Bruntwork · Eon Labs · bigblack · driver: native Claude Code /loop (20-min) · 2026-06-07

Forex Orthogonal-Feature → Production Implementation

Implement the 14 ξ-confirmed parameterless-orthogonal forex candidates as REAL production feature columns in fxview_cache.forex_bars, by copying the engineering of the crypto bar_close #509 promotion — not the crypto. ONE PR per feature; every PR human-merged.

▶ CAMPAIGN COMPLETE (2026-06-09, laptop run) — 9 of 9 implementable features shipped + bootstrap. Feature #9 (FINAL) spectral_log_slope SHIPPED → PR #99 (#1 anderson = #91, #2 edge = #92, #3 hvg = #93, #4 kendall = #94, #5 lziv = #95, #6 mann_kendall = #96, #7 sevcik = #97, #8 sign_entropy = #98). First spectral / log-log-PSD feature. Chain bit-faithful: SSoT oracle (discovery-probe k_spectral_slope over the RAW "close" 200-bar trailing window: mean-center → one-sided real-DFT power → drop DC → OLS slope of log(P) vs log(f) over surviving positive-power bins ≥8; slope verbatim NOT negated) → Rust kernel (direct DFT std cos/sin mod-n twiddle + closed-form OLS; no new dep) → producer column → populate (200-bar close ring, no gap-null like sevcik) → batch≡streaming parity (11200 windows, max|dev| 4.641e-14 ≪ 1e-6) → 3-way FOSS (probe-form vs clean-room hand-DFT+hand-OLS vs scipy.fft.rfft+scipy.stats.linregress: 0 mismatches, 0 mask divergences, max|dev| 2.7e-14 / 1.4e-14 over 11200) → feature_metadata. Load-bearing finding: the probe's (P>0) mask drops numpy pocketfft exact-zero bins (even-n Nyquist) that a direct DFT leaves at ~1e-60 → one bin at log(P)≈−138 caused a 1.4-slope divergence; fixed with a relative numerical-zero floor P>ε²·max_P (machine-defined ε², NOT a tuned knob, NOT a frequency-band restriction → parameterless preserved), reproducing numpy's bin set on all 11200 windows. FOSS trap: card lists librosa poly_features but librosa fits magnitude-over-LINEAR-freq, NOT log-log PSD slope — scipy (FFT+linregress, BSD-3) is the correct discrete ref. check:all-gates green on macOS except (a) worktree git-hook-installed + (b) the STANDING foreign research-orthogonality substrate-skip (operator-follow-up; see handoff). §11 STOP reached: all 9 implementable PRs + bootstrap opened; 5 deferred NOT attempted. Bootstrap #90 + feature PRs await operator merge (order #90 → #91 → #92 → #93 → #94 → #95 → #96 → #97 → #98 → #99).

10 / 10

PRs (bootstrap + 9) ✓

9 / 14

implementable now

deferred (3 substrate + 2 param)

12 / 12

bootstrap tests pass

⛔ Operator Morning Handoff

(1) Current state. Bootstrap PR #90 is OPEN (branch feat/odb-ext/bootstrap, rebased onto 967a0d8 after hygiene PR #89 merged), awaiting review + squash-merge. Scaffold + the Challenge-and-Held F1/F2/F4a fixes applied. 8/8 SOTA-harness + 4/4 cohort anchor-pin tests pass; rust:check + rust-bash-line-ref-bounds green; test_schema_contract unaffected (no producer wire-format / schema.sql change by design). F3 caveat: the fail-loud anchors are macOS-verified but UNVERIFIED on Linux/bigblack pending the pre-existing signal_ring test-build breakage (out of campaign scope) — see the PR body + BOOTSTRAP_STATUS.md.

(2) Follow-up gaps + distance to closed.

MERGED · PR #89 Repo-wide gate blocker cleared: the foreign check:rust-bash-line-ref-bounds drift in fxview-stall-detector-watchdog (not this campaign) was fixed in standalone PR #89 and merged (origin/main → 967a0d8); the gate is green again.
bigblack-only · ignored on laptop origin/main Linux test-build is broken (pre-existing, out of campaign scope; ignored on the laptop run per override §8). cargo test -p fxview-core fails with 40 errors across 6 #[cfg(target_os="linux")] signal_ring/shm test modules — a dropped use import (CanonicalSignalRingEntryTotalTwoCacheLines) + a missing constructor arg from the Wave-A/275/396 test-modularization refactors. Invisible on the macOS dev/CI platform; blocks mise run check:all-gates on bigblack, so NO PR (bootstrap or per-feature) can pass gates here until fixed. Minimal fix: add the missing use to the 3 Linux-gated test files + supply the producer_started_unix_microseconds: u64 arg. Owned by the SHM/signal-ring loop, not this campaign (the loop did not touch it — "never guess / never patch in-flight code it doesn't own").
DESIGN DECISION · 5 deferred 9 of 14 get kernels now. 3 substrate-blocked (local_variation_lv, pietra_hoover_durations, spread_sign_reversal_rate — need per-quote intra-bar data not persisted; decide: intra-bar sidecar feature vs drop). 2 parameter-flagged (dhvg_indeg_outdeg_kld, mean_cosine_turning_angle — NOT parameterless-by-inspection per the 2026-06-09 audit: free knobs; operator pin-or-drop). Full per-feature rationale, decision menus + resurrection conditions: see the DEFERRAL REGISTER callout below · SSoT DEFERRALS.md (PR #100).

DEFERRAL REGISTER · why 5 of 14 were not implemented (recorded 2026-06-12). Canonical SSoT: findings/audits/2026-06-03-forex-parameterless-orthogonal-candidate-discovery/DEFERRALS.md (PR #100); each discovery candidate card now carries an Implementation disposition section and discovery KEEPERS.md carries the 14-keeper outcome table. This callout is a mirror, not the SSoT.

substrate-blocked · 2026-06-07 bootstrap local_variation_lv (card 47) — needs the per-quote inter-arrival intervals τ_i; forex_bars persists only quote_count (=n) + duration_us (=Στ_i): the per-quote dispersion LV measures is irrecoverable from {n, Σd}. Operator: intra-bar sidecar re-route vs drop.
substrate-blocked · 2026-06-07 bootstrap pietra_hoover_durations (card 120) — the Pietra L1 deviation Σ|d_i−d̄| over per-quote durations cannot be reconstructed from {quote_count, duration_us, perbar_fx_quote_max_gap_us} = {n, Σd, max d}. Operator: intra-bar sidecar re-route vs drop.
substrate-blocked · 2026-06-07 bootstrap spread_sign_reversal_rate (card 111) — needs the ordered per-tick spread series s_t=ask_t−bid_t; only per-bar spread moments are persisted, and the ordering is irrecoverable from any moment set. Operator: intra-bar sidecar re-route vs drop.
parameter-flagged · 2026-06-09 re-audit dhvg_indeg_outdeg_kld (card 30) — unpinned port-level conventions = de-facto free knobs: KLD orientation (the card defines D(P_in‖P_out), Lacasa 2012 defines D(P_out‖P_in)), KL argument order, zero-bin smoothing/ε policy (KLD undefined where one distribution has P(k)=0). Operator: pin-or-drop — pin each knob in the Python-SSoT pinning table, then the feature re-qualifies for the standard per-feature PR pipeline.
parameter-flagged · 2026-06-09 re-audit mean_cosine_turning_angle (card 17) — unpinned conventions: dt (real-µs close_time_us deltas vs uniform spacing), normalization (Re(mean resultant vector) vs modulus), all-flat disposition (+1 vs NULL), gap-break policy across the >8h weekend/Ouroboros reset; plus reference impl trajr is GPL-3 → clean-room mandatory. Operator: pin-or-drop.

Resurrection conditions: substrate-blocked → a persisted intra-bar substrate lands via an operator-approved sidecar + schema + rebuild campaign; parameter-flagged → an operator-recorded pin for every listed knob (+ clean-room attestation for card 17). The compile-time cohort anchor (exactly_eleven_implementable_three_blocked) encodes the substrate axis only (11 implementable + 3 blocked); the 2026-06-09 parameterless axis explains 11→9.

(3) SOTA robustness — applied vs deferred.

Applied (bootstrap): sha-pinned REAL fixture (replay = the test) + fail-closed count-validated loader; cohort SHA256 anchor + Rust↔JSON drift gate (Lesson #9 fail-loud); tiered EXACT-vs-ε tolerance helpers; causal de-gap window-reset guard (the "exceed crypto" property).
Deferred to per-feature PRs: per-feature pinned-SSoT oracle runners + 3-way FOSS triangulation (uv-hermetic, Python 3.14); golden snapshot (insta); batch≡streaming parity; the kernels themselves.
Errata caught: sevcik_fd FOSS ref is neurokit2 not antropy; anderson_darling_a2 uses inter-bar close logret (not the card's per-tick prose). GPL clean-room mandatory: mean_cosine_turning_angle (trajr), mann_kendall_z (cran/Kendall).

(4) Hygiene debt (foreign, pre-existing) — operator follow-up. Full ledger: findings/audits/2026-06-07-…/HYGIENE_DEBT.md (on PR #90).

FIXED · PR #89 (merged) rust-bash-line-ref-bounds — 7 blank-line refs in fxview-stall-detector-watchdog (foreign Audit Class C). Deeper follow-up: fuller Class-C re-audit of remaining stale-but-non-blank refs — rust-enhancement loop (iter#129/#154/#155).
OPEN · Terry-owned signal_ring/shm Linux test-build break (41 errors, Phase-23.E, blocklist, latent since 2026-05-18). NOT fixed (blocklist; dodged on macOS). Follow-up: add the missing use + ctor arg to the 3 Linux-gated test files so the campaign's anchors run on bigblack (the F3 caveat).
NEW (iter 03) · rust-enhancement loop check:research-orthogonality-null-utils-tests — 33 of 34 (foreign env-drift). The 34th is a smoke test that SKIPs when /tmp/mql5-orthogonality-substrate/eurusd_5dbps_5k_fp_candidates.parquet is absent (env artifact, not in-repo, regenerable only with ClickHouse); the gate's MIN_TESTS=34 counts only passed → trips. Owned by the rust-enhancement loop (iter#59 gate); this campaign never touches scripts/research/. Per §10 = operator-follow-up; NOT weakened. Follow-up: count passed+skipped for the substrate-gated test, OR commit a tiny substrate fixture, OR pin the substrate path.

14-Feature Ledger

#	Feature	Card	Status	Tier	SSoT / note
—	PR #1 bootstrap	—	PR #90 OPEN	—	#90 · fixture + harness + cohort anchor + C&H F1/F2/F4a (598d1d8 + bf907d2)
1	anderson_darling_a2	110	PR #91 OPEN	ε 1e-9	scipy.stats.anderson (RAW, ddof=1) · parity max\|dev\| 4.5e-13 (11200 windows) + batch≡streaming + 3-way FOSS
2	dhvg_indeg_outdeg_kld	30	deferred · param-flagged	EXACT+ε	NOT parameterless (KLD orientation + arg + zero-bin ε) — pin-or-drop
3	edge_spread_bps	52	PR #92 OPEN	ε 1e-9	bidask.edge · parity max\|dev\| 9.78e-11 (11200) + batch≡streaming + 3-way FOSS (edge vs edge_rolling 3.54e-14)
4	hvg_clustering	54	PR #93 OPEN	EXACT+ε	ts2vg HVG + networkx.average_clustering · parity max\|dev\| 9.99e-16 (11200) + batch≡streaming + 3-way FOSS (0 edge-set mismatches vs clean-room numpy HVG)
5	kendall_taub_ret_spread	79	PR #94 OPEN	EXACT+ε	scipy.kendalltau(variant='b') · integer-exact concordance counts · parity max\|dev\| 2.78e-17 (11200) + batch≡streaming + 3-way FOSS (0 mismatches / max\|dev\| 0.0 vs clean-room numpy tau-b)
6	local_variation_lv	47	substrate-blocked	ε	needs per-quote intervals (not persisted)
7	lziv_complexity_signs	40	PR #95 OPEN	EXACT+ε	antropy.lziv_complexity · EXACT integer LZ76 count · parity max\|dev\| 2.22e-16 (11200) + batch≡streaming + 3-way FOSS (0 mismatches / max\|dev\| 0.0 vs clean-room substring-LZ76)
8	mann_kendall_z	20	PR #96 OPEN	EXACT+ε	probe k_mann_kendall · integer-exact S, SIMPLE var (NO tie correction) · parity max\|dev\| 3.553e-15 (11200) + batch≡streaming + 3-way FOSS (0 S-mismatches vs clean-room + pymannkendall integer S; tie-pervasive substrate → no-tie-correction is load-bearing)
9	mean_cosine_turning_angle	17	deferred · param-flagged	ε 1e-9	NOT parameterless (dt-convention + normalization + all-flat) — pin-or-drop · trajr GPL clean-room
10	pietra_hoover_durations	120	substrate-blocked	ε	needs per-quote durations (not persisted)
11	sevcik_fd	10	PR #97 OPEN	ε 1e-9	probe k_sevcik · unit-square curve-length FD · parity max\|dev\| 6.661e-16 (11200) + batch≡streaming + 3-way FOSS (0 mismatches vs clean-room + neurokit2.fractal_sevcik; errata: antropy has NO Sevcik FD, real ref is neurokit2)
12	sign_two_state_entropy_rate	83	PR #98 OPEN	EXACT+ε	probe k_sign_entropy_rate · integer-exact 2×2 transition counts · 2-state Markov entropy rate H=−Σ statp·P·log2 P · parity max\|dev\| 1.110e-16 (11200) + batch≡streaming + 3-way FOSS (0 mismatches vs clean-room conditional-entropy + pyitlib H(X\|X₋₁) ML; FOSS trap: card lists NPEET but NPEET is kNN-continuous — wrong tool; pyitlib is the discrete ref)
13	spectral_log_slope	75	PR #99 OPEN	ε ~1e-6	probe k_spectral_slope · full-band log-spectral slope (1/f^α): mean-center → direct real DFT → drop DC → OLS log(P) vs log(f) · parity max\|dev\| 4.641e-14 (11200) + batch≡streaming + 3-way FOSS (0 mismatches vs clean-room hand-DFT + scipy.fft/linregress; P>0 mask reconciled by ε²·max_P numerical-zero floor; FOSS trap: card lists librosa but librosa fits magnitude-over-linear-freq — scipy is the correct log-log ref)
14	spread_sign_reversal_rate	111	substrate-blocked	EXACT+ε	needs per-quote spread seq (not persisted)

Iterations

Presentation layer only — never the SSoT. Repo SSoT: findings/audits/2026-06-07-forex-orthogonal-feature-implementation/ on branch feat/odb-ext/bootstrap (commit a61cc58).