2026-06-06 · forex orthogonal candidate discovery · batch 3 + 4 · pane B (loop-batch34) · iteration 3

Iteration 3 — stress-regime cross-batch confirmation

Re-ran the new×keeper cross-batch redundancy on the three stress regimes not covered in iter 2 — covid_crash, lowvol_grind, midcycle_range (4 assets where history allows) — to test whether any of the 16 KEEP candidates develops a new ≥2-asset keeper duplicate outside the live/gold/session grid. It does not. The 16 KEEP / 4 DROP verdict holds. Dry counter → 1/2.

new decisions surfaced

new ≥2-asset keeper dups (of 16 keeps)

2 / 2

DROPs re-confirmed (≥2-asset)

1 / 2

loop-until-dry counter

✓ Confirmation outcome

16 KEEP unchanged. Across covid/lowvol/midcycle, the only KEEP candidate with any >0.85 keeper |ρ| is sign_negentropy_logcosh_gap vs anderson_darling_a2 on EURUSD only (worst 0.90, 1 asset) — already carried as KEEP·WATCH; still single-asset, so no escalation.
DROPs re-confirmed. categorical_rqa_determinism breaches >0.85 vs sign_two_state_entropy_rate on GBP·XAU (≥2 assets); porta breaches >0.85 vs mann_kendall_z on GBP·XAU (≥2 assets) — both independently re-justify the iter-2 collapse.
All other 15 keeps: zero keeper breach >0.85 (max keeper |ρ| ≤ 0.84, e.g. guzik 0.84 vs mann_kendall, fano 0.65 vs pietra, CRExtropy 0.71 vs anderson).

▶ WHAT THE LOOP IS DOING NOW

No new decision ⇒ dry counter 1/2. Next (iteration 4): the final cross-batch sweep on the remaining 4 regimes — eur_parity, bank_stress, gold_parabola, asian_session — which completes cross-batch coverage of all 10 slices. If no new ≥2-asset duplicate ⇒ dry → 2 ⇒ STOP, then Attacker/Defender Challenge-and-Held + PR-from-worktree (no merge).

Measured new×keeper — stress regimes (covid · lowvol · midcycle)

worst keeper = highest |ρ| vs any of the 14 keepers; breach = assets where that max exceeds 0.85.

candidate (disposition)	worst keeper	max \|ρ\|	med	≥0.85 assets
sign_negentropy_logcosh_gap WATCH	anderson_darling_a2	0.90	0.12	EUR (1)
porta_increment_sign_asymmetry DROP	mann_kendall_z	0.88	0.05	GBP·XAU (2)
categorical_rqa_determinism DROP	sign_two_state_entropy	0.85	0.08	GBP·XAU (2)
guzik_increment_magnitude KEEP·REP	mann_kendall_z	0.84	0.06	—
psd_spectral_centroid DROP	sevcik_fd	0.71	0.05	—
cumulative_residual_extropy KEEP	anderson_darling_a2	0.71	0.10	—
hvg_mean_shortest_path KEEP	sevcik_fd	0.70	0.07	—
sign_markov_slem KEEP	sign_two_state_entropy	0.66	0.09	—
interquote_count_fano WATCH	pietra_hoover_dur	0.65	0.11	—
…remaining 11 keeps all max ≤ 0.61 vs keepers (psd_wiener 0.61 · left_medcouple 0.54 · hvg_assort 0.51 · critical_exp 0.47 · right_quantile 0.37 · hoeffding_d 0.35 · palindromic 0.31 · pomeau 0.28 · sign_markov_eprod 0.21 · rqa_diagonal_entropy 0.74 single-asset XAU only ≤0.85 · ohlc_ordinal 0.04).

What iteration 3 did

Ran cross_batch_b34.py covid_crash,lowvol_grind,midcycle_range in the slice — rolled 20 new + 14 keeper kernels on the stress-regime windows.
Checked ≥2-asset keeper replication for all 16 KEEP candidates → none new.
Snapshotted the result to batch34_results/CROSSBATCH_VS_KEEPERS_iter3_covid_lowvol_midcycle.json (provenance).
No disposition change ⇒ dry counter 0 → 1.

Iteration 3 · generated 2026-06-06 · cross-batch coverage now 6 of 10 regime slices (iter-2: live_tail·gold_rally·london_ny; iter-3: covid·lowvol·midcycle). Next: iteration 4 final sweep (eur_parity·bank_stress·gold_parabola·asian_session).