The Bitcoin CME gap fill (spec cg_spec.md): spot BTCUSDT at the CME weekly close (Friday 17:00 New York) and reopen (Sunday 18:00). Run 2026-10-21T15:45:59.071Z; 500 control runs per variant. Trades: |gap| >= 1%, toward the gap, 0.05% a fill.
data: 4789278 minutes 2017-08-17 04:00 .. 2026-09-30 23:59 UTC; 458 weekends measured (from Friday 2017-12-22)
  gap size: median |gap| 1.14%, 54.6% of weekends 1% or more, largest 13.58%
  fills (all: 458 gaps): within 24 h 56.3%, 7 days 80.3%, 30 days 89.1%, by the end of the data 96.9%; median 16.47 hours to fill
  fills (big: 250 gaps): within 24 h 38.4%, 7 days 70.0%, 30 days 83.6%, by the end of the data 96.0%; median 43.45 hours to fill
  fills (up: 148 gaps): within 24 h 35.1%, 7 days 68.9%, 30 days 81.1%, by the end of the data 93.2%; median 42.63 hours to fill
  fills (down: 102 gaps): within 24 h 43.1%, 7 days 71.6%, 30 days 87.3%, by the end of the data 100.0%; median 45.82 hours to fill

=== CG-S1: stop 1x the gap, target Friday's close, out by the next Friday 17:00 New York ===
  DEV      n= 122 net -14.17 R  avg -0.116 R  PF 0.79 t -1.31
  VAL      n=  65 net -18.60 R  avg -0.286 R  PF 0.55 t -2.38
  HOLDOUT  n=  63 net -20.51 R  avg -0.326 R  PF 0.52 t -2.65
  ALL      n=250 net -53.28 R avg -0.213 R (gross -0.167 R, cost 0.046 R a trade); net % of price a trade -0.551%; PF 0.64 t -3.45
  winners 41.2% (95% 35.3-47.4%); ended by target 100 (40.0%), stop 142 (56.8%), Friday close 8 (3.2%); hours held median 11.10; without the 10 best -63.16 R
  longs (gap down) n=102 net -31.13 R | shorts (gap up) n=148 net -22.15 R
  by year (n, net R): 2017 2 -0.05 | 2018 38 -2.29 | 2019 33 -8.36 | 2020 29 -2.81 | 2021 36 1.39 | 2022 26 -8.36 | 2023 23 -12.28 | 2024 26 -9.42 | 2025 22 -7.16 | 2026 15 -3.93
  control (entries moved to other days, same time of day, same distances, 500 runs, 2 s): mean trades 249.92, median avg -0.048 R, 5-95% -0.144 .. 0.060 R; real avg -0.213 R at percentile 0.20
  [FAIL] 1 PF >= 1.2 in validation: VAL PF 0.55
  [PASS] 2 >= 50 trades in each period: 122 / 65 / 63
  [FAIL] 3 t >= 2.0 on development: DEV t -1.31
  [FAIL] 4 above 95th pct of the control: percentile 0.20
  [FAIL] 5 positive without top-10 trades: -63.16 R
  [FAIL] 6 holdout net positive (flagged used): HOLDOUT net -20.51 R PF 0.52
  VERDICT Buried (1/6 gates)
  rule 3: 4 trades touch a data gap of 10 minutes or more (btc_gaps.json); without them avg -0.202 R, gates 1/6 -> Buried (verdict unchanged)

=== CG-S2: stop 2x the gap, target Friday's close, out by the next Friday 17:00 New York ===
  DEV      n= 122 net -10.91 R  avg -0.089 R  PF 0.75 t -1.44
  VAL      n=  65 net -13.44 R  avg -0.207 R  PF 0.55 t -2.26
  HOLDOUT  n=  63 net -6.87 R  avg -0.109 R  PF 0.72 t -1.22
  ALL      n=250 net -31.22 R avg -0.125 R (gross -0.102 R, cost 0.023 R a trade); net % of price a trade -0.644%; PF 0.68 t -2.80
  winners 57.2% (95% 51.0-63.2%); ended by target 136 (54.4%), stop 88 (35.2%), Friday close 26 (10.4%); hours held median 19.22; without the 10 best -36.16 R
  longs (gap down) n=102 net -15.13 R | shorts (gap up) n=148 net -16.09 R
  by year (n, net R): 2017 2 -0.53 | 2018 38 -4.15 | 2019 33 -0.12 | 2020 29 -2.78 | 2021 36 -3.82 | 2022 26 -9.18 | 2023 23 -3.78 | 2024 26 -5.43 | 2025 22 -0.98 | 2026 15 -0.46
  control (entries moved to other days, same time of day, same distances, 500 runs, 2 s): mean trades 249.92, median avg -0.028 R, 5-95% -0.098 .. 0.031 R; real avg -0.125 R at percentile 0.60
  [FAIL] 1 PF >= 1.2 in validation: VAL PF 0.55
  [PASS] 2 >= 50 trades in each period: 122 / 65 / 63
  [FAIL] 3 t >= 2.0 on development: DEV t -1.44
  [FAIL] 4 above 95th pct of the control: percentile 0.60
  [FAIL] 5 positive without top-10 trades: -36.16 R
  [FAIL] 6 holdout net positive (flagged used): HOLDOUT net -6.87 R PF 0.72
  VERDICT Buried (1/6 gates)
  rule 3: 7 trades touch a data gap of 10 minutes or more (btc_gaps.json); without them avg -0.121 R, gates 1/6 -> Buried (verdict unchanged)

=== CG-NS: no stop, target Friday's close, out by the next Friday 17:00 New York ===
  DEV      n= 122 net -49.31 R  avg -0.404 R  PF 0.62 t -1.47
  VAL      n=  65 net 4.04 R  avg 0.062 R  PF 1.10 t 0.32
  HOLDOUT  n=  63 net -0.54 R  avg -0.009 R  PF 0.99 t -0.03
  ALL      n=250 net -45.81 R avg -0.183 R (gross -0.137 R, cost 0.046 R a trade); net % of price a trade -0.455%; PF 0.78 t -1.15
  winners 72.4% (95% 66.6-77.6%); ended by target 173 (69.2%), Friday close 77 (30.8%); hours held median 46.57; without the 10 best -55.69 R
  longs (gap down) n=102 net 22.41 R | shorts (gap up) n=148 net -68.23 R
  by year (n, net R): 2017 2 1.95 | 2018 38 -18.91 | 2019 33 6.26 | 2020 29 -28.51 | 2021 36 -4.87 | 2022 26 6.49 | 2023 23 -7.68 | 2024 26 -8.74 | 2025 22 7.53 | 2026 15 0.67
  control (entries moved to other days, same time of day, same distances, 500 runs, 2 s): mean trades 249.92, median avg -0.174 R, 5-95% -0.450 .. 0.059 R; real avg -0.183 R at percentile 47.60
  [FAIL] 1 PF >= 1.2 in validation: VAL PF 1.10
  [PASS] 2 >= 50 trades in each period: 122 / 65 / 63
  [FAIL] 3 t >= 2.0 on development: DEV t -1.47
  [FAIL] 4 above 95th pct of the control: percentile 47.60
  [FAIL] 5 positive without top-10 trades: -55.69 R
  [FAIL] 6 holdout net positive (flagged used): HOLDOUT net -0.54 R PF 0.99
  VERDICT Buried (1/6 gates)
  rule 3: 11 trades touch a data gap of 10 minutes or more (btc_gaps.json); without them avg -0.132 R, gates 1/6 -> Buried (verdict unchanged)