2026-06-06 · forex orthogonal candidate discovery · batch 3 + 4 · pane B (loop-batch34) · iteration 5

Iteration 5 — pooled-union cross-batch verification

With cross-batch now covering all 10 regime slices (iter 2 + 3 + 4), this iteration pools the >0.85 keeper-breach assets per candidate across every grid — to catch any candidate that reaches ≥2 distinct assets only when unioned across regimes (the failure mode that the negentropy DROP exposed in iter 4). Result: no new ≥2-asset duplicate among the 15 KEEPs. 15 KEEP / 5 DROP holds. Dry counter → 1/2.

new decisions

13 / 15

keeps with zero keeper breach

keep with pooled breach (XAU-only, 1 asset)

1 / 2

loop-until-dry counter

✓ Pooled-union outcome

13 of 15 keeps have zero keeper |ρ| > 0.85 in any of the 174-cell screen or the 10-slice cross-batch grids (pooled worst-keeper ≤ 0.84: guzik 0.84 · CRExtropy 0.80 · hvg_mean_path 0.72 · psd_wiener 0.71 · interquote_fano 0.69 · sign_markov_slem 0.66 · left_medcouple 0.60 · hvg_assort 0.58 · right_quantile 0.57 · critical_exp 0.49 · pomeau/hoeffding 0.42 · palindromic 0.31 · sign_markov_eprod 0.22).
1 of 15 — categorical_rqa_diagonal_entropy — carries a pooled keeper breach, but on XAUUSD only (1 asset, worst 0.96 vs sign_two_state_entropy_rate in gold-saturated regimes). Below the ≥2-asset DROP bar ⇒ stays KEEP·REP·WATCH.
No split-redundancy: no KEEP candidate breaches different keepers on different assets across grids — every breach that exists is against a single consistent keeper, so the per-worst-keeper union is the complete picture.
The 4 cross-batch/intra-batch DROPs re-confirm their ≥2-asset (or all-4-asset) pooled breaches.

Pooled cross-batch table (union over all 10 slices)

pooled worst = highest |ρ| vs any keeper across the 174-cell + 10-slice grids; pooled breach = union of assets where that keeper exceeds 0.85.

candidate	pooled worst	keeper	pooled ≥0.85 assets	status
categorical_rqa_determinism	0.98	sign_two_state_entropy	GBP·XAG·XAU (3)	DROP
categorical_rqa_diagonal_entropy	0.96	sign_two_state_entropy	XAU (1)	KEEP·REP·WATCH
sign_negentropy_logcosh_gap	0.90	anderson_darling_a2	EUR·XAG·XAU (3)	DROP
porta_increment_sign_asymmetry	0.88	mann_kendall_z	EUR·GBP·XAU (3)	DROP
guzik_increment_magnitude	0.84	mann_kendall_z	none	KEEP·REP
cumulative_residual_extropy	0.80	anderson / mean_cosine	none	KEEP
psd_spectral_centroid	0.76	sevcik_fd	none	DROP (intra)
hvg_mean_shortest_path	0.72	sevcik_fd	none	KEEP
psd_wiener_spectral_flatness	0.71	sevcik_fd	none	KEEP·REP
interquote_count_fano_factor	0.69	pietra_hoover_dur	none	KEEP·WATCH
…remaining 9 keeps all pooled worst ≤ 0.66 (sign_markov_slem 0.66 · left_medcouple 0.60 · hvg_assort 0.58 · right_quantile 0.57 · critical_exp 0.49 · pomeau 0.42 · hoeffding_d 0.42 · palindromic 0.31 · sign_markov_eprod 0.22) — none breach.

▶ WHAT THE LOOP IS DOING NOW

No new decision ⇒ dry counter 1/2. Next (iteration 6): the final independent audit — scale-aware value-health (per-symbol output range, no near-point-mass), parameterless-by-inspection re-confirmation for all 15 survivors, and the intra-batch pooled cand×cand >0.95 check. If no new decision ⇒ dry → 2 ⇒ STOP → Attacker/Defender Challenge-and-Held → PR-from-worktree (no merge).

Iteration 5 · generated 2026-06-06 · pooled from CROSSBATCH logs across all 10 regime slices · 15 KEEP / 5 DROP unchanged.