The hardened probe ran the full 164-cell ξ matrix (14 low-n excluded). Verdict: 14 survivors confirmed independent, candle_shoelace_area and updown_cjs dropped. A second Attacker/Defender pass (10 challenges) hardened the adjudication — and downgraded candle's drop from "absorbed" to a CI-aware combined-evidence call.
independent under asymmetric ξ. Every per-symbol median ξ ≤ 0.29 (≪ the 0.50 absorption cut). The few worst-CELL maxima near 0.50 (sevcik 0.526, local_variation 0.473) are single-cell spikes that do not replicate (per-symbol medians ≤0.15). keepers.json now carries implementation_ready:true for these 14.| candidate | ξ verdict | per-symbol median (E/G/Xag/Xau) | disposition |
|---|---|---|---|
| edge_spread_bps | independent | 0.16 / 0.14 / 0.13 / 0.13 | KEEP |
| sevcik_fd | independent | 0.15 / 0.14 / 0.14 / 0.13 | KEEP |
| anderson_darling_a2 | independent | 0.20 / 0.18 / 0.19 / 0.22 | KEEP |
| hvg_clustering | independent | ≤ 0.23 | KEEP |
| kendall_taub_ret_spread | independent | ≤ 0.24 | KEEP |
| spectral_log_slope | independent | ≤ 0.24 | KEEP |
| lziv_complexity_signs | independent | ≤ 0.26 | KEEP |
| dhvg_indeg_outdeg_kld | independent | ≤ 0.22 | KEEP |
| pietra_hoover_durations | independent | 0.15 / 0.16 / 0.17 / 0.15 | KEEP |
| local_variation_lv | independent | 0.15 / 0.14 / 0.15 / 0.14 | KEEP |
| spread_sign_reversal_rate | independent | ≤ 0.33 | KEEP |
| sign_two_state_entropy_rate | independent | ≤ 0.24 | KEEP |
| mean_cosine_turning_angle | independent | 0.18 / 0.16 / 0.22 / 0.17 | KEEP |
| mann_kendall_z | independent | ≤ 0.29 | KEEP (clique rep) |
| candle_shoelace_area | absorbed_pointest | 0.52 / 0.50 / 0.48 / 0.49 | DROP pending operator |
| updown_cjs | drop_degenerate | 0.13 / 0.13 / 0.14 / 0.14 | DROP (degenerate) |
candle_shoelace_area → DROP (pending operator concurrence). ξ point-estimate is absorbed on EUR (0.517) + GBP (0.502), but the Politis–Romano subsample CI confirms only EURUSD (CI [0.507,0.525] excludes 0.50); GBP's CI [0.488,0.512] brackets 0.50 (a 0.23σ coin-flip). So the machine verdict is absorbed_pointest, not the binding ≥2-symbol absorbed. The DROP rests on three convergent signals: Spearman's persistent ρ=0.72 monotone shadow (strong), EUR-CI-confirmed ξ absorption (modal absorber intra_max_drawdown, 33% of cells — not a single-column shadow), and paper-id ④ PENDING. Flagged for your sign-off.updown_cjs → DROP (degenerate, settled). Not absorbed (ξ≈0.13) but degenerate on all 4 symbols (CJS divergence ~1e-6; even XAG's 7×-larger range of 1.4e-5 is 4 orders below a meaningful divergence). The metals-only-keep hypothesis is rejected — the up/down distributions are statistically indistinguishable on every symbol. Machine verdict drop_degenerate.semivar_signed_ratio, medcouple_skewness, foster_stuart_records_d, turning_point_rate_z were Spearman-DROPs never ξ-checked. A targeted directed-ξ run (those 4 + mann_kendall_z) is finishing to confirm the collapse under the non-monotone metric (CONFIRM iff member→mann_kendall ξ>0.50 on ≥2 symbols). Result lands in PR 2.ATTACKER attacked the verdicts; conceded gaps were fixed; DEFENDER adjudicated against the re-aggregated verdict. The review materially improved the call — it caught that candle's GBP "replication" was a coin-flip and that its named absorber was a 3-cell artifact.
| # | challenge | verdict |
|---|---|---|
| 1 | CRIT: machine never emitted DROP for updown_cjs (flag only) | FIXED drop_degenerate emitted |
| 2 | candle GBP 0.502 = 0.23σ from 0.50; replication = EUR + coin-flip | FIXED PR-CI → EUR-only confirmed |
| 4 | absorber misattributed (lee_ready_imbalance = 3 cells) | FIXED modal = intra_max_drawdown 33% |
| 5 | updown 1e-4 floor arbitrary; metals untested | FIXED degenerate all 4 incl. XAG; DROP holds |
| 8 | ≥2-symbol replication unweighted by N | FIXED CI widths track 1/√n |
| 6,7,10 | KEEPs clean / Holm contradiction / forward direction | HELD with evidence |
| 3 | is candle DROP justified on EUR+Spearman alone? | operator combined-evidence DROP, sign-off |
| 9 | 4 clique members dropped on Spearman only | confirming directed-ξ run in progress |
keepers.json already has implementation_ready:true for the 14.Generated 2026-06-06 · iteration 3 · 164-cell hardened ξ run · verdict doc 2026-06-06-xi-final-verdict-and-challenge-held.md