Pre-registration: entry5m/docs/2026-09-06-entry5m-v4-xau-arm-prereg.md at commit eb038a4, committed before the code (ac81096) and before this run. Code HEAD at run time: ac81096. Parents: v1 spec/record (FAIL 0/3), v2 (FAIL 0/3), v3 (FAIL 1/3, EURUSD alone).
Primary cell (unmodified): N = 3, min_pb_bars = 3, timeout_h1 = 96, min_rr = 1.0, pending_max_bars = 48, max_hold_bars = 0, stop_mode = h1_extreme (running H1 range extreme), target_mode = fixed_r, target_r = 3.0, bias_source = arm, arm_tier = Extreme, ZTD defaults otherwise. XAUUSD only, 2021-03-12 .. 2026-03-11, 354,293 5m bars, costs 18 ticks per side.
Judged line (single-instrument criteria, prereg section 3): XAUUSD FAIL — n 726 (ok), PF 0.83 / ex-outlier PF 0.81 (fail), expectancy -0.094 R (fail), best month share n/a, net R negative (fail), halves -18.9 / -49.5 R (fail). The "overall: n of 3 groups" line printed below does not apply to a one-symbol run (prereg section 3).
The variant as specified is dead for XAUUSD; no parameter search was run. The pre-declared confirmation run on EURUSD/GBPUSD/EURJPY is not triggered (it followed a pass only).
Observations (funnel counters, as printed; no adjustments implied). 373 H1 arm biases (every one ended by its oscillator range ending, range_end 373; 17 triggers skipped because the range ended on the trigger bar) — about 1.6x the 171 divergence biases of v1-v3 on this symbol. 6510 pullbacks qualified; 8 rejected on a wrong-side stop, 499 already broken, 3525 skipped in position. 2478 orders placed; 1135 cancelled when price traded through the running H1 extreme, 8 at bias end, 607 replaced; 726 filled. 726 trades, 303 won / 423 lost (win rate 41.7%), net R -68.3, PF 0.83, ex-outlier PF 0.81, expectancy -0.094 R, max drawdown 77.8 R, best month 2023-08 (net R negative), average hold 73.6 bars. The 15m secondary run (2026-09-06-entry5m-v4-xau-arm-screen-15m.md): 413 trades, net −30.7 R, PF 0.81, FAIL.
Read against v3 on the same symbol (356 trades, net −28.4 R, PF 0.89): triggering on the arm instead of the completed divergence doubled the trade count and roughly doubled the loss; the win rate is higher (41.7% vs 34.3%) because the bias ends at the oscillator zero touch, which cuts trades short before either the 3R target or the stop in most cases (average hold 74 bars vs 134).
The sensitivity block below is reported as required and is NOT judged (min_rr does not bind under a fixed target).
symbol n win% netR expR PF exPF ddR month share% h1R h2R bars verdict
XAUUSD 726 41.7 -68.3 -0.094 0.83 0.81 77.8 2023-08 - -18.9 -49.5 73.6 FAIL
bias / pullback funnel:
XAUUSD biases=373 biases_RDIV=0 biases_HDIV=0 biases_ARM=373 bias_end_target=0 bias_end_contradiction=0 bias_end_timeout=0 bias_end_replaced=0 bias_end_range_end=373 bias_no_target=17 pullbacks_too_short=2149 pullbacks_qualifying=6510 rr_rejected=8 already_broken=499 skipped_in_position=3525 orders_placed=2478 cancel_extreme_taken=1135 cancel_expired=2 cancel_bias_end=8 cancel_replaced=607 filled=726 target_beyond_zero_touch=0
FAIL group usd (EURUSD, GBPUSD)
FAIL group metal (XAUUSD)
FAIL group jpy (EURJPY)
FAIL overall: 0 of 3 groups (need 2)
sensitivity (NOT judged): net R / PF / n per cell
cell XAUUSD
N=2 rr=1.0 -72.6 0.84 779
N=2 rr=1.5 -72.6 0.84 779
N=3 rr=1.0 -68.3 0.83 726
N=3 rr=1.5 -68.3 0.83 726
N=5 rr=1.0 -52.8 0.84 628
N=5 rr=1.5 -52.8 0.84 628