Each card opens that campaign, its closing reconciliation and its iteration pages. Chips are lifecycle โ where the work sits in its process, never a judgement of what it found.
Operator ruling, 2026-08-14: this campaign is settled. Recorded as a ruling rather than re-derived from the folder โ the file-based reading that preceded it is kept verbatim in _status_evidence_superseded so the disagreement staysโฆ
FINISHED SHOWCASE. index: 'Audit ยท 2026-05-16 -> 2026-05-20 ยท SHOWCASE PHASE 2 EMPIRICAL โ 6 SPIKES ยท 9 COMMITS ยท BIGBLACK-REPLICATED', 'Symbols replicated 3 / 3 โ BTC + ETH + ADA all IAAFT-validated REAL'. Six named spike pages rโฆ
Campaigns live under campaigns/ โ one directory for all of them, each with its lifecycle status and the evidence behind it.
State-of-the-art metrics for measuring feature orthogonality in financial time-series forecasting. The full catalog spans 3 tiers ร 8 categories, with FOSS implementation status verified per metric. Catalog answers a different question from the 64-candidate sibling audit: this is the tools, that was the things being measured.
The CODEC/FOCI cascade was run through the Selection gates (S0โS5). Chatterjee ฮพ (A1 pairwise) is the only validated sieve โ S1/S2/S3/S5 PASS โ but value-add over the Pearson/Spearman baseline is unproven (S4 deferred: the redundant/orthogonal labels are Pearson-derived, so the test is circular). CODEC (A2) is skipped โ nearly inert on the crypto panel (0.16 flags/cell, T(canary)โ0) and no faithful FOSS null. O-information (A3) is de-scoped to a cheap bias-centered permutation-null watch (the naive |ฮฉ|โฅ0.5 flag manufactures false synergy). Feasible replacements for the conditional/synergy gap: GCM, RCoT (A2) and the bias-centered O-info permutation null (A3) โ see the priority list. Max achievable verdict is Selection-partial until a Pearson-independent S4 label exists; deployment stays ๐ BLOCKED on Terry.
Fills gap: Non-linear redundancy (ฮพ: S1/S2/S3/S5 PASS; value-add pending S4)
Top pick: Chatterjee ฮพโ โ xicorpy (MIT) โ VALIDATED (Selection-partial)
27 metrics ยท 11 permissive FOSS ยท 10 GPL ยท 1 unpackaged
Fills gap: Conditional redundancy (CODEC SKIPPED โ inert on substrate + no faithful FOSS null)
Top pick: GCM / RCoT โ GeneralisedCovarianceMeasure (CRAN) ยท RCIT/RCoT (causal-learn)
11 metrics ยท 2 permissive FOSS ยท 7 GPL
Fills gap: Synergy (higher-order) โ de-scoped to a cheap permutation-null watch
Top pick: O-information + permutation null โ hoi (BSD-3) โ bias-centered shuffle null; naive |ฮฉ|โฅ0.5 flag dropped
6 metrics ยท 3 permissive FOSS ยท 2 GPL ยท 1 unpackaged
Fills gap: Matrix-level orthogonality
Top pick: Marchenko-Pastur + RIE โ pyRMT (BSD-style)
15 metrics ยท 10 permissive FOSS ยท 5 GPL
Fills gap: Causal direction
Top pick: Reduced TE (Kirkley 2025) โ author GitHub (TBD)
15 metrics ยท 6 permissive FOSS ยท 7 GPL ยท 1 unpackaged
Fills gap: Stability across regimes
Top pick: Sliced Wasserstein dependency โ POT (MIT)
4 metrics ยท 1 permissive FOSS ยท 1 GPL ยท 2 unpackaged
Fills gap: FDR aggregation (meta)
Top pick: DeepLINK-T + e-BH โ knockpy + e-BH (MIT + open)
14 metrics ยท 8 permissive FOSS ยท 3 GPL ยท 2 unpackaged
Fills gap: Portfolio-level orthogonality
Top pick: Lรณpez de Prado cMDA + HRP โ mlfinlab / PyPortfolioOpt (mixed)
8 metrics ยท 5 permissive FOSS
One-off documents, kept apart from the 8-category taxonomy above.