
  [ok] tests before: 12 / 12
  [ok] tests costs: 14 / 14
  [ok] tests survived: 23 / 23
  [ok] tests tie: 5 / 5
  [ok] Q2 n: 292 / 292
  [ok] Q2 all median: 0.8666098068765773 / 0.8666098068765773
  [ok] Q2 all p25: 0.3518082211102017 / 0.3518082211102017
  [ok] Q2 all p75: 2.2810181278971253 / 2.2810181278971253
  [ok] Q2 index median: 0.4672205247045387 / 0.4672205247045387
  [ok] Q2 gold median: 0.9888074650542313 / 0.9888074650542313
  [ok] Q2 fx median: 1.1303898786470232 / 1.1303898786470232
  [ok] Q2 btc median: 0.5864978018491181 / 0.5864978018491181
  [ok] Q2 spy median: 0.19455916137103252 / 0.19455916137103252
  [ok] Q3 n: 208 / 208
  [ok] Q3 rho: 0.4602481142954178 / 0.4602481142954178
  [ok] Q3 rho with the hub: 0.5390657078015039 / 0.5390657078015039
  [ok] Q4 k=0: 292 / 292
  [ok] Q4 k=0.5: 212 / 212
  [ok] Q4 k=1: 161 / 161
  [ok] Q4 k=1.5: 111 / 111
  [ok] Q4 k=2: 93 / 93

IDENTICAL