Navigation

odb.epatner.com · 2026-06-06 · pane B (loop-batch34)

Forex Orthogonal Candidate Discovery — batch 3 + 4

20 newly-shortlisted parameterless candidates (batch 3 + batch 4) registered in the multi-slice probe and run through the same Spearman multi-slice orthogonality screen as batches 1-2 — 174 cell-slices (4 assets × up to 7 ODB thresholds × up to 10 regime slices) + cross-batch vs the 14 batch-1-2 keepers. Real bars only · read-only ClickHouse · no synthetic series.

15
KEEP (4 DROP · 1 HOLD)
10/10
regime slices cross-batch-covered
9 / 3
Challenge HELD / CONCEDED (0 flips)
2 / 2
loop-until-dry → STOP

■ FINAL — loop dry (K=2), STOP

Iteration 6 complete. Final audit (value-health · parameterless · intra-batch among survivors) found no new decision ⇒ second consecutive dry iteration ⇒ STOP. Adversarial Attacker/Defender review: 9 HELD, 3 CONCEDED→FIXED, 0 verdict flips (guzik WATCH · single pooled denominator · ohlc DROP→HOLD). Locked: 15 KEEP · 4 DROP · 1 HOLD. PR-from-worktree opened for review (not merged).

Iteration log: iter 1 → · iter 2 → · iter 3 → · iter 4 → · iter 5 → · iter 6 — FINAL · Challenge-and-Held →

Live dispositions — the 20 candidates

vs-prod = |ρ| across all 174 cell-slices vs the 143 shipped columns; vs-keeper = |ρ| vs the 14 batch-1-2 keepers (4 assets × 3 slices). Breach = assets where max|ρ|>0.85. Legend: KEEP · KEEP·REP family representative · KEEP·WATCH flagged for ξ gate · DROP.

#candidatefamilysubprod medprod maxprod breachworst keeper (max)keeper breachdisposition
25pomeau_irreversibility_lag1time-irrevlogret0.130.39mann_kendall_z (0.23)KEEP
28guzik_increment_magnitude_asymmetrytime-irrevlogret0.230.63mann_kendall_z (0.84)1bp underKEEP·REP·WATCH
67categorical_rqa_determinism_logret_signcat-RQAsigns0.130.47sign_two_state_entropy (0.94)XAG·XAUDROP·collapse
70categorical_rqa_diagonal_entropy_logret_signcat-RQAsigns0.130.45sign_two_state_entropy (0.96)XAU (1)KEEP·REP·WATCH
118critical_exponent_sign_wordwordssigns0.180.33hvg_clustering (0.49)KEEP
119palindromic_defect_sign_wordwordssigns0.130.33sevcik_fd (0.27)KEEP
96sign_markov_entropy_production_3statesign-Markovsigns0.070.26spectral_log_slope (0.22)KEEP
97sign_markov_slem_mixing_3statesign-Markovsigns0.170.65sevcik_fd (0.66)KEEP
72psd_wiener_spectral_flatnessspectralclose0.130.51spectral_log_slope (0.58)KEEP·REP
74psd_spectral_centroid_normfreqspectralclose0.140.58sevcik_fd (0.69)DROP·collapse
57hvg_mean_shortest_path_lengthvis-graphclose0.170.54sevcik_fd (0.72)KEEP
55hvg_degree_assortativity_coefficientvis-graphclose0.170.30hvg_clustering (0.58)KEEP
63hoeffding_d_midreturn_durationcopula-freeret_dur★0.180.49anderson_darling_a2 (0.42)KEEP
44ohlc_ordinal_pattern_indexcandle ordinalohlc_bar0.850.94EUR·GBP·XAG·XAU (105c)anderson_darling_a2 (0.04)HOLD·location
86interquote_count_fano_factordur count-dispqcount★0.520.82pietra_hoover_dur (0.65)KEEP·WATCH
82sign_negentropy_logcosh_gapnegentropylogret0.320.62anderson_darling_a2 (0.90)EUR·XAG·XAUDROP·connector
38cumulative_residual_extropyinformationlogret0.270.61mean_cosine_turning (0.80)KEEP
07right_quantile_weight_tailweightdist taillogret0.160.56anderson_darling_a2 (0.40)KEEP
102left_medcouple_lower_tailweightdist taillogret0.170.67mann_kendall_z (0.60)KEEP
27porta_increment_sign_asymmetrytime-irrevlogret0.250.62mann_kendall_z (0.87)EUR·GBP·XAUDROP·collapse

The 5 not-shortlisted (4 DROP + 1 HOLD)

droppedwhyreplicated oncollapses to
psd_spectral_centroid_normfreqintra-batch BAN, pair med 0.95 / max 0.98all 4 assets >0.95psd_wiener_spectral_flatness
porta_increment_sign_asymmetry>0.85 vs guzik AND vs keeper mann_kendall_zguzik all 4 · mann_kendall 3guzik (richer, magnitude-weighted)
categorical_rqa_determinism_logret_signcross-batch dup of keeper, worst 0.94XAG·XAU vs sign_two_state…diagonal_entropy
sign_negentropy_logcosh_gapcross-batch dup of keeper anderson_darling_a2 (iter 4)EUR·XAG·XAU (pooled)— (CRExtropy is info-family rep)
ohlc_ordinal_pattern_index HOLDdiffuse shadow of order-flow/directional subspace, med 0.85, never >0.95all 4 assets, 105/174 cells— (held out; relabeled DROP→HOLD in review)

Emergent families grid (15 survivors)

⏳ Time-irreversibility

25 pomeau · 28 guzik·REP

porta collapsed in (med 0.78 all 4 assets).

♻️ Categorical-RQA

70 diagonal-entropy·REP

determinism dropped (0.94 vs sign_two_state, XAG·XAU).

🔤 Combinatorics-on-words

118 critical-exp · 119 palindromic-defect

cleanest: max 0.33 vs prod.

🔀 Sign-Markov

96 entropy-prod · 97 SLEM

96 most-independent of all 20 (med 0.07).

📡 Spectral

72 wiener-flatness·REP

centroid collapsed in (med 0.95 all 4).

🕸️ Visibility-graph

57 mean-path · 55 assortativity

<0.85 vs hvg_clustering/dhvg keepers.

🔗 Copula-free non-monotone

63 Hoeffding-D (ret_dur★)

return↔duration coupling; new substrate.

⏱️ Duration count-dispersion

86 Fano (qcount★)·WATCH

med 0.52 vs intensity; 0 cells >0.85.

🔢 Information

38 CRExtropy

82 negentropy dropped iter 4 (dup of anderson, ≥2 assets).

📐 Distributional tail

07 right-quantile-weight · 102 left-medcouple

robust parameter-free tails.

How a disposition is decided (the DISTRIBUTION rule)

No single |ρ| cutoff (this is about the decision threshold, not the ODB price threshold — every candidate is measured across all 4 assets × up to 7 ODB thresholds × up to 10 regime slices = 174 cell-slices). Across that whole grid: (1) median location of |ρ| vs the nearest shipped column; (2) max tail |ρ|; (3) breach-mass — #cell-slices >0.85 / >0.95; (4) ≥2-asset replication — a breach on ≥2 of the 4 assets is real, not a single-cell artifact; (5) scale-aware value-health — per-symbol output range, so orthogonal-but-informationless (near point-mass) ⇒ DROP. Persistent high median ⇒ DROP/collapse to a representative; >0.95 tail replicated on ≥2 assets ⇒ DROP the connector; else KEEP.

assetODB thresholds (bps)regime slicescell-slices
EURUSD1·2·3·5·10·25·50all 1070
GBPUSD5·10·25·50all 1040
XAGUSD5·10·25·50all 1040
XAUUSD5·10·25·506 (gold history shorter)24
total grid174

Slices: covid_crash · lowvol_grind · eur_parity · bank_stress · gold_rally · midcycle_range · gold_parabola · live_tail · asian_session · london_ny. All 20 pass the scale-aware value-health gate (min per-symbol range ≫ 1e-6; none point-mass).

Iterations

Pane B (loop-batch34) · isolated from pane A (loop-xi-keeper) · HTML outside the repo at ~/sites · artifacts in worktree batch34_results/ · updated 2026-06-06.