SirStone
be74e369eb
problem gives 60%. Thread closed with a number, not a shrug.
Gate 2 (fd2f7f6) said STOP because every correction lost hits. But the shifts it
applied were the conditional MEDIANS of the error (4-16 deg) while the gate metric
is HITS, and the naive guess is UNBIASED - so shifting an already-centred
distribution can only destroy near-target mass. That suggested the experiment was
MISCALIBRATED rather than the idea being dead. One cheap test settled it.
ARMS (pooled, both primary fixtures, 25 rounds, N~9433/h; deltas in pp vs naive):
h naive TM-med TM-hit TM-x0.25 PERF-SIGN shuffled
15 39.3 29.0(-10.3) 35.1(-4.2) 37.9(-1.5) **47.0 (+7.7)** 35.1(-4.2)
20 28.2 19.2(-9.0) 22.8(-5.4) 25.6(-2.6) **33.4 (+5.1)** 22.0(-6.2)
25 20.9 13.1(-7.8) 15.4(-5.5) 18.8(-2.2) **25.6 (+4.7)** 16.0(-4.9)
30 16.4 10.5(-5.9) 11.4(-5.0) 14.1(-2.3) **18.3 (+1.9)** 11.2(-5.2)
1. **THE FORM CAN BUY HITS.** Perfect side knowledge + the hit-optimal shift gains
**+4.9 pp mean** (+1.9..+7.7), positive in ~19/25 rounds and in BOTH fixtures -
and it improves the median AND p90 residual. So the premise is NOT structurally
impossible.
2. **BUT THE TM'S 60% SIDE ACCURACY LOSES** (-4.2 pp quadrant, -3.3..-5.6 side).
Shrinking the shift toward zero monotonically reduces the loss but never turns
it positive.
3. **BREAK-EVEN IS ~80% SIDE ACCURACY** (synthetic sweep: 0.70 -> -1.6 pp, **0.80
-> 0.0**, 1.00 -> +4.9 pp). The observed side signal tops out at **~60% (TM) /
53-56% (turn-only rule)** against 50% chance, and the headroom study found it is
essentially ONE WEAK FEATURE. **No realistic predictor of this problem clears
the 80% wall.**
WHY IT IS SO SENSITIVE - the hits-vs-shift curve (h15, calibration): shift 0 ->
36.2%, +-2deg -> 27.4/26.8, +-4deg -> 14.2, +-6deg -> 11.7, +-10deg -> 8.3.
Shifting unconditionally is catastrophic. The whole value comes from shifting ONLY
when the side is known, because the signed-error distribution becomes ONE-SIDED
once you condition on the true side. And the hit-optimal PERF-SIGN shift is only
**+-2 to 3.5 deg** - an order of magnitude smaller than the 4-16 deg medians Gate 2
used, which is exactly why Gate 2's calibration could never win.
VERDICT: **MISCALIBRATED, but practically dead at achievable accuracy.**
Recommendation: stop the per-bucket-shift direction, and record the RIGHT reason -
an **accuracy wall at ~80%**, not an impossibility of the form. If ever revived,
the only viable path is a side predictor that materially exceeds 60%; a better
calibration cannot fix it (we already used the hit-optimal one).
METHOD NOTE: the hit-optimal shift is fitted on the calibration slice (the
pipeline's existing out-of-sample offset split) and applied on the later eval
slice - NOT fitted on the TM training slice, to avoid in-sample label leakage.
INTEGRITY: TM-med reproduces Gate 2's committed deltas to the decimal
(-10.3/-9.0/-7.8/-5.9); PS-shift0 and TM-x0.0 are exactly naive; the shuffled
control never improves. Guards all pass (test_tm_diag 48, test_tm_automata_diag 55,
test_tm_clause_shape 66, diag_synthetic, diag_automata_validation, test_gun_harness,
test_vbullet_metric, test_power_selection, test_adaptive_radar,
test_tfil_ring_weights, test_power_policy, test_ram_decision, test_rack_membership,
test_selector_tiebreak, test_tm_pattern_registration, test_vbullet_admit_gate).
Caveats: DrussGT-only; enemy movement is a closed-loop response to our CURRENT
movement; offline observation is perfect while live we see the enemy only on scans
-> all absolute hit fractions are optimistic upper bounds, only deltas are
meaningful.
1205 lines
55 KiB
Plaintext
1205 lines
55 KiB
Plaintext
==========================================================================================
|
|
GATE 2b - is the shift-per-bucket appliance STRUCTURALLY DEAD or MISCALIBRATED?
|
|
==========================================================================================
|
|
fixtures : tr_drussgt_vs_modularbot.jsonl, tr_drussgt_vs_modularbot_shield.jsonl
|
|
bits : 53 (49 draft + 4 horizon one-hot)
|
|
TM : 50 clauses, 64 states, s=3.0, 5 epochs, fresh per round
|
|
split : within-round, train 50%, calibrate 70.%, eval rest
|
|
appliance : shift on {-20..+20 deg, 0.5 step} maximising HIT COUNT
|
|
on the calibration slice; applied out-of-sample
|
|
estimated hit : |residual| < atan(18px / range) -- ABSOLUTE IS OPTIMISTIC
|
|
DECISIVE arm : PERF-SIGN uses the TRUE sign (cheating) + hit-optimal shift
|
|
bullet block : INFERRED from self-energy drops (no gun heading recorded)
|
|
==========================================================================================
|
|
|
|
## FIXTURE tr_drussgt_vs_modularbot.jsonl: 15 rounds, 20026 ticks
|
|
round 1 (L= 1252) nEval/h = [361, 356, 351, 346]
|
|
h arm med p90 hit%
|
|
15 naive 4.20 11.73 30.2
|
|
15 TM-med 4.54 12.06 29.9
|
|
15 TM-hit 4.93 12.23 24.4
|
|
15 TM-side 4.32 11.92 28.5
|
|
15 TM-x0.25 4.43 11.86 30.5
|
|
15 TM-x0.5 4.66 11.81 29.4
|
|
15 PERF-SIGN 2.60 10.23 46.3
|
|
15 shuffled 4.56 12.96 25.5
|
|
20 naive 6.41 18.49 16.0
|
|
20 TM-med 6.19 18.21 19.1
|
|
20 TM-hit 6.50 18.52 18.3
|
|
20 TM-side 6.60 18.53 17.1
|
|
20 TM-x0.25 6.41 19.15 20.2
|
|
20 TM-x0.5 6.27 18.62 21.1
|
|
20 PERF-SIGN 3.69 16.34 36.2
|
|
20 shuffled 6.50 18.86 17.7
|
|
25 naive 8.45 26.51 12.3
|
|
25 TM-med 9.14 22.74 8.3
|
|
25 TM-hit 9.40 23.40 10.0
|
|
25 TM-side 9.81 23.53 11.1
|
|
25 TM-x0.25 8.46 25.22 16.8
|
|
25 TM-x0.5 8.54 24.12 16.2
|
|
25 PERF-SIGN 4.40 22.40 31.1
|
|
25 shuffled 9.17 26.16 11.4
|
|
30 naive 8.91 33.37 13.3
|
|
30 TM-med 11.18 29.74 6.6
|
|
30 TM-hit 11.97 28.64 7.2
|
|
30 TM-side 12.27 29.66 8.1
|
|
30 TM-x0.25 9.85 33.52 13.0
|
|
30 TM-x0.5 9.93 31.89 12.4
|
|
30 PERF-SIGN 6.03 25.37 20.5
|
|
30 shuffled 9.36 37.04 13.9
|
|
|
|
round 2 (L= 1416) nEval/h = [410, 405, 400, 395]
|
|
h arm med p90 hit%
|
|
15 naive 2.49 13.34 47.8
|
|
15 TM-med 4.42 11.90 28.8
|
|
15 TM-hit 3.73 12.97 40.5
|
|
15 TM-side 2.75 13.24 45.4
|
|
15 TM-x0.25 2.83 12.91 42.7
|
|
15 TM-x0.5 3.24 12.53 39.3
|
|
15 PERF-SIGN 1.50 11.84 54.9
|
|
15 shuffled 3.39 13.33 42.7
|
|
20 naive 4.91 21.45 35.6
|
|
20 TM-med 7.21 17.83 25.2
|
|
20 TM-hit 5.21 20.49 33.1
|
|
20 TM-side 4.42 20.84 36.5
|
|
20 TM-x0.25 5.17 20.31 31.6
|
|
20 TM-x0.5 5.25 19.62 30.9
|
|
20 PERF-SIGN 5.58 16.00 33.1
|
|
20 shuffled 7.91 20.89 25.7
|
|
25 naive 7.54 29.02 28.8
|
|
25 TM-med 8.97 27.06 21.8
|
|
25 TM-hit 8.81 27.65 18.5
|
|
25 TM-side 7.24 28.57 25.2
|
|
25 TM-x0.25 7.25 27.67 20.5
|
|
25 TM-x0.5 8.04 26.35 21.0
|
|
25 PERF-SIGN 7.12 23.73 25.2
|
|
25 shuffled 8.53 29.93 26.2
|
|
30 naive 12.27 36.16 21.8
|
|
30 TM-med 10.55 31.84 13.7
|
|
30 TM-hit 14.50 41.65 13.9
|
|
30 TM-side 11.12 35.98 15.7
|
|
30 TM-x0.25 12.39 36.59 17.2
|
|
30 TM-x0.5 12.65 37.68 15.9
|
|
30 PERF-SIGN 11.02 29.88 14.9
|
|
30 shuffled 17.75 43.76 7.3
|
|
|
|
round 3 (L= 845) nEval/h = [239, 234, 229, 224]
|
|
h arm med p90 hit%
|
|
15 naive 1.09 13.57 59.4
|
|
15 TM-med 1.08 13.53 59.8
|
|
15 TM-hit 0.98 13.43 58.2
|
|
15 TM-side 1.76 13.66 56.5
|
|
15 TM-x0.25 1.03 13.67 59.8
|
|
15 TM-x0.5 0.98 13.54 59.4
|
|
15 PERF-SIGN 1.87 11.57 65.3
|
|
15 shuffled 0.96 13.79 57.7
|
|
20 naive 2.94 23.90 46.2
|
|
20 TM-med 3.27 22.89 43.2
|
|
20 TM-hit 3.44 22.99 41.9
|
|
20 TM-side 3.44 22.99 41.9
|
|
20 TM-x0.25 2.94 23.52 45.7
|
|
20 TM-x0.5 2.94 23.02 44.0
|
|
20 PERF-SIGN 1.95 21.90 54.7
|
|
20 shuffled 2.97 23.52 46.6
|
|
25 naive 7.12 32.44 29.3
|
|
25 TM-med 6.66 31.73 21.4
|
|
25 TM-hit 9.15 31.77 16.2
|
|
25 TM-side 7.12 32.44 27.5
|
|
25 TM-x0.25 6.67 32.12 25.8
|
|
25 TM-x0.5 6.40 32.24 21.8
|
|
25 PERF-SIGN 5.12 30.44 44.1
|
|
25 shuffled 6.01 32.06 29.7
|
|
30 naive 10.18 42.99 17.9
|
|
30 TM-med 12.32 40.72 20.1
|
|
30 TM-hit 12.25 41.36 19.2
|
|
30 TM-side 9.98 42.05 15.6
|
|
30 TM-x0.25 9.23 41.92 13.4
|
|
30 TM-x0.5 9.23 41.66 15.6
|
|
30 PERF-SIGN 7.98 40.55 22.3
|
|
30 shuffled 11.09 42.12 15.6
|
|
|
|
round 4 (L= 1446) nEval/h = [419, 414, 409, 404]
|
|
h arm med p90 hit%
|
|
15 naive 4.97 11.21 23.9
|
|
15 TM-med 4.77 12.28 24.3
|
|
15 TM-hit 5.56 12.71 22.0
|
|
15 TM-side 5.00 11.30 24.3
|
|
15 TM-x0.25 4.73 10.83 26.5
|
|
15 TM-x0.5 4.86 11.03 26.0
|
|
15 PERF-SIGN 3.26 8.09 33.7
|
|
15 shuffled 5.88 13.48 19.6
|
|
20 naive 7.53 16.45 13.5
|
|
20 TM-med 6.59 16.82 12.3
|
|
20 TM-hit 6.07 16.22 15.2
|
|
20 TM-side 6.82 16.19 15.9
|
|
20 TM-x0.25 6.66 15.64 16.2
|
|
20 TM-x0.5 6.15 15.56 15.5
|
|
20 PERF-SIGN 4.30 12.85 29.0
|
|
20 shuffled 7.19 16.72 14.0
|
|
25 naive 9.99 23.25 10.5
|
|
25 TM-med 9.03 22.95 11.7
|
|
25 TM-hit 10.30 24.95 8.6
|
|
25 TM-side 9.31 21.88 11.0
|
|
25 TM-x0.25 9.12 22.81 9.3
|
|
25 TM-x0.5 8.95 23.07 11.5
|
|
25 PERF-SIGN 6.11 18.11 20.5
|
|
25 shuffled 10.51 28.54 12.5
|
|
30 naive 12.60 29.60 6.9
|
|
30 TM-med 11.19 29.12 8.7
|
|
30 TM-hit 10.37 29.63 11.1
|
|
30 TM-side 11.72 29.66 9.4
|
|
30 TM-x0.25 11.42 28.81 7.2
|
|
30 TM-x0.5 10.49 29.33 8.4
|
|
30 PERF-SIGN 6.41 18.64 14.4
|
|
30 shuffled 13.68 34.90 6.2
|
|
|
|
round 5 (L= 1893) nEval/h = [553, 548, 543, 538]
|
|
h arm med p90 hit%
|
|
15 naive 3.33 13.59 44.7
|
|
15 TM-med 3.13 13.70 45.4
|
|
15 TM-hit 3.34 13.60 44.1
|
|
15 TM-side 3.55 13.70 42.1
|
|
15 TM-x0.25 3.34 13.60 44.7
|
|
15 TM-x0.5 3.35 13.60 44.8
|
|
15 PERF-SIGN 1.93 11.60 54.4
|
|
15 shuffled 3.76 14.26 42.7
|
|
20 naive 5.75 21.91 34.9
|
|
20 TM-med 5.23 21.62 31.8
|
|
20 TM-hit 5.73 22.28 33.9
|
|
20 TM-side 5.79 21.59 32.8
|
|
20 TM-x0.25 5.62 20.81 35.2
|
|
20 TM-x0.5 5.41 21.21 34.5
|
|
20 PERF-SIGN 4.25 20.41 42.5
|
|
20 shuffled 6.70 22.56 30.1
|
|
25 naive 8.09 29.86 27.6
|
|
25 TM-med 6.89 29.83 17.5
|
|
25 TM-hit 7.03 28.99 23.9
|
|
25 TM-side 7.44 29.54 22.1
|
|
25 TM-x0.25 7.72 29.57 27.1
|
|
25 TM-x0.5 7.40 29.22 27.3
|
|
25 PERF-SIGN 6.24 28.02 37.0
|
|
25 shuffled 7.99 30.33 25.0
|
|
30 naive 11.17 35.88 21.2
|
|
30 TM-med 9.42 36.62 15.4
|
|
30 TM-hit 10.31 36.05 20.4
|
|
30 TM-side 11.25 36.37 19.0
|
|
30 TM-x0.25 10.76 35.58 19.7
|
|
30 TM-x0.5 10.42 35.30 20.4
|
|
30 PERF-SIGN 8.94 33.45 30.1
|
|
30 shuffled 11.82 37.70 18.2
|
|
|
|
round 6 (L= 1080) nEval/h = [309, 304, 299, 294]
|
|
h arm med p90 hit%
|
|
15 naive 6.66 13.01 20.7
|
|
15 TM-med 6.26 12.61 17.5
|
|
15 TM-hit 6.46 12.93 21.0
|
|
15 TM-side 6.45 13.23 20.7
|
|
15 TM-x0.25 6.71 12.93 21.0
|
|
15 TM-x0.5 6.68 12.93 21.0
|
|
15 PERF-SIGN 5.16 11.51 27.5
|
|
15 shuffled 6.61 12.81 19.7
|
|
20 naive 9.00 20.62 12.2
|
|
20 TM-med 8.75 18.74 13.2
|
|
20 TM-hit 9.11 20.21 11.8
|
|
20 TM-side 9.16 20.23 13.2
|
|
20 TM-x0.25 8.59 19.92 11.2
|
|
20 TM-x0.5 8.30 18.93 11.2
|
|
20 PERF-SIGN 7.50 19.12 18.8
|
|
20 shuffled 9.09 20.62 12.2
|
|
25 naive 11.36 26.48 8.0
|
|
25 TM-med 9.05 23.09 11.4
|
|
25 TM-hit 10.24 26.44 7.4
|
|
25 TM-side 11.69 27.48 7.4
|
|
25 TM-x0.25 10.91 26.51 9.7
|
|
25 TM-x0.5 10.89 25.81 10.0
|
|
25 PERF-SIGN 10.09 25.32 13.4
|
|
25 shuffled 11.05 27.32 9.0
|
|
30 naive 14.01 32.46 6.5
|
|
30 TM-med 10.03 29.06 13.6
|
|
30 TM-hit 13.32 36.28 5.4
|
|
30 TM-side 13.67 34.07 5.4
|
|
30 TM-x0.25 13.50 32.39 7.1
|
|
30 TM-x0.5 13.07 33.78 6.5
|
|
30 PERF-SIGN 12.14 30.74 11.2
|
|
30 shuffled 12.97 32.16 5.8
|
|
|
|
round 7 (L= 1460) nEval/h = [424, 419, 414, 409]
|
|
h arm med p90 hit%
|
|
15 naive 4.85 14.20 33.5
|
|
15 TM-med 5.08 13.66 27.1
|
|
15 TM-hit 5.68 14.20 29.0
|
|
15 TM-side 4.78 14.20 33.7
|
|
15 TM-x0.25 4.87 14.06 30.4
|
|
15 TM-x0.5 4.98 14.31 30.4
|
|
15 PERF-SIGN 3.35 12.70 42.7
|
|
15 shuffled 6.25 16.14 24.8
|
|
20 naive 7.32 20.48 21.7
|
|
20 TM-med 8.11 19.04 11.5
|
|
20 TM-hit 7.97 20.48 19.3
|
|
20 TM-side 7.36 20.71 22.0
|
|
20 TM-x0.25 7.34 20.58 20.8
|
|
20 TM-x0.5 7.37 20.48 20.5
|
|
20 PERF-SIGN 5.82 18.98 32.5
|
|
20 shuffled 8.64 22.33 17.7
|
|
25 naive 10.81 28.52 16.4
|
|
25 TM-med 10.16 26.15 8.7
|
|
25 TM-hit 11.15 28.67 17.1
|
|
25 TM-side 10.71 28.54 16.7
|
|
25 TM-x0.25 11.02 28.54 17.1
|
|
25 TM-x0.5 10.91 28.54 17.4
|
|
25 PERF-SIGN 9.31 27.02 23.9
|
|
25 shuffled 10.86 30.63 13.3
|
|
30 naive 14.36 34.95 9.5
|
|
30 TM-med 13.26 32.06 5.9
|
|
30 TM-hit 14.61 33.93 8.3
|
|
30 TM-side 14.95 34.44 7.8
|
|
30 TM-x0.25 14.37 34.70 10.0
|
|
30 TM-x0.5 13.59 33.80 7.8
|
|
30 PERF-SIGN 12.86 33.45 18.1
|
|
30 shuffled 15.53 37.65 9.8
|
|
|
|
round 8 (L= 1213) nEval/h = [349, 344, 339, 334]
|
|
h arm med p90 hit%
|
|
15 naive 3.89 15.17 37.2
|
|
15 TM-med 4.06 13.55 38.7
|
|
15 TM-hit 4.33 14.87 37.0
|
|
15 TM-side 4.88 15.52 35.5
|
|
15 TM-x0.25 3.59 14.45 34.1
|
|
15 TM-x0.5 4.02 13.76 37.5
|
|
15 PERF-SIGN 2.39 13.67 47.9
|
|
15 shuffled 4.27 14.97 37.0
|
|
20 naive 7.39 21.18 30.5
|
|
20 TM-med 6.32 21.44 32.0
|
|
20 TM-hit 6.07 23.20 31.7
|
|
20 TM-side 7.04 25.02 29.9
|
|
20 TM-x0.25 6.11 20.33 28.2
|
|
20 TM-x0.5 6.39 21.14 29.9
|
|
20 PERF-SIGN 5.89 19.68 38.1
|
|
20 shuffled 7.76 21.29 33.1
|
|
25 naive 10.75 26.76 28.3
|
|
25 TM-med 7.12 29.72 23.9
|
|
25 TM-hit 6.77 30.96 24.2
|
|
25 TM-side 7.99 31.04 24.5
|
|
25 TM-x0.25 9.12 26.47 23.6
|
|
25 TM-x0.5 8.20 27.55 23.3
|
|
25 PERF-SIGN 9.70 25.26 32.2
|
|
25 shuffled 10.70 26.45 30.1
|
|
30 naive 13.98 31.31 24.9
|
|
30 TM-med 6.90 32.68 20.7
|
|
30 TM-hit 8.03 33.07 22.2
|
|
30 TM-side 9.22 33.06 20.4
|
|
30 TM-x0.25 12.70 31.94 22.8
|
|
30 TM-x0.5 11.07 32.56 20.7
|
|
30 PERF-SIGN 12.48 29.81 28.1
|
|
30 shuffled 13.01 32.78 18.6
|
|
|
|
round 9 (L= 1759) nEval/h = [513, 508, 503, 498]
|
|
h arm med p90 hit%
|
|
15 naive 2.58 14.40 49.3
|
|
15 TM-med 4.59 13.54 19.9
|
|
15 TM-hit 2.58 14.23 49.3
|
|
15 TM-side 2.70 14.50 48.1
|
|
15 TM-x0.25 2.58 14.38 49.5
|
|
15 TM-x0.5 2.57 14.03 49.7
|
|
15 PERF-SIGN 1.78 12.40 60.0
|
|
15 shuffled 2.82 16.40 47.4
|
|
20 naive 5.52 22.86 37.0
|
|
20 TM-med 7.28 20.73 11.8
|
|
20 TM-hit 9.98 20.53 8.9
|
|
20 TM-side 5.42 22.41 33.9
|
|
20 TM-x0.25 5.26 21.56 25.8
|
|
20 TM-x0.5 6.63 20.64 11.2
|
|
20 PERF-SIGN 5.42 18.99 33.3
|
|
20 shuffled 6.92 23.11 26.4
|
|
25 naive 8.13 31.70 33.6
|
|
25 TM-med 10.37 29.16 8.5
|
|
25 TM-hit 9.59 29.43 7.4
|
|
25 TM-side 8.85 32.87 7.2
|
|
25 TM-x0.25 8.13 30.60 25.2
|
|
25 TM-x0.5 8.36 30.00 12.7
|
|
25 PERF-SIGN 5.21 27.31 24.9
|
|
25 shuffled 9.92 30.06 17.5
|
|
30 naive 9.83 39.66 28.1
|
|
30 TM-med 13.29 36.45 8.4
|
|
30 TM-hit 11.41 37.37 6.0
|
|
30 TM-side 9.77 40.57 15.1
|
|
30 TM-x0.25 10.29 38.90 20.3
|
|
30 TM-x0.5 10.43 38.33 9.6
|
|
30 PERF-SIGN 10.56 30.54 11.6
|
|
30 shuffled 13.28 41.47 15.9
|
|
|
|
round 10 (L= 1380) nEval/h = [400, 395, 390, 385]
|
|
h arm med p90 hit%
|
|
15 naive 4.01 13.67 39.5
|
|
15 TM-med 4.47 13.06 27.0
|
|
15 TM-hit 3.90 13.40 32.0
|
|
15 TM-side 3.82 13.34 30.5
|
|
15 TM-x0.25 4.14 13.54 40.2
|
|
15 TM-x0.5 4.16 13.42 34.2
|
|
15 PERF-SIGN 3.17 11.23 36.5
|
|
15 shuffled 4.31 15.73 37.2
|
|
20 naive 7.63 20.99 22.8
|
|
20 TM-med 7.74 21.04 16.7
|
|
20 TM-hit 7.55 21.65 19.7
|
|
20 TM-side 7.52 20.37 16.5
|
|
20 TM-x0.25 7.29 20.99 25.1
|
|
20 TM-x0.5 7.18 20.97 24.6
|
|
20 PERF-SIGN 5.27 17.22 20.5
|
|
20 shuffled 7.41 21.15 22.3
|
|
25 naive 10.69 28.86 15.4
|
|
25 TM-med 9.74 28.43 8.2
|
|
25 TM-hit 8.99 29.45 13.8
|
|
25 TM-side 8.36 28.36 16.4
|
|
25 TM-x0.25 10.57 28.96 19.2
|
|
25 TM-x0.5 9.43 28.93 18.5
|
|
25 PERF-SIGN 6.08 24.46 14.6
|
|
25 shuffled 10.36 29.31 11.5
|
|
30 naive 13.79 35.78 11.2
|
|
30 TM-med 13.76 34.64 7.0
|
|
30 TM-hit 12.52 36.57 11.9
|
|
30 TM-side 12.63 34.28 11.2
|
|
30 TM-x0.25 13.47 35.90 12.7
|
|
30 TM-x0.5 12.91 36.28 13.0
|
|
30 PERF-SIGN 9.67 31.90 19.0
|
|
30 shuffled 13.23 40.25 9.6
|
|
|
|
round 11 (L= 1127) nEval/h = [324, 319, 314, 309]
|
|
h arm med p90 hit%
|
|
15 naive 6.80 14.92 20.7
|
|
15 TM-med 6.86 15.11 19.1
|
|
15 TM-hit 8.24 17.47 20.4
|
|
15 TM-side 6.70 14.85 21.0
|
|
15 TM-x0.25 6.67 14.12 16.0
|
|
15 TM-x0.5 6.99 14.85 17.9
|
|
15 PERF-SIGN 3.97 11.82 27.2
|
|
15 shuffled 7.93 15.17 13.9
|
|
20 naive 9.68 22.92 15.4
|
|
20 TM-med 9.87 22.80 11.9
|
|
20 TM-hit 10.39 23.44 12.5
|
|
20 TM-side 10.58 27.27 10.7
|
|
20 TM-x0.25 9.26 22.24 12.5
|
|
20 TM-x0.5 9.61 22.49 15.7
|
|
20 PERF-SIGN 7.98 16.55 18.2
|
|
20 shuffled 10.36 24.28 9.7
|
|
25 naive 10.88 30.80 13.7
|
|
25 TM-med 11.11 28.15 10.2
|
|
25 TM-hit 10.96 28.03 11.1
|
|
25 TM-side 12.05 33.22 10.2
|
|
25 TM-x0.25 10.67 29.83 11.1
|
|
25 TM-x0.5 10.49 29.19 11.1
|
|
25 PERF-SIGN 9.86 22.35 14.0
|
|
25 shuffled 12.39 32.57 11.5
|
|
30 naive 13.77 38.10 10.0
|
|
30 TM-med 12.41 33.61 9.7
|
|
30 TM-hit 13.28 35.20 7.8
|
|
30 TM-side 14.86 40.87 8.4
|
|
30 TM-x0.25 12.42 36.61 6.5
|
|
30 TM-x0.5 12.67 35.72 5.5
|
|
30 PERF-SIGN 11.54 27.84 11.0
|
|
30 shuffled 13.87 38.88 5.8
|
|
|
|
round 12 (L= 1258) nEval/h = [363, 358, 353, 348]
|
|
h arm med p90 hit%
|
|
15 naive 2.76 13.10 47.4
|
|
15 TM-med 2.72 12.78 46.8
|
|
15 TM-hit 2.88 13.16 45.7
|
|
15 TM-side 2.83 13.00 45.7
|
|
15 TM-x0.25 2.78 13.12 47.4
|
|
15 TM-x0.5 2.82 13.07 46.8
|
|
15 PERF-SIGN 1.96 11.10 56.7
|
|
15 shuffled 3.28 13.54 41.0
|
|
20 naive 5.47 19.38 39.4
|
|
20 TM-med 4.74 19.08 39.9
|
|
20 TM-hit 4.13 18.92 36.9
|
|
20 TM-side 5.48 19.50 39.4
|
|
20 TM-x0.25 5.07 19.36 39.9
|
|
20 TM-x0.5 4.62 18.92 38.8
|
|
20 PERF-SIGN 2.99 17.36 44.7
|
|
20 shuffled 5.80 19.50 33.2
|
|
25 naive 8.00 26.76 32.0
|
|
25 TM-med 7.08 26.69 31.2
|
|
25 TM-hit 6.95 27.50 24.9
|
|
25 TM-side 7.52 27.50 29.7
|
|
25 TM-x0.25 7.47 27.31 33.7
|
|
25 TM-x0.5 6.93 27.24 33.4
|
|
25 PERF-SIGN 5.40 24.13 37.4
|
|
25 shuffled 8.52 27.50 29.5
|
|
30 naive 10.11 33.73 24.1
|
|
30 TM-med 10.04 32.63 14.4
|
|
30 TM-hit 8.71 33.81 26.7
|
|
30 TM-side 13.08 36.25 15.2
|
|
30 TM-x0.25 9.48 33.57 26.1
|
|
30 TM-x0.5 9.07 33.26 27.0
|
|
30 PERF-SIGN 7.32 30.69 27.9
|
|
30 shuffled 11.52 33.81 14.9
|
|
|
|
round 13 (L= 1330) nEval/h = [385, 380, 375, 370]
|
|
h arm med p90 hit%
|
|
15 naive 5.81 16.12 32.7
|
|
15 TM-med 8.07 16.09 16.1
|
|
15 TM-hit 6.07 15.20 20.0
|
|
15 TM-side 5.60 15.63 33.0
|
|
15 TM-x0.25 5.71 15.60 32.5
|
|
15 TM-x0.5 5.78 15.45 27.0
|
|
15 PERF-SIGN 3.81 12.02 39.5
|
|
15 shuffled 5.75 15.63 27.3
|
|
20 naive 9.24 23.32 21.8
|
|
20 TM-med 12.84 24.67 13.7
|
|
20 TM-hit 14.16 25.89 13.4
|
|
20 TM-side 10.30 25.00 13.7
|
|
20 TM-x0.25 9.98 22.98 12.4
|
|
20 TM-x0.5 11.54 23.51 13.7
|
|
20 PERF-SIGN 6.89 21.12 32.4
|
|
20 shuffled 10.76 24.56 18.9
|
|
25 naive 12.70 31.07 13.1
|
|
25 TM-med 16.49 33.55 9.9
|
|
25 TM-hit 17.74 33.22 7.7
|
|
25 TM-side 13.73 32.50 13.6
|
|
25 TM-x0.25 13.06 30.05 8.5
|
|
25 TM-x0.5 14.35 30.52 8.3
|
|
25 PERF-SIGN 10.21 28.53 19.5
|
|
25 shuffled 13.18 31.22 14.9
|
|
30 naive 14.84 36.94 10.0
|
|
30 TM-med 20.28 42.39 8.6
|
|
30 TM-hit 19.28 40.64 7.0
|
|
30 TM-side 16.56 40.65 12.7
|
|
30 TM-x0.25 15.59 37.84 7.0
|
|
30 TM-x0.5 16.53 37.51 7.8
|
|
30 PERF-SIGN 11.84 34.44 18.4
|
|
30 shuffled 15.03 36.94 10.0
|
|
|
|
round 14 (L= 1474) nEval/h = [428, 423, 418, 413]
|
|
h arm med p90 hit%
|
|
15 naive 3.33 13.96 42.5
|
|
15 TM-med 5.17 11.86 23.4
|
|
15 TM-hit 3.18 12.11 43.0
|
|
15 TM-side 3.26 12.58 42.8
|
|
15 TM-x0.25 3.31 12.89 42.5
|
|
15 TM-x0.5 3.33 12.42 42.5
|
|
15 PERF-SIGN 2.25 11.79 50.7
|
|
15 shuffled 3.29 13.96 42.1
|
|
20 naive 6.05 21.34 29.1
|
|
20 TM-med 8.08 19.45 15.6
|
|
20 TM-hit 4.96 19.95 32.6
|
|
20 TM-side 5.44 20.96 31.7
|
|
20 TM-x0.25 6.07 19.75 30.0
|
|
20 TM-x0.5 5.68 19.70 30.0
|
|
20 PERF-SIGN 4.04 18.96 36.6
|
|
20 shuffled 7.04 21.43 23.9
|
|
25 naive 9.48 28.59 24.2
|
|
25 TM-med 11.16 26.70 11.7
|
|
25 TM-hit 7.73 27.76 23.0
|
|
25 TM-side 7.97 30.31 23.2
|
|
25 TM-x0.25 9.29 27.37 23.9
|
|
25 TM-x0.5 9.28 27.00 24.6
|
|
25 PERF-SIGN 7.56 21.35 22.2
|
|
25 shuffled 13.05 27.93 11.5
|
|
30 naive 12.80 37.10 20.8
|
|
30 TM-med 15.48 35.95 10.2
|
|
30 TM-hit 12.59 35.19 8.0
|
|
30 TM-side 11.55 38.10 21.8
|
|
30 TM-x0.25 11.85 36.02 16.9
|
|
30 TM-x0.5 11.59 34.42 9.7
|
|
30 PERF-SIGN 9.85 28.54 14.5
|
|
30 shuffled 13.71 36.23 14.5
|
|
|
|
round 15 (L= 1093) nEval/h = [313, 308, 303, 298]
|
|
h arm med p90 hit%
|
|
15 naive 4.30 18.06 40.9
|
|
15 TM-med 7.14 14.79 14.1
|
|
15 TM-hit 4.81 17.25 37.4
|
|
15 TM-side 4.53 16.87 36.4
|
|
15 TM-x0.25 4.28 17.53 41.5
|
|
15 TM-x0.5 4.20 17.61 38.0
|
|
15 PERF-SIGN 2.30 16.06 51.4
|
|
15 shuffled 4.58 17.48 39.9
|
|
20 naive 7.49 26.79 34.1
|
|
20 TM-med 9.24 21.96 11.4
|
|
20 TM-hit 7.82 25.26 30.2
|
|
20 TM-side 9.08 25.94 29.2
|
|
20 TM-x0.25 7.87 26.03 36.0
|
|
20 TM-x0.5 7.47 25.71 31.8
|
|
20 PERF-SIGN 5.55 20.63 15.6
|
|
20 shuffled 9.11 28.75 7.5
|
|
25 naive 10.87 35.16 23.1
|
|
25 TM-med 13.81 29.13 7.9
|
|
25 TM-hit 10.99 31.50 25.4
|
|
25 TM-side 11.68 29.23 6.3
|
|
25 TM-x0.25 11.01 34.22 22.4
|
|
25 TM-x0.5 10.97 33.25 22.4
|
|
25 PERF-SIGN 7.48 29.73 32.0
|
|
25 shuffled 13.51 40.29 15.5
|
|
30 naive 15.47 43.12 18.5
|
|
30 TM-med 20.37 36.64 5.0
|
|
30 TM-hit 15.73 39.72 15.4
|
|
30 TM-side 15.47 39.30 16.4
|
|
30 TM-x0.25 15.76 41.51 17.1
|
|
30 TM-x0.5 15.74 41.15 17.4
|
|
30 PERF-SIGN 11.61 32.14 9.7
|
|
30 shuffled 16.96 42.39 9.1
|
|
|
|
|
|
## PER-FIXTURE pooled (tr_drussgt_vs_modularbot.jsonl)
|
|
h arm N med|err| p90|err| hit% miss%
|
|
15 naive 5790 4.26 14.10 38.3 61.7
|
|
15 TM-med 5790 4.85 13.39 28.9 71.1
|
|
15 TM-hit 5790 4.46 14.20 35.2 64.8
|
|
15 TM-side 5790 4.26 13.78 36.6 63.4
|
|
15 TM-x0.0 5790 4.26 14.10 38.3 61.7
|
|
15 TM-x0.25 5790 4.18 13.78 37.6 62.4
|
|
15 TM-x0.5 5790 4.28 13.63 36.5 63.5
|
|
15 PERF-SIGN 5790 2.61 11.88 46.6 53.4
|
|
15 PS-shift0 5790 4.26 14.10 38.3 61.7
|
|
15 shuffled 5790 4.61 14.63 34.8 65.2
|
|
20 naive 5715 6.97 21.49 27.4 72.6
|
|
20 TM-med 5715 7.61 20.46 20.2 79.8
|
|
20 TM-hit 5715 7.61 21.44 23.6 76.4
|
|
20 TM-side 5715 7.06 21.73 25.8 74.2
|
|
20 TM-x0.0 5715 6.97 21.49 27.4 72.6
|
|
20 TM-x0.25 5715 6.72 20.93 25.9 74.1
|
|
20 TM-x0.5 5715 6.84 20.68 24.5 75.5
|
|
20 PERF-SIGN 5715 5.30 18.38 32.6 67.4
|
|
20 PS-shift0 5715 6.97 21.49 27.4 72.6
|
|
20 shuffled 5715 7.76 22.06 22.6 77.4
|
|
25 naive 5640 9.74 29.02 21.5 78.5
|
|
25 TM-med 5640 9.89 27.86 13.9 86.1
|
|
25 TM-hit 5640 9.76 28.99 16.0 84.0
|
|
25 TM-side 5640 9.72 29.02 16.8 83.2
|
|
25 TM-x0.0 5640 9.74 29.02 21.5 78.5
|
|
25 TM-x0.25 5640 9.44 28.60 19.9 80.1
|
|
25 TM-x0.5 5640 9.33 28.43 18.8 81.2
|
|
25 PERF-SIGN 5640 7.12 25.48 26.0 74.0
|
|
25 PS-shift0 5640 9.74 29.02 21.5 78.5
|
|
25 shuffled 5640 10.68 29.83 17.9 82.1
|
|
30 naive 5565 12.40 36.16 16.7 83.3
|
|
30 TM-med 5565 12.75 34.51 11.0 89.0
|
|
30 TM-hit 5565 12.70 36.23 12.6 87.4
|
|
30 TM-side 5565 12.61 36.37 13.7 86.3
|
|
30 TM-x0.0 5565 12.40 36.16 16.7 83.3
|
|
30 TM-x0.25 5565 12.03 35.77 14.8 85.2
|
|
30 TM-x0.5 5565 11.94 35.62 13.2 86.8
|
|
30 PERF-SIGN 5565 9.77 30.85 18.3 81.7
|
|
30 PS-shift0 5565 12.40 36.16 16.7 83.3
|
|
30 shuffled 5565 13.63 37.96 11.9 88.1
|
|
|
|
## PER-FIXTURE learnability (tr_drussgt_vs_modularbot.jsonl)
|
|
h N TMquad% shufquad% TMside% majQuad% pred[Ls Lb Rs Rb] / true[Ls Lb Rs Rb]
|
|
15 5790 35.0 22.2 58.2 25.4 [1237 1608 898 2047 ] / [1448 1427 1471 1444 ]
|
|
20 5715 35.3 22.3 58.9 25.3 [1273 1542 849 2051 ] / [1392 1438 1445 1440 ]
|
|
25 5640 34.6 24.0 59.0 25.6 [1227 1558 802 2053 ] / [1368 1414 1445 1413 ]
|
|
30 5565 35.4 23.7 61.1 26.1 [1254 1469 834 2008 ] / [1340 1361 1452 1412 ]
|
|
|
|
## FIXTURE tr_drussgt_vs_modularbot_shield.jsonl: 10 rounds, 12629 ticks
|
|
round 1 (L= 593) nEval/h = [163, 158, 153, 148]
|
|
h arm med p90 hit%
|
|
15 naive 2.08 16.95 61.3
|
|
15 TM-med 2.08 16.95 61.3
|
|
15 TM-hit 2.08 16.95 61.3
|
|
15 TM-side 2.08 16.95 61.3
|
|
15 TM-x0.25 2.08 16.95 61.3
|
|
15 TM-x0.5 2.08 16.95 61.3
|
|
15 PERF-SIGN 2.08 16.95 61.3
|
|
15 shuffled 2.08 16.95 61.3
|
|
20 naive 4.14 25.55 39.9
|
|
20 TM-med 4.14 25.55 39.9
|
|
20 TM-hit 4.14 25.55 39.9
|
|
20 TM-side 4.14 25.55 39.9
|
|
20 TM-x0.25 4.14 25.55 39.9
|
|
20 TM-x0.5 4.14 25.55 39.9
|
|
20 PERF-SIGN 4.14 25.55 39.9
|
|
20 shuffled 4.14 25.55 39.9
|
|
25 naive 6.92 34.33 22.9
|
|
25 TM-med 6.92 34.33 22.9
|
|
25 TM-hit 7.00 33.98 20.3
|
|
25 TM-side 7.00 33.98 20.3
|
|
25 TM-x0.25 6.91 34.36 22.9
|
|
25 TM-x0.5 7.03 34.23 20.9
|
|
25 PERF-SIGN 6.78 33.98 22.9
|
|
25 shuffled 7.00 33.98 20.3
|
|
30 naive 10.18 41.79 18.2
|
|
30 TM-med 10.18 41.79 18.2
|
|
30 TM-hit 10.18 41.79 18.2
|
|
30 TM-side 10.18 41.79 18.2
|
|
30 TM-x0.25 10.18 41.79 18.2
|
|
30 TM-x0.5 10.18 41.79 18.2
|
|
30 PERF-SIGN 10.08 40.74 19.6
|
|
30 shuffled 10.18 41.79 18.2
|
|
|
|
round 2 (L= 1081) nEval/h = [310, 305, 300, 295]
|
|
h arm med p90 hit%
|
|
15 naive 2.43 9.95 45.5
|
|
15 TM-med 3.47 8.70 26.8
|
|
15 TM-hit 2.78 9.58 37.4
|
|
15 TM-side 2.29 9.62 47.4
|
|
15 TM-x0.25 2.43 9.78 46.5
|
|
15 TM-x0.5 2.58 9.76 43.9
|
|
15 PERF-SIGN 1.49 8.45 55.5
|
|
15 shuffled 3.78 14.45 39.7
|
|
20 naive 4.96 15.77 32.5
|
|
20 TM-med 4.85 12.53 21.3
|
|
20 TM-hit 4.74 15.32 30.2
|
|
20 TM-side 4.65 15.77 34.4
|
|
20 TM-x0.25 4.85 15.65 33.8
|
|
20 TM-x0.5 4.73 15.83 33.4
|
|
20 PERF-SIGN 3.22 14.27 35.7
|
|
20 shuffled 5.03 17.36 27.9
|
|
25 naive 8.08 23.11 26.3
|
|
25 TM-med 7.10 18.45 15.7
|
|
25 TM-hit 6.56 18.83 16.7
|
|
25 TM-side 7.22 21.36 21.3
|
|
25 TM-x0.25 6.97 21.62 21.0
|
|
25 TM-x0.5 6.65 20.40 15.7
|
|
25 PERF-SIGN 6.93 17.29 24.0
|
|
25 shuffled 10.10 25.17 14.0
|
|
30 naive 11.82 30.98 18.6
|
|
30 TM-med 8.81 24.99 9.8
|
|
30 TM-hit 11.94 32.49 9.2
|
|
30 TM-side 10.21 25.96 16.9
|
|
30 TM-x0.25 10.47 30.05 10.2
|
|
30 TM-x0.5 9.89 29.49 8.1
|
|
30 PERF-SIGN 9.68 28.49 19.3
|
|
30 shuffled 12.16 30.47 15.6
|
|
|
|
round 3 (L= 1237) nEval/h = [357, 352, 347, 342]
|
|
h arm med p90 hit%
|
|
15 naive 4.45 17.62 41.5
|
|
15 TM-med 7.33 14.79 17.1
|
|
15 TM-hit 8.23 14.27 17.1
|
|
15 TM-side 6.05 16.78 24.9
|
|
15 TM-x0.25 3.62 16.38 32.2
|
|
15 TM-x0.5 4.92 15.47 17.6
|
|
15 PERF-SIGN 3.55 12.63 43.7
|
|
15 shuffled 7.07 21.51 24.6
|
|
20 naive 7.78 25.57 32.1
|
|
20 TM-med 11.92 22.17 9.7
|
|
20 TM-hit 12.52 24.04 9.9
|
|
20 TM-side 9.69 25.29 9.9
|
|
20 TM-x0.25 6.75 23.57 11.9
|
|
20 TM-x0.5 8.36 22.02 9.9
|
|
20 PERF-SIGN 5.80 20.42 37.8
|
|
20 shuffled 11.00 27.24 17.0
|
|
25 naive 11.50 31.52 19.9
|
|
25 TM-med 16.11 30.15 6.9
|
|
25 TM-hit 15.99 27.52 9.8
|
|
25 TM-side 13.31 32.73 7.2
|
|
25 TM-x0.25 10.33 29.39 11.8
|
|
25 TM-x0.5 10.29 28.48 7.2
|
|
25 PERF-SIGN 8.18 26.52 31.7
|
|
25 shuffled 14.94 36.70 4.6
|
|
30 naive 14.83 38.77 14.3
|
|
30 TM-med 20.87 34.41 5.8
|
|
30 TM-hit 16.61 35.13 5.8
|
|
30 TM-side 15.39 42.28 6.4
|
|
30 TM-x0.25 13.01 35.75 15.2
|
|
30 TM-x0.5 12.75 35.15 6.1
|
|
30 PERF-SIGN 8.95 26.30 7.3
|
|
30 shuffled 19.25 46.58 3.8
|
|
|
|
round 4 (L= 1296) nEval/h = [374, 369, 364, 359]
|
|
h arm med p90 hit%
|
|
15 naive 1.72 12.89 54.8
|
|
15 TM-med 3.62 14.68 27.0
|
|
15 TM-hit 1.97 13.89 54.5
|
|
15 TM-side 1.91 13.33 53.7
|
|
15 TM-x0.25 1.69 13.14 55.1
|
|
15 TM-x0.5 1.65 13.50 54.0
|
|
15 PERF-SIGN 1.84 11.33 61.8
|
|
15 shuffled 2.34 13.13 49.2
|
|
20 naive 3.40 21.79 42.8
|
|
20 TM-med 7.10 24.36 16.5
|
|
20 TM-hit 3.90 22.29 38.8
|
|
20 TM-side 3.73 22.02 39.6
|
|
20 TM-x0.25 3.52 21.92 43.9
|
|
20 TM-x0.5 3.47 22.22 44.7
|
|
20 PERF-SIGN 2.67 19.22 45.8
|
|
20 shuffled 4.98 21.72 22.8
|
|
25 naive 5.26 30.32 32.1
|
|
25 TM-med 9.22 34.28 13.5
|
|
25 TM-hit 5.86 31.25 24.2
|
|
25 TM-side 6.84 30.71 9.1
|
|
25 TM-x0.25 5.42 30.56 32.1
|
|
25 TM-x0.5 5.43 30.55 31.6
|
|
25 PERF-SIGN 2.91 27.68 36.8
|
|
25 shuffled 7.26 29.97 17.6
|
|
30 naive 6.85 38.04 23.7
|
|
30 TM-med 10.38 42.47 15.3
|
|
30 TM-hit 8.11 39.17 11.1
|
|
30 TM-side 9.46 38.46 7.5
|
|
30 TM-x0.25 6.73 38.70 24.2
|
|
30 TM-x0.5 7.53 38.37 20.3
|
|
30 PERF-SIGN 5.01 32.04 22.3
|
|
30 shuffled 9.72 37.96 9.5
|
|
|
|
round 5 (L= 1550) nEval/h = [450, 445, 440, 435]
|
|
h arm med p90 hit%
|
|
15 naive 4.76 14.38 37.1
|
|
15 TM-med 4.14 13.43 37.3
|
|
15 TM-hit 4.17 13.46 36.0
|
|
15 TM-side 4.32 14.18 37.6
|
|
15 TM-x0.25 4.39 13.75 34.4
|
|
15 TM-x0.5 4.50 13.31 34.7
|
|
15 PERF-SIGN 2.76 12.38 46.2
|
|
15 shuffled 4.31 14.47 38.7
|
|
20 naive 7.25 22.56 22.5
|
|
20 TM-med 7.31 19.53 21.8
|
|
20 TM-hit 6.87 19.92 18.0
|
|
20 TM-side 7.50 22.31 21.8
|
|
20 TM-x0.25 7.17 21.56 22.5
|
|
20 TM-x0.5 6.91 20.92 20.0
|
|
20 PERF-SIGN 5.25 20.56 37.5
|
|
20 shuffled 7.50 21.93 21.3
|
|
25 naive 9.70 30.01 11.4
|
|
25 TM-med 9.72 26.88 12.3
|
|
25 TM-hit 9.33 26.70 12.0
|
|
25 TM-side 9.03 29.72 15.5
|
|
25 TM-x0.25 9.42 28.77 10.9
|
|
25 TM-x0.5 9.06 27.75 11.6
|
|
25 PERF-SIGN 6.31 26.30 21.6
|
|
25 shuffled 8.87 29.42 18.9
|
|
30 naive 12.35 36.18 8.0
|
|
30 TM-med 13.08 34.35 8.3
|
|
30 TM-hit 13.31 34.33 9.7
|
|
30 TM-side 12.96 36.80 10.6
|
|
30 TM-x0.25 11.54 35.43 9.2
|
|
30 TM-x0.5 12.07 34.68 8.7
|
|
30 PERF-SIGN 7.33 31.78 17.7
|
|
30 shuffled 13.00 36.19 9.0
|
|
|
|
round 6 (L= 1284) nEval/h = [371, 366, 361, 356]
|
|
h arm med p90 hit%
|
|
15 naive 4.35 13.01 29.9
|
|
15 TM-med 5.08 12.79 14.3
|
|
15 TM-hit 4.38 12.73 30.5
|
|
15 TM-side 4.44 12.78 29.4
|
|
15 TM-x0.25 4.19 13.05 30.7
|
|
15 TM-x0.5 4.14 12.93 31.5
|
|
15 PERF-SIGN 2.57 11.38 41.5
|
|
15 shuffled 4.44 13.43 30.2
|
|
20 naive 7.35 18.96 18.9
|
|
20 TM-med 9.10 20.33 11.7
|
|
20 TM-hit 7.27 18.22 16.4
|
|
20 TM-side 7.77 18.48 15.8
|
|
20 TM-x0.25 7.33 19.22 18.9
|
|
20 TM-x0.5 7.26 18.58 17.8
|
|
20 PERF-SIGN 5.29 16.93 27.0
|
|
20 shuffled 7.61 20.85 15.8
|
|
25 naive 11.26 25.97 10.5
|
|
25 TM-med 13.50 30.78 9.4
|
|
25 TM-hit 10.81 25.06 10.2
|
|
25 TM-side 11.13 25.48 9.1
|
|
25 TM-x0.25 11.13 25.29 11.1
|
|
25 TM-x0.5 10.64 25.14 10.8
|
|
25 PERF-SIGN 7.49 17.37 13.9
|
|
25 shuffled 11.54 28.86 7.8
|
|
30 naive 15.01 33.20 7.9
|
|
30 TM-med 15.25 36.96 7.9
|
|
30 TM-hit 14.61 34.12 5.3
|
|
30 TM-side 14.53 33.14 5.3
|
|
30 TM-x0.25 15.10 33.00 7.0
|
|
30 TM-x0.5 14.50 33.25 5.9
|
|
30 PERF-SIGN 10.37 24.48 9.8
|
|
30 shuffled 17.36 38.99 4.2
|
|
|
|
round 7 (L= 1364) nEval/h = [395, 390, 385, 380]
|
|
h arm med p90 hit%
|
|
15 naive 2.46 13.59 49.1
|
|
15 TM-med 5.33 12.88 36.2
|
|
15 TM-hit 5.90 14.89 33.9
|
|
15 TM-side 2.44 13.55 49.4
|
|
15 TM-x0.25 3.66 13.50 38.5
|
|
15 TM-x0.5 4.70 12.88 37.0
|
|
15 PERF-SIGN 1.93 11.59 59.0
|
|
15 shuffled 2.68 13.92 47.8
|
|
20 naive 5.01 21.20 34.4
|
|
20 TM-med 8.50 20.85 20.5
|
|
20 TM-hit 8.41 21.77 24.4
|
|
20 TM-side 4.82 21.36 35.1
|
|
20 TM-x0.25 5.52 20.92 24.9
|
|
20 TM-x0.5 7.35 20.36 23.6
|
|
20 PERF-SIGN 3.01 19.20 49.0
|
|
20 shuffled 5.38 21.36 38.2
|
|
25 naive 6.96 29.57 24.2
|
|
25 TM-med 12.04 29.71 14.0
|
|
25 TM-hit 11.75 29.61 18.4
|
|
25 TM-side 8.86 30.31 16.1
|
|
25 TM-x0.25 6.94 29.32 19.5
|
|
25 TM-x0.5 9.52 29.07 22.6
|
|
25 PERF-SIGN 5.06 27.22 41.3
|
|
25 shuffled 7.81 29.22 23.9
|
|
30 naive 10.37 35.68 23.4
|
|
30 TM-med 16.49 38.72 13.7
|
|
30 TM-hit 14.92 39.07 12.9
|
|
30 TM-side 11.80 39.61 13.4
|
|
30 TM-x0.25 9.43 37.27 15.5
|
|
30 TM-x0.5 10.13 37.74 17.4
|
|
30 PERF-SIGN 8.32 33.53 35.5
|
|
30 shuffled 10.82 35.67 27.6
|
|
|
|
round 8 (L= 1546) nEval/h = [449, 444, 439, 434]
|
|
h arm med p90 hit%
|
|
15 naive 4.25 14.89 40.1
|
|
15 TM-med 5.51 16.01 31.4
|
|
15 TM-hit 5.57 15.51 32.7
|
|
15 TM-side 8.92 18.26 10.7
|
|
15 TM-x0.25 4.19 14.38 36.5
|
|
15 TM-x0.5 4.95 14.63 32.3
|
|
15 PERF-SIGN 4.98 10.49 32.7
|
|
15 shuffled 10.99 19.29 18.5
|
|
20 naive 7.75 23.03 33.8
|
|
20 TM-med 7.73 25.17 10.8
|
|
20 TM-hit 9.48 29.07 6.8
|
|
20 TM-side 11.88 28.52 8.3
|
|
20 TM-x0.25 6.88 21.78 20.0
|
|
20 TM-x0.5 7.40 22.06 9.9
|
|
20 PERF-SIGN 8.10 12.16 16.2
|
|
20 shuffled 12.43 28.17 5.6
|
|
25 naive 11.52 31.02 24.4
|
|
25 TM-med 11.97 33.75 8.4
|
|
25 TM-hit 13.73 35.77 8.0
|
|
25 TM-side 13.75 35.51 6.4
|
|
25 TM-x0.25 10.42 29.90 13.2
|
|
25 TM-x0.5 9.67 30.11 6.2
|
|
25 PERF-SIGN 8.21 18.65 10.5
|
|
25 shuffled 15.01 37.19 6.6
|
|
30 naive 14.47 39.88 17.1
|
|
30 TM-med 13.82 41.47 7.1
|
|
30 TM-hit 14.73 41.51 6.2
|
|
30 TM-side 18.06 44.33 3.9
|
|
30 TM-x0.25 12.20 38.44 13.4
|
|
30 TM-x0.5 12.11 38.29 6.9
|
|
30 PERF-SIGN 11.55 23.35 10.1
|
|
30 shuffled 16.57 44.29 6.0
|
|
|
|
round 9 (L= 1286) nEval/h = [371, 366, 361, 356]
|
|
h arm med p90 hit%
|
|
15 naive 5.97 14.60 24.0
|
|
15 TM-med 5.56 13.46 27.0
|
|
15 TM-hit 5.37 13.80 26.7
|
|
15 TM-side 5.73 14.42 27.0
|
|
15 TM-x0.25 5.75 14.42 24.3
|
|
15 TM-x0.5 5.74 14.05 25.1
|
|
15 PERF-SIGN 3.97 12.60 35.0
|
|
15 shuffled 6.08 14.79 23.7
|
|
20 naive 9.31 21.51 15.8
|
|
20 TM-med 8.73 20.98 17.5
|
|
20 TM-hit 9.52 22.48 14.5
|
|
20 TM-side 9.96 22.77 12.6
|
|
20 TM-x0.25 9.05 21.15 17.5
|
|
20 TM-x0.5 9.10 21.12 16.1
|
|
20 PERF-SIGN 7.25 19.56 24.6
|
|
20 shuffled 9.25 21.24 19.4
|
|
25 naive 12.35 27.72 12.2
|
|
25 TM-med 11.16 27.17 8.3
|
|
25 TM-hit 13.17 28.72 9.1
|
|
25 TM-side 12.17 27.07 10.8
|
|
25 TM-x0.25 12.00 27.22 11.4
|
|
25 TM-x0.5 12.02 27.20 7.5
|
|
25 PERF-SIGN 10.42 26.19 19.1
|
|
25 shuffled 12.62 27.73 10.0
|
|
30 naive 14.72 31.71 11.0
|
|
30 TM-med 14.05 30.50 7.3
|
|
30 TM-hit 16.13 31.61 9.3
|
|
30 TM-side 16.31 32.14 7.0
|
|
30 TM-x0.25 14.69 31.28 8.7
|
|
30 TM-x0.5 15.72 31.27 4.8
|
|
30 PERF-SIGN 12.69 29.14 16.6
|
|
30 shuffled 14.78 32.29 7.9
|
|
|
|
round 10 (L= 1392) nEval/h = [403, 398, 393, 388]
|
|
h arm med p90 hit%
|
|
15 naive 4.35 15.34 39.0
|
|
15 TM-med 4.67 14.34 28.8
|
|
15 TM-hit 4.13 14.57 34.7
|
|
15 TM-side 4.29 15.20 29.8
|
|
15 TM-x0.25 4.32 14.89 38.7
|
|
15 TM-x0.5 4.18 14.73 36.5
|
|
15 PERF-SIGN 2.35 13.34 50.4
|
|
15 shuffled 4.44 14.98 39.5
|
|
20 naive 7.44 22.49 28.9
|
|
20 TM-med 8.41 21.46 20.1
|
|
20 TM-hit 7.10 22.18 30.4
|
|
20 TM-side 7.19 22.61 27.9
|
|
20 TM-x0.25 7.34 22.29 28.4
|
|
20 TM-x0.5 7.22 22.18 28.6
|
|
20 PERF-SIGN 5.58 20.68 37.7
|
|
20 shuffled 7.55 23.14 17.6
|
|
25 naive 11.21 30.56 20.1
|
|
25 TM-med 11.67 30.44 13.5
|
|
25 TM-hit 10.51 29.88 20.9
|
|
25 TM-side 10.93 29.52 19.3
|
|
25 TM-x0.25 10.94 30.19 21.6
|
|
25 TM-x0.5 10.82 29.49 21.9
|
|
25 PERF-SIGN 9.71 29.06 29.3
|
|
25 shuffled 11.80 31.40 11.2
|
|
30 naive 15.74 37.20 17.8
|
|
30 TM-med 14.96 37.52 8.8
|
|
30 TM-hit 16.07 39.13 11.9
|
|
30 TM-side 15.74 37.35 16.8
|
|
30 TM-x0.25 13.88 36.78 10.6
|
|
30 TM-x0.5 13.44 37.54 10.6
|
|
30 PERF-SIGN 13.74 35.20 24.5
|
|
30 shuffled 16.69 41.02 3.9
|
|
|
|
|
|
## PER-FIXTURE pooled (tr_drussgt_vs_modularbot_shield.jsonl)
|
|
h arm N med|err| p90|err| hit% miss%
|
|
15 naive 3643 3.83 14.38 41.0 59.0
|
|
15 TM-med 3643 4.71 13.67 29.3 70.7
|
|
15 TM-hit 3643 4.47 14.37 35.0 65.0
|
|
15 TM-side 3643 4.66 14.50 35.1 64.9
|
|
15 TM-x0.0 3643 3.83 14.38 41.0 59.0
|
|
15 TM-x0.25 3643 3.75 13.92 38.3 61.7
|
|
15 TM-x0.5 3643 4.32 13.63 35.8 64.2
|
|
15 PERF-SIGN 3643 2.48 11.87 47.6 52.4
|
|
15 PS-shift0 3643 3.83 14.38 41.0 59.0
|
|
15 shuffled 3643 4.82 15.41 35.7 64.3
|
|
20 naive 3593 6.50 21.96 29.5 70.5
|
|
20 TM-med 3593 7.63 21.39 17.7 82.3
|
|
20 TM-hit 3593 8.00 21.89 21.5 78.5
|
|
20 TM-side 3593 7.96 22.26 23.2 76.8
|
|
20 TM-x0.0 3593 6.50 21.96 29.5 70.5
|
|
20 TM-x0.25 3593 6.27 21.22 25.1 74.9
|
|
20 TM-x0.5 3593 6.86 21.05 23.0 77.0
|
|
20 PERF-SIGN 3593 5.00 18.74 34.6 65.4
|
|
20 PS-shift0 3593 6.50 21.96 29.5 70.5
|
|
20 shuffled 3593 7.92 22.91 21.2 78.8
|
|
25 naive 3543 9.68 29.41 20.1 79.9
|
|
25 TM-med 3543 10.94 29.71 11.8 88.2
|
|
25 TM-hit 3543 11.21 28.74 14.5 85.5
|
|
25 TM-side 3543 10.88 29.50 13.0 87.0
|
|
25 TM-x0.0 3543 9.68 29.41 20.1 79.9
|
|
25 TM-x0.25 3543 9.14 28.62 17.0 83.0
|
|
25 TM-x0.5 3543 9.30 28.15 15.1 84.9
|
|
25 PERF-SIGN 3543 7.51 25.11 25.0 75.0
|
|
25 PS-shift0 3543 9.68 29.41 20.1 79.9
|
|
25 shuffled 3543 11.39 30.95 13.1 86.9
|
|
30 naive 3493 12.97 36.08 15.7 84.3
|
|
30 TM-med 3493 13.78 36.34 9.7 90.3
|
|
30 TM-hit 3493 14.23 36.47 9.4 90.6
|
|
30 TM-side 3493 13.68 37.39 10.0 90.0
|
|
30 TM-x0.0 3493 12.97 36.08 15.7 84.3
|
|
30 TM-x0.25 3493 12.20 35.70 12.9 87.1
|
|
30 TM-x0.5 3493 11.91 35.68 10.2 89.8
|
|
30 PERF-SIGN 3493 9.61 30.76 18.2 81.8
|
|
30 PS-shift0 3493 12.97 36.08 15.7 84.3
|
|
30 shuffled 3493 14.36 38.36 10.0 90.0
|
|
|
|
## PER-FIXTURE learnability (tr_drussgt_vs_modularbot_shield.jsonl)
|
|
h N TMquad% shufquad% TMside% majQuad% pred[Ls Lb Rs Rb] / true[Ls Lb Rs Rb]
|
|
15 3643 35.5 24.1 61.7 26.3 [1057 1007 549 1030 ] / [959 887 906 891 ]
|
|
20 3593 34.0 25.6 61.1 25.4 [1008 1037 537 1011 ] / [913 893 898 889 ]
|
|
25 3543 33.7 23.4 60.9 25.6 [981 1000 564 998 ] / [860 908 870 905 ]
|
|
30 3493 32.5 26.5 60.8 26.1 [938 1032 541 982 ] / [810 912 869 902 ]
|
|
|
|
==========================================================================================
|
|
|
|
## POOLED PRIMARY (tr_drussgt_vs_modularbot*): arm table
|
|
h arm N med|err| p90|err| hit% miss%
|
|
15 naive 9433 4.08 14.24 39.3 60.7
|
|
15 TM-med 9433 4.81 13.53 29.0 71.0
|
|
15 TM-hit 9433 4.46 14.29 35.1 64.9
|
|
15 TM-side 9433 4.40 14.12 36.0 64.0
|
|
15 TM-x0.0 9433 4.08 14.24 39.3 60.7
|
|
15 TM-x0.25 9433 3.99 13.84 37.9 62.1
|
|
15 TM-x0.5 9433 4.30 13.63 36.3 63.7
|
|
15 PERF-SIGN 9433 2.57 11.88 47.0 53.0
|
|
15 PS-shift0 9433 4.08 14.24 39.3 60.7
|
|
15 shuffled 9433 4.68 14.97 35.1 64.9
|
|
20 naive 9308 6.81 21.60 28.2 71.8
|
|
20 TM-med 9308 7.61 20.85 19.2 80.8
|
|
20 TM-hit 9308 7.75 21.62 22.8 77.2
|
|
20 TM-side 9308 7.36 21.96 24.8 75.2
|
|
20 TM-x0.0 9308 6.81 21.60 28.2 71.8
|
|
20 TM-x0.25 9308 6.58 21.08 25.6 74.4
|
|
20 TM-x0.5 9308 6.85 20.82 23.9 76.1
|
|
20 PERF-SIGN 9308 5.22 18.50 33.4 66.6
|
|
20 PS-shift0 9308 6.81 21.60 28.2 71.8
|
|
20 shuffled 9308 7.81 22.33 22.0 78.0
|
|
25 naive 9183 9.71 29.13 20.9 79.1
|
|
25 TM-med 9183 10.33 28.63 13.1 86.9
|
|
25 TM-hit 9183 10.22 28.90 15.4 84.6
|
|
25 TM-side 9183 10.12 29.21 15.3 84.7
|
|
25 TM-x0.0 9183 9.71 29.13 20.9 79.1
|
|
25 TM-x0.25 9183 9.31 28.61 18.8 81.2
|
|
25 TM-x0.5 9183 9.31 28.29 17.4 82.6
|
|
25 PERF-SIGN 9183 7.30 25.26 25.6 74.4
|
|
25 PS-shift0 9183 9.71 29.13 20.9 79.1
|
|
25 shuffled 9183 10.96 30.19 16.0 84.0
|
|
30 naive 9058 12.62 36.12 16.4 83.6
|
|
30 TM-med 9058 13.09 35.17 10.5 89.5
|
|
30 TM-hit 9058 13.27 36.28 11.4 88.6
|
|
30 TM-side 9058 12.94 36.88 12.3 87.7
|
|
30 TM-x0.0 9058 12.62 36.12 16.4 83.6
|
|
30 TM-x0.25 9058 12.12 35.75 14.1 85.9
|
|
30 TM-x0.5 9058 11.93 35.63 12.1 87.9
|
|
30 PERF-SIGN 9058 9.72 30.82 18.3 81.7
|
|
30 PS-shift0 9058 12.62 36.12 16.4 83.6
|
|
30 shuffled 9058 13.90 38.14 11.2 88.8
|
|
|
|
## DELTAS (percentage points of estimated hit fraction)
|
|
h naive% dTMmed dTMhit dTMside dTMs0 dTMs25 dTMs50 dPS dPS0 dShuf (hit pp vs naive)
|
|
15 39.3 -10.3 -4.2 -3.3 +0.0 -1.5 -3.0 +7.7 +0.0 -4.2
|
|
20 28.2 -9.0 -5.4 -3.4 +0.0 -2.6 -4.3 +5.1 +0.0 -6.2
|
|
25 20.9 -7.8 -5.5 -5.6 +0.0 -2.2 -3.6 +4.7 +0.0 -4.9
|
|
30 16.4 -5.9 -5.0 -4.1 +0.0 -2.3 -4.3 +1.9 +0.0 -5.2
|
|
|
|
## POOLED learnability diagnostics (is the learner learning?)
|
|
h N TMquad% shufquad% TMside% majQuad% pred[Ls Lb Rs Rb] / true[Ls Lb Rs Rb]
|
|
15 9433 35.2 22.9 59.6 25.5 [2294 2615 1447 3077 ] / [2407 2314 2377 2335 ]
|
|
20 9308 34.8 23.6 59.8 25.2 [2281 2579 1386 3062 ] / [2305 2331 2343 2329 ]
|
|
25 9183 34.3 23.8 59.7 25.3 [2208 2558 1366 3051 ] / [2228 2322 2315 2318 ]
|
|
30 9058 34.3 24.8 61.0 25.6 [2192 2501 1375 2990 ] / [2150 2273 2321 2314 ]
|
|
|
|
## FITTED HIT-OPTIMAL SHIFT MAGNITUDES
|
|
(median across rounds of the fitted HIT-OPTIMAL shift, deg)
|
|
h TM[Ls Lb Rs Rb] PS[left right] shuf[Ls Lb Rs Rb]
|
|
15 [ 0.00 1.50 0.00 -1.00 ] [ 2.00 -2.00 ] [ 0.00 0.00 0.00 -0.50 ]
|
|
20 [ 0.00 2.00 -0.50 -1.00 ] [ 2.00 -2.00 ] [ 0.00 0.00 -0.50 0.00 ]
|
|
25 [ 0.50 5.50 0.00 -2.71 ] [ 2.50 -2.50 ] [ 0.00 0.00 1.00 1.50 ]
|
|
30 [ 1.50 8.00 0.00 -5.00 ] [ 3.50 -3.00 ] [ 0.00 1.00 1.00 1.50 ]
|
|
|
|
## HIT FRACTION vs CONSTANT SHIFT MAGNITUDE (calibration)
|
|
shift h=15 h=20 h=25 h=30 (hit% on calibration, pooled)
|
|
-20.0 1.2 4.0 4.2 4.4
|
|
-18.0 2.1 4.5 4.7 4.8
|
|
-16.0 3.4 5.1 5.7 5.2
|
|
-14.0 5.1 6.3 5.9 5.7
|
|
-12.0 6.6 7.4 6.4 5.6
|
|
-10.0 8.4 8.1 7.2 6.2
|
|
-8.0 10.2 8.6 7.6 6.4
|
|
-6.0 11.3 8.9 7.6 6.3
|
|
-4.0 14.2 11.8 10.2 7.8
|
|
-2.0 26.8 19.2 14.4 10.9
|
|
+0.0 36.2 25.3 18.0 12.8
|
|
+2.0 27.4 19.9 15.4 12.9
|
|
+4.0 14.1 12.5 11.3 11.0
|
|
+6.0 11.7 9.2 9.0 8.5
|
|
+8.0 10.7 8.6 7.7 7.2
|
|
+10.0 8.3 8.0 6.8 5.7
|
|
+12.0 6.4 7.3 6.2 5.4
|
|
+14.0 4.7 6.4 5.8 5.5
|
|
+16.0 3.6 5.6 5.6 5.1
|
|
+18.0 2.1 4.6 5.6 5.2
|
|
+20.0 1.1 3.7 4.8 4.8
|
|
|
|
## SIDE-ACCURACY SWEEP (synthetic q-accurate predictor, out-of-sample)
|
|
q fitShift(h=20)[L R] hit% h15 h20 h25 h30 d(hit) pp: h15 h20 h25 h30 mean
|
|
0.50 [ 0.00 -0.50] 37.8 25.1 16.8 11.5 -1.5 -3.1 -4.1 -4.9 -3.4
|
|
0.55 [ 0.50 0.00] 37.6 25.8 16.6 11.5 -1.7 -2.4 -4.3 -4.8 -3.3
|
|
0.60 [ 0.50 -0.50] 38.0 25.7 17.0 12.0 -1.3 -2.5 -3.9 -4.3 -3.0
|
|
0.65 [ 1.00 -1.00] 39.2 25.1 17.7 12.6 -0.1 -3.1 -3.3 -3.8 -2.6
|
|
0.70 [ 1.50 -1.00] 39.4 26.5 19.0 13.6 +0.1 -1.8 -2.0 -2.8 -1.6
|
|
0.80 [ 1.50 -1.50] 40.5 28.5 20.7 15.0 +1.2 +0.2 -0.3 -1.4 -0.0
|
|
1.00 [ 2.00 -2.00] 47.0 33.4 25.6 18.3 +7.7 +5.1 +4.7 +1.9 +4.9
|
|
|
|
==========================================================================================
|
|
## PER-ROUND hit-delta integrity (pp vs naive)
|
|
fixture round h TMmed TMhit TMside TMs25 TMs50 PS shuf
|
|
tr_drussgt_vs_modularbot.jsonl 1 15 -0.3 -5.8 -1.7 +0.3 -0.8 +16.1 -4.7
|
|
tr_drussgt_vs_modularbot.jsonl 1 20 +3.1 +2.2 +1.1 +4.2 +5.1 +20.2 +1.7
|
|
tr_drussgt_vs_modularbot.jsonl 1 25 -4.0 -2.3 -1.1 +4.6 +4.0 +18.8 -0.9
|
|
tr_drussgt_vs_modularbot.jsonl 1 30 -6.6 -6.1 -5.2 -0.3 -0.9 +7.2 +0.6
|
|
tr_drussgt_vs_modularbot.jsonl 2 15 -19.0 -7.3 -2.4 -5.1 -8.5 +7.1 -5.1
|
|
tr_drussgt_vs_modularbot.jsonl 2 20 -10.4 -2.5 +1.0 -4.0 -4.7 -2.5 -9.9
|
|
tr_drussgt_vs_modularbot.jsonl 2 25 -7.0 -10.2 -3.5 -8.2 -7.8 -3.5 -2.5
|
|
tr_drussgt_vs_modularbot.jsonl 2 30 -8.1 -7.8 -6.1 -4.6 -5.8 -6.8 -14.4
|
|
tr_drussgt_vs_modularbot.jsonl 3 15 +0.4 -1.3 -2.9 +0.4 +0.0 +5.9 -1.7
|
|
tr_drussgt_vs_modularbot.jsonl 3 20 -3.0 -4.3 -4.3 -0.4 -2.1 +8.5 +0.4
|
|
tr_drussgt_vs_modularbot.jsonl 3 25 -7.9 -13.1 -1.7 -3.5 -7.4 +14.8 +0.4
|
|
tr_drussgt_vs_modularbot.jsonl 3 30 +2.2 +1.3 -2.2 -4.5 -2.2 +4.5 -2.2
|
|
tr_drussgt_vs_modularbot.jsonl 4 15 +0.5 -1.9 +0.5 +2.6 +2.1 +9.8 -4.3
|
|
tr_drussgt_vs_modularbot.jsonl 4 20 -1.2 +1.7 +2.4 +2.7 +1.9 +15.5 +0.5
|
|
tr_drussgt_vs_modularbot.jsonl 4 25 +1.2 -2.0 +0.5 -1.2 +1.0 +10.0 +2.0
|
|
tr_drussgt_vs_modularbot.jsonl 4 30 +1.7 +4.2 +2.5 +0.2 +1.5 +7.4 -0.7
|
|
tr_drussgt_vs_modularbot.jsonl 5 15 +0.7 -0.5 -2.5 +0.0 +0.2 +9.8 -2.0
|
|
tr_drussgt_vs_modularbot.jsonl 5 20 -3.1 -0.9 -2.0 +0.4 -0.4 +7.7 -4.7
|
|
tr_drussgt_vs_modularbot.jsonl 5 25 -10.1 -3.7 -5.5 -0.6 -0.4 +9.4 -2.6
|
|
tr_drussgt_vs_modularbot.jsonl 5 30 -5.8 -0.7 -2.2 -1.5 -0.7 +8.9 -3.0
|
|
tr_drussgt_vs_modularbot.jsonl 6 15 -3.2 +0.3 +0.0 +0.3 +0.3 +6.8 -1.0
|
|
tr_drussgt_vs_modularbot.jsonl 6 20 +1.0 -0.3 +1.0 -1.0 -1.0 +6.6 +0.0
|
|
tr_drussgt_vs_modularbot.jsonl 6 25 +3.3 -0.7 -0.7 +1.7 +2.0 +5.4 +1.0
|
|
tr_drussgt_vs_modularbot.jsonl 6 30 +7.1 -1.0 -1.0 +0.7 +0.0 +4.8 -0.7
|
|
tr_drussgt_vs_modularbot.jsonl 7 15 -6.4 -4.5 +0.2 -3.1 -3.1 +9.2 -8.7
|
|
tr_drussgt_vs_modularbot.jsonl 7 20 -10.3 -2.4 +0.2 -1.0 -1.2 +10.7 -4.1
|
|
tr_drussgt_vs_modularbot.jsonl 7 25 -7.7 +0.7 +0.2 +0.7 +1.0 +7.5 -3.1
|
|
tr_drussgt_vs_modularbot.jsonl 7 30 -3.7 -1.2 -1.7 +0.5 -1.7 +8.6 +0.2
|
|
tr_drussgt_vs_modularbot.jsonl 8 15 +1.4 -0.3 -1.7 -3.2 +0.3 +10.6 -0.3
|
|
tr_drussgt_vs_modularbot.jsonl 8 20 +1.5 +1.2 -0.6 -2.3 -0.6 +7.6 +2.6
|
|
tr_drussgt_vs_modularbot.jsonl 8 25 -4.4 -4.1 -3.8 -4.7 -5.0 +3.8 +1.8
|
|
tr_drussgt_vs_modularbot.jsonl 8 30 -4.2 -2.7 -4.5 -2.1 -4.2 +3.3 -6.3
|
|
tr_drussgt_vs_modularbot.jsonl 9 15 -29.4 +0.0 -1.2 +0.2 +0.4 +10.7 -1.9
|
|
tr_drussgt_vs_modularbot.jsonl 9 20 -25.2 -28.1 -3.1 -11.2 -25.8 -3.7 -10.6
|
|
tr_drussgt_vs_modularbot.jsonl 9 25 -25.0 -26.2 -26.4 -8.3 -20.9 -8.7 -16.1
|
|
tr_drussgt_vs_modularbot.jsonl 9 30 -19.7 -22.1 -13.1 -7.8 -18.5 -16.5 -12.2
|
|
tr_drussgt_vs_modularbot.jsonl 10 15 -12.5 -7.5 -9.0 +0.8 -5.2 -3.0 -2.2
|
|
tr_drussgt_vs_modularbot.jsonl 10 20 -6.1 -3.0 -6.3 +2.3 +1.8 -2.3 -0.5
|
|
tr_drussgt_vs_modularbot.jsonl 10 25 -7.2 -1.5 +1.0 +3.8 +3.1 -0.8 -3.8
|
|
tr_drussgt_vs_modularbot.jsonl 10 30 -4.2 +0.8 +0.0 +1.6 +1.8 +7.8 -1.6
|
|
tr_drussgt_vs_modularbot.jsonl 11 15 -1.5 -0.3 +0.3 -4.6 -2.8 +6.5 -6.8
|
|
tr_drussgt_vs_modularbot.jsonl 11 20 -3.4 -2.8 -4.7 -2.8 +0.3 +2.8 -5.6
|
|
tr_drussgt_vs_modularbot.jsonl 11 25 -3.5 -2.5 -3.5 -2.5 -2.5 +0.3 -2.2
|
|
tr_drussgt_vs_modularbot.jsonl 11 30 -0.3 -2.3 -1.6 -3.6 -4.5 +1.0 -4.2
|
|
tr_drussgt_vs_modularbot.jsonl 12 15 -0.6 -1.7 -1.7 +0.0 -0.6 +9.4 -6.3
|
|
tr_drussgt_vs_modularbot.jsonl 12 20 +0.6 -2.5 +0.0 +0.6 -0.6 +5.3 -6.1
|
|
tr_drussgt_vs_modularbot.jsonl 12 25 -0.8 -7.1 -2.3 +1.7 +1.4 +5.4 -2.5
|
|
tr_drussgt_vs_modularbot.jsonl 12 30 -9.8 +2.6 -8.9 +2.0 +2.9 +3.7 -9.2
|
|
tr_drussgt_vs_modularbot.jsonl 13 15 -16.6 -12.7 +0.3 -0.3 -5.7 +6.8 -5.5
|
|
tr_drussgt_vs_modularbot.jsonl 13 20 -8.2 -8.4 -8.2 -9.5 -8.2 +10.5 -2.9
|
|
tr_drussgt_vs_modularbot.jsonl 13 25 -3.2 -5.3 +0.5 -4.5 -4.8 +6.4 +1.9
|
|
tr_drussgt_vs_modularbot.jsonl 13 30 -1.4 -3.0 +2.7 -3.0 -2.2 +8.4 +0.0
|
|
tr_drussgt_vs_modularbot.jsonl 14 15 -19.2 +0.5 +0.2 +0.0 +0.0 +8.2 -0.5
|
|
tr_drussgt_vs_modularbot.jsonl 14 20 -13.5 +3.5 +2.6 +0.9 +0.9 +7.6 -5.2
|
|
tr_drussgt_vs_modularbot.jsonl 14 25 -12.4 -1.2 -1.0 -0.2 +0.5 -1.9 -12.7
|
|
tr_drussgt_vs_modularbot.jsonl 14 30 -10.7 -12.8 +1.0 -3.9 -11.1 -6.3 -6.3
|
|
tr_drussgt_vs_modularbot.jsonl 15 15 -26.8 -3.5 -4.5 +0.6 -2.9 +10.5 -1.0
|
|
tr_drussgt_vs_modularbot.jsonl 15 20 -22.7 -3.9 -4.9 +1.9 -2.3 -18.5 -26.6
|
|
tr_drussgt_vs_modularbot.jsonl 15 25 -15.2 +2.3 -16.8 -0.7 -0.7 +8.9 -7.6
|
|
tr_drussgt_vs_modularbot.jsonl 15 30 -13.4 -3.0 -2.0 -1.3 -1.0 -8.7 -9.4
|
|
tr_drussgt_vs_modularbot_shield.jsonl 1 15 +0.0 +0.0 +0.0 +0.0 +0.0 +0.0 +0.0
|
|
tr_drussgt_vs_modularbot_shield.jsonl 1 20 +0.0 +0.0 +0.0 +0.0 +0.0 +0.0 +0.0
|
|
tr_drussgt_vs_modularbot_shield.jsonl 1 25 +0.0 -2.6 -2.6 +0.0 -2.0 +0.0 -2.6
|
|
tr_drussgt_vs_modularbot_shield.jsonl 1 30 +0.0 +0.0 +0.0 +0.0 +0.0 +1.4 +0.0
|
|
tr_drussgt_vs_modularbot_shield.jsonl 2 15 -18.7 -8.1 +1.9 +1.0 -1.6 +10.0 -5.8
|
|
tr_drussgt_vs_modularbot_shield.jsonl 2 20 -11.1 -2.3 +2.0 +1.3 +1.0 +3.3 -4.6
|
|
tr_drussgt_vs_modularbot_shield.jsonl 2 25 -10.7 -9.7 -5.0 -5.3 -10.7 -2.3 -12.3
|
|
tr_drussgt_vs_modularbot_shield.jsonl 2 30 -8.8 -9.5 -1.7 -8.5 -10.5 +0.7 -3.1
|
|
tr_drussgt_vs_modularbot_shield.jsonl 3 15 -24.4 -24.4 -16.5 -9.2 -23.8 +2.2 -16.8
|
|
tr_drussgt_vs_modularbot_shield.jsonl 3 20 -22.4 -22.2 -22.2 -20.2 -22.2 +5.7 -15.1
|
|
tr_drussgt_vs_modularbot_shield.jsonl 3 25 -13.0 -10.1 -12.7 -8.1 -12.7 +11.8 -15.3
|
|
tr_drussgt_vs_modularbot_shield.jsonl 3 30 -8.5 -8.5 -7.9 +0.9 -8.2 -7.0 -10.5
|
|
tr_drussgt_vs_modularbot_shield.jsonl 4 15 -27.8 -0.3 -1.1 +0.3 -0.8 +7.0 -5.6
|
|
tr_drussgt_vs_modularbot_shield.jsonl 4 20 -26.3 -4.1 -3.3 +1.1 +1.9 +3.0 -20.1
|
|
tr_drussgt_vs_modularbot_shield.jsonl 4 25 -18.7 -8.0 -23.1 +0.0 -0.5 +4.7 -14.6
|
|
tr_drussgt_vs_modularbot_shield.jsonl 4 30 -8.4 -12.5 -16.2 +0.6 -3.3 -1.4 -14.2
|
|
tr_drussgt_vs_modularbot_shield.jsonl 5 15 +0.2 -1.1 +0.4 -2.7 -2.4 +9.1 +1.6
|
|
tr_drussgt_vs_modularbot_shield.jsonl 5 20 -0.7 -4.5 -0.7 +0.0 -2.5 +15.1 -1.1
|
|
tr_drussgt_vs_modularbot_shield.jsonl 5 25 +0.9 +0.7 +4.1 -0.5 +0.2 +10.2 +7.5
|
|
tr_drussgt_vs_modularbot_shield.jsonl 5 30 +0.2 +1.6 +2.5 +1.1 +0.7 +9.7 +0.9
|
|
tr_drussgt_vs_modularbot_shield.jsonl 6 15 -15.6 +0.5 -0.5 +0.8 +1.6 +11.6 +0.3
|
|
tr_drussgt_vs_modularbot_shield.jsonl 6 20 -7.1 -2.5 -3.0 +0.0 -1.1 +8.2 -3.0
|
|
tr_drussgt_vs_modularbot_shield.jsonl 6 25 -1.1 -0.3 -1.4 +0.6 +0.3 +3.3 -2.8
|
|
tr_drussgt_vs_modularbot_shield.jsonl 6 30 +0.0 -2.5 -2.5 -0.8 -2.0 +2.0 -3.7
|
|
tr_drussgt_vs_modularbot_shield.jsonl 7 15 -12.9 -15.2 +0.3 -10.6 -12.2 +9.9 -1.3
|
|
tr_drussgt_vs_modularbot_shield.jsonl 7 20 -13.8 -10.0 +0.8 -9.5 -10.8 +14.6 +3.8
|
|
tr_drussgt_vs_modularbot_shield.jsonl 7 25 -10.1 -5.7 -8.1 -4.7 -1.6 +17.1 -0.3
|
|
tr_drussgt_vs_modularbot_shield.jsonl 7 30 -9.7 -10.5 -10.0 -7.9 -6.1 +12.1 +4.2
|
|
tr_drussgt_vs_modularbot_shield.jsonl 8 15 -8.7 -7.3 -29.4 -3.6 -7.8 -7.3 -21.6
|
|
tr_drussgt_vs_modularbot_shield.jsonl 8 20 -23.0 -27.0 -25.5 -13.7 -23.9 -17.6 -28.2
|
|
tr_drussgt_vs_modularbot_shield.jsonl 8 25 -15.9 -16.4 -18.0 -11.2 -18.2 -13.9 -17.8
|
|
tr_drussgt_vs_modularbot_shield.jsonl 8 30 -9.9 -10.8 -13.1 -3.7 -10.1 -6.9 -11.1
|
|
tr_drussgt_vs_modularbot_shield.jsonl 9 15 +3.0 +2.7 +3.0 +0.3 +1.1 +11.1 -0.3
|
|
tr_drussgt_vs_modularbot_shield.jsonl 9 20 +1.6 -1.4 -3.3 +1.6 +0.3 +8.7 +3.6
|
|
tr_drussgt_vs_modularbot_shield.jsonl 9 25 -3.9 -3.0 -1.4 -0.8 -4.7 +6.9 -2.2
|
|
tr_drussgt_vs_modularbot_shield.jsonl 9 30 -3.7 -1.7 -3.9 -2.2 -6.2 +5.6 -3.1
|
|
tr_drussgt_vs_modularbot_shield.jsonl 10 15 -10.2 -4.2 -9.2 -0.2 -2.5 +11.4 +0.5
|
|
tr_drussgt_vs_modularbot_shield.jsonl 10 20 -8.8 +1.5 -1.0 -0.5 -0.3 +8.8 -11.3
|
|
tr_drussgt_vs_modularbot_shield.jsonl 10 25 -6.6 +0.8 -0.8 +1.5 +1.8 +9.2 -8.9
|
|
tr_drussgt_vs_modularbot_shield.jsonl 10 30 -9.0 -5.9 -1.0 -7.2 -7.2 +6.7 -13.9
|
|
|
|
==========================================================================================
|
|
## VERDICT INPUTS (pooled hit pp vs naive)
|
|
h= 15 PERF-SIGN +7.7 pp TM-side -3.3 pp TM-quad -4.2 pp shuffled -4.2 pp
|
|
h= 20 PERF-SIGN +5.1 pp TM-side -3.4 pp TM-quad -5.4 pp shuffled -6.2 pp
|
|
h= 25 PERF-SIGN +4.7 pp TM-side -5.6 pp TM-quad -5.5 pp shuffled -4.9 pp
|
|
h= 30 PERF-SIGN +1.9 pp TM-side -4.1 pp TM-quad -5.0 pp shuffled -5.2 pp
|