โ€บNavigation
ERRATUM & UPDATE (2026-06-24, operator-directed) โ€” findings below are unchanged; this corrects framing only.
1. Scope decision: the campaign is crypto-only "for now" (original operator directive 2026-05-30, re-confirmed 2026-06-24; provisional and reversible per pre-spec.md v2). Forex is temporarily out of scope, so this question is now RESOLVED-AS-SCOPED, not a live blocker. The forex ICP wiring defect described below is logged for if/when forex is reintroduced. See Iteration 3.
2. ClickHouse wording fix: this loop does have read-only ClickHouse access (--readonly=2). This page's original wording implied the firing had no ClickHouse access; that was imprecise and has been corrected in place (so you will no longer find that phrasing below). It only ever meant no query was needed this firing โ€” the answer was read from a committed artifact that an earlier read-only ClickHouse run produced.

Iteration 1 โ€” Q1 (forex instance) CURRENT STATUS: RESOLVED-AS-SCOPED was: BLOCKED โ€” wiring defect

Live status: RESOLVED-AS-SCOPED โ€” forex is out of scope (operator directive, crypto-only). The "BLOCKED" analysis below was accurate at the time and is retained for the record; it is superseded by the erratum and by Iteration 3.

Read-only ClickHouse access is present (--readonly=2); no query was required this firing โ€” the answer was read from a committed artifact ยท matrix-evaluation-for-orthogonality loop ยท PR #543

Question (backlog Q1)

Can forex features be graded on regime-invariance at all? The grounding metric is the ICP environment count (target 10 environments) followed by a per-feature pass/fail under Type-I-controlled ICP (ฮฑ = 0.05).

Method (read-only, zero new compute)

This question is answered by characterizing a committed artifact, not by running a new measurement โ€” so no ClickHouse query and no leakage_guard.py run are warranted (R7 applies to new measurements only). The artifact is the committed cross-regime ICP map:

findings/evolution/audits/2026-05-26-forward-orthogonality-prediction/icp_v3_cross_regime.json (2330 bytes, committed on the PR branch).

The file holds exactly two cells, keyed <substrate>|<symbol>|<threshold_dbps>:

Cell keyPopulated feature slotsPer-feature ICP valueInterpretation
crypto|BTCUSDT|25092 slots here (this icp_v3 artifact; full_grid_aggregate.json counts the deduplicated 91-feature panel)every slot = 0.00/92 (same null finding as 0/91 on the dedup panel) features regime-invariant โ†’ STABLE-empty replicated for crypto
forex|EURUSD|50 โ€” the cell is the empty object {}noneno ICP ever computed for forex โ€” empty because the forex active-set columns are crypto-named (harness wiring bug), NOT a forex result

Findings โ€” every number with its trust basis

NumberValueTrust basisStatus
Forex ICP environment count0 (target 10)Direct structural artifact read โ€” a ground-truth provenance fact, NOT a statistical measurement. No null, no error-bound applies because nothing was measured.DEFECT
Forex per-feature pass/failundefined โ€” cell is {}N/A โ€” cannot run Type-I-controlled ICP on an absent environment set.UNAVAILABLE
Crypto ICP invariance0 / 92 (same null finding as 0/91 on the dedup panel)Type-I-controlled ICP (ฮฑ = 0.05); committed + replicated in SPIKE-2026-06-16-RECONFIRMATION (ICP 0/92, leakage gate PASS). The 92-slot count is this icp_v3 artifact / SPIKE; full_grid_aggregate.json reports the same finding as 0/91 on a deduplicated 91-feature panel.GROUNDED

Verdict (superseded โ€” see current status above)

Current status: RESOLVED-AS-SCOPED. Under the operator's crypto-only directive, forex is out of scope, so there is no live blocker. The verdict immediately below is the original reading, retained for the record.

[Original, now superseded] Forex features cannot yet be graded on regime-invariance. The grading was BLOCKED by a confirmed wiring defect: the forex cell in icp_v3_cross_regime.json is an empty object {} โ€” 0 of the target 10 environments are populated, so no per-feature Type-I-controlled ICP pass/fail exists.

Anti-inference guard (R7). The empty forex cell must not be read as "all forex features are fragile" or "forex orthogonality failed." Absence-of-measurement is not a measurement. The only defensible reading is: the forex ICP environment loop is not wired, mirroring the open defect already logged in E4-FINAL-VERDICT.md ("forex ICP wiring (0 envs)") and FULL-GRID-SCALE-VERDICT.md ("forex degeneracy = harness defect: crypto-named active set").

By contrast the crypto cell is fully populated and is uniformly 0/92 (the same null finding as 0/91 on the dedup panel) โ€” a real, leakage-gated, replicated result consistent with the campaign's grounded "STABLE-ORTHOGONAL empirically empty" finding. So the instrument works; only the forex side is unwired.

What this implies / what's next

Read-only & resource-cap proof

GuardEvidence this firing
ClickHouse READ-ONLY (R1)Read-only access present (--readonly=2); no query was required this firing โ€” answer characterized from a committed artifact (0 queries, 0 writes).
Watchdog precheck (R2)load1 = 4.95 โ‰ค 24 at evaluation time (no heavy compute launched).
Resource cap (R2)No compute envelope needed โ€” read of a 2330-byte committed JSON; peak RSS negligible.
Leakage discipline (R7)No new measurement produced โ†’ leakage_guard.py not required. Crypto 0/92 inherits its committed, replicated leakage-PASS basis.
Write surfaces (R3)Only this dashboard folder + the PR #543 body. No source/doc edits; defect handed to operator, not patched.

SSoT: icp_v3_cross_regime.json ยท cross-refs E4-FINAL-VERDICT.md, FULL-GRID-SCALE-VERDICT.md, STATUS-GROUNDING-AND-THRESHOLDS.md ยท audit twin findings/evolution/audits/2026-05-26-forward-orthogonality-prediction/ ยท read-only evaluator ยท do NOT merge PR #543