2026-06-06 · forex orthogonal candidate discovery · batch 3 + 4 · pane B (loop-batch34) · iteration 4
The final cross-batch sweep on eur_parity · bank_stress · gold_parabola · asian_session completes new×keeper coverage of all 10 regime slices — and it surfaced a genuine new decision the live/gold/session grid had under-sampled: sign_negentropy_logcosh_gap is a cross-batch duplicate of the keeper anderson_darling_a2. Updated verdict: 15 KEEP / 5 DROP. Dry counter resets to 0.
Both sign_negentropy_logcosh_gap (Hyvärinen logcosh distance-from-Gaussian) and the keeper anderson_darling_a2 are non-Gaussianity / distributional-shape statistics on the same logret substrate. Pooled across all 10 slices, negentropy breaches >0.85 vs anderson on EUR · XAG · XAU (≥2 assets, worst 0.90) — a replicated cross-batch duplicate ⇒ DROP. Its intra-batch partner cumulative_residual_extropy (38) survives as the negentropy/information representative (worst keeper 0.77, no breach).
| grid (slices) | negentropy vs anderson — breach assets >0.85 | worst |ρ| |
|---|---|---|
| iter 2 — live_tail · gold_rally · london_ny | EUR (1) | 0.87 |
| iter 3 — covid · lowvol · midcycle | EUR (1) | 0.90 |
| iter 4 — eur_parity · bank_stress · gold_parabola · asian | EUR · XAG · XAU (3) | 0.88 |
| pooled union (all 10 slices) | EUR · XAG · XAU → ≥2 assets | 0.90 |
Why it matters: the live/gold/session grid alone showed only EUR (1 asset) — under the ≥2-asset bar. The parity/stress/parabola regimes exposed the XAG+XAU replication. This is precisely what loop-until-dry guards against; the iteration was not a no-op.
New decision ⇒ dry counter reset to 0/2. Cross-batch coverage is now complete (all 10 slices). Next: iter 5 — pooled-union re-verification across all 10 slices (does any OTHER candidate reach ≥2-asset when unioned?); iter 6 — final value-health + parameterless re-audit. Two consecutive no-new-decision iterations ⇒ dry → 2 ⇒ STOP → Attacker/Defender Challenge-and-Held → PR-from-worktree (no merge).
| # | candidate | family | disposition | basis |
|---|---|---|---|---|
| 96 | sign_markov_entropy_production_3state | sign-Markov | KEEP | most independent (prod med 0.07) |
| 25 | pomeau_irreversibility_lag1 | time-irrev | KEEP | prod med 0.13, keeper ≤0.42 |
| 119 | palindromic_defect_sign_word | words | KEEP | keeper ≤0.31 |
| 118 | critical_exponent_sign_word | words | KEEP | keeper ≤0.49 |
| 97 | sign_markov_slem_mixing_3state | sign-Markov | KEEP | keeper ≤0.66 |
| 55 | hvg_degree_assortativity | vis-graph | KEEP | prod max 0.30; keeper ≤0.58 |
| 57 | hvg_mean_shortest_path_length | vis-graph | KEEP | keeper ≤0.72 |
| 63 | hoeffding_d_midreturn_duration | copula-free | KEEP | ret_dur★; keeper ≤0.42 |
| 07 | right_quantile_weight_tailweight | dist tail | KEEP | keeper ≤0.57 |
| 102 | left_medcouple_lower_tailweight | dist tail | KEEP | keeper ≤0.60 |
| 38 | cumulative_residual_extropy | information | KEEP | now sole info-family survivor; keeper ≤0.80 |
| 86 | interquote_count_fano_factor | dur count-disp | KEEP·WATCH | qcount★; prod med 0.52, 0 cells >0.85 |
| 28 | guzik_increment_magnitude | time-irrev | KEEP·REP | absorbs porta; keeper ≤0.84 |
| 72 | psd_wiener_spectral_flatness | spectral | KEEP·REP | absorbs centroid; keeper ≤0.71 |
| 70 | categorical_rqa_diagonal_entropy | cat-RQA | KEEP·REP·WATCH | absorbs determinism; XAU-only 0.96 vs sign_two_state (1 asset) |
| DROP (5) | ||||
| 74 | psd_spectral_centroid_normfreq | spectral | DROP·collapse | 0.95 med vs flatness, all 4 assets |
| 27 | porta_increment_sign_asymmetry | time-irrev | DROP·collapse | vs guzik all 4 + vs mann_kendall ≥2 assets |
| 67 | categorical_rqa_determinism | cat-RQA | DROP·collapse | vs sign_two_state ≥2 assets (worst 0.98) |
| 82 | sign_negentropy_logcosh_gap | negentropy | DROP·connector | NEW (iter 4): vs anderson_darling_a2 on EUR·XAG·XAU |
| 44 | ohlc_ordinal_pattern_index | candle ordinal | DROP·location | prod med 0.85, all 4 assets, 105/174 cells |
Iteration 4 · generated 2026-06-06 · cross-batch coverage complete (10/10 slices). Artifacts: CROSSBATCH_VS_KEEPERS_iter{3,4}_*.json · updated batch34_dispositions.json (15 KEEP / 5 DROP). Next: iteration 5 — pooled-union re-verification.