Asked six leftover questions about the forward-orthogonality work and required every answer to carry its own trust basis. Came back with four answered and two answerable only after builds the read-only evaluator was not allowed to make.
| Quantity | Value | Meaning |
|---|---|---|
| Backlog questions reported | 6 of 6 (4 resolved, 2 partial, 0 pending) | evaluator pass complete |
| Regime-invariant crypto features | 0 of 91 (and 0 of 92 on the second panel definition) | same finding, two panel counts, not two runs |
| Certified conditional set | 4 of 91 at q=0.10; PBO 0.00; deflation z 22.15 vs 1.53 | the only positively-certified subset |
| Unclassified remainder | 87 of 91 | blocked on a missing per-environment measurement |
| Live grounding | 871,376 real BTCUSDT at 250 dbps bars through 2026-06-24; 3 SELECTs, capped 5 cores / 3 GiB | read-only |
| Redundancy void-band test | median 0.991 over 3,344 feature-cells; bins 0.80->0.95 = 91 / 110 / 198 / 2,141 | density rises monotonically โ no gap at 0.85 |
| Firing | What it did |
|---|---|
| 2 | Found the fragile-vs-conditional split cannot be made: the harness never emitted per-environment pass counts |
| 3 | First live database grounding; the regime-invariance question is answered, and the answer is empty |
| 4 | An empty result here is the expected outcome of a provably hard problem, not a trustworthy no-signal ceiling |
| 5 | A three-layer alarm with a formal false-alarm bound, specified but not built |
| 6 | Both flagged anomalies sit outside any gate |
| 7 | The 0.85 cutoff cannot be derived from the data; it stays an explicit policy hold |
Two operator builds, both outside the evaluator's write surfaces: extend the ICP harness to emit per-feature k-of-10 pass counts so the 87 unclassified features can be split; and install R + nonlinearICP to build the relaxed-ICP harness with a shuffled-null power calibration. Also build the specified monitor. Stale content to fix: the page banner still carries a do-not-merge instruction for pull request 543, which merged on 2026-08-07.
'evaluator_pass: COMPLETE (all 6 reported); remaining closure requires two operator builds, out of read-only evaluator write-surfaces.' questions_pending: 0.
Lifecycle status is process state, not a judgement of the findings. Results are stated as numbers with their uncertainty. Generated between CLOSING-RECONCILIATION markers โ regenerate rather than edit.
A read-only evaluator + reporter for the forward-orthogonality matrix campaign. It mutates no production system: each firing answers one backlog question and attaches a trust basis to every number, then commits the finding to PR #543. Audit twin: findings/evolution/audits/2026-05-26-forward-orthogonality-prediction/.
pre-spec.md's scope-expansion path. The evaluator has read-only ClickHouse access (clickhouse-client --readonly=2). From iteration 3 onward, numbers are grounded in live read-only queries against opendeviationbar_cache.open_deviation_bars on bigblack, shown verbatim; iterations 1โ2 read committed result files produced by earlier read-only runs and ran zero live queries.full_grid_aggregate.json uses a deduplicated 91-feature panel (โ 0/91), while icp_v3_cross_regime.json and the SPIKE-2026-06-16 replication use a 92-slot panel (โ 0/92). Both return zero regime-invariant features. It is the same finding on two panel definitions, not two different runs. aggression_ratio, bar_dispersion_entropy, bar_katz_fd, bar_petrosian_fd); PBO 0.00 crypto; deflation z 22.15 vs 1.53. CONDITIONAL / regime-scoped โ not stable.adv_auc โ 0.522 derived (matched-null). TAU = 0.85 is un-nulled policy (confirmed iter 7); 0.95 BAN ground-truth-anchored.| # | Matrix | Bucket | Citation | Serves |
|---|---|---|---|---|
| 1 | CTM covariate-shift โ WATCH | A โ anytime-valid spine | arXiv:2505.04608 | Q4 live expiry monitor |
| 2 | e-detectors | A โ anytime-valid spine | arXiv:2203.03532 | Q4 false-alarm bound |
| 3 | Nonlinear / relaxed ICP | B โ STABLE-empty falsifier | arXiv:1706.08576 | Q3 (is STABLE-empty a linear artifact?) |
| 4 | Group-invariance CTM | B โ STABLE-empty falsifier | arXiv:2401.15461 | Q3 / Q4 |
| Q | Question | Trust basis | Status | Iteration |
|---|---|---|---|---|
| 1 | Can features be graded on regime-invariance at all? | Type-I-controlled ICP (ฮฑ=0.05) | RESOLVED-AS-SCOPED โ empty result Crypto graded; result is 0/91โ0/92 invariant = none (STABLE-empty), an empty finding, not a positive result. Forex resolved by descoping, not measurement. | iter 1 (superseded) ยท iter 3 |
| 2 | Crypto: split FRAGILE vs CONDITIONAL per feature | per-feature k-of-N ICP (ฮฑ=0.05) | PARTIAL. Of 91: 0 STABLE, exactly 4 certified CONDITIONAL (FDR), 87 unclassified โ per-environment k-of-N is a missing measurement โ operator. | iter 2 |
| 3 | Is STABLE empty, or a strict-linear artifact? | G10 shuffled-null power calibration | PARTIAL. The empty exact set is provably degenerate (NP-hardness, arXiv:2501.17354). Relaxed/ฮต-tolerant invariance undecided: deciding nonlinear ICP (F014, arXiv:1706.08576) specified but unbuilt (no Rscript) โ operator. | iter 4 |
| 4 | Live "orthogonality expired" alarm with a formal false-alarm bound? | Anytime-valid / Ville bound (โคฮฑ at every t) | RESOLVED โ spec delivered (build = operator) Buildable design: WATCH (2505.04608) + e-detectors (2203.03532) + group-invariance CTM (2401.15461), alarm at wealth โฅ 1/ฮฑ, + E6 replay plan. Monitor UNBUILT โ probe stays EVALUATING โ operator builds + E6-validates. | iter 5 |
| 5 | Are the flagged anomalies disqualifying? (wold_R 0.278; forex PBO 0.20) | Replication + PBO/CSCV | RESOLVED โ neither disqualifies wold_R was already reported-not-gated โ its crypto inversion changes no gate; forex PBO 0.20 is a forex watch item, parked (forex out of scope). Crypto FDR certificate untouched. | iter 6 |
| 6 | Can TAU=0.85 be derived, or must it stay policy? | Ground-truth anchor (info-loss-budget) | RESOLVED โ policy, not derived TAU=0.85 fails the void-band test (redundancy median 0.991, no gap) โ policy. Keep the anchored 0.95 BAN; a principled cutoff needs the info-loss-budget derivation โ operator. | iter 7 |
cran/nonlinearICP, build the relaxed-ICP (F014) harness, pair with a G10 shuffled-null power calibration, read the ฮต-frontier. No metric was promoted to a gate (R7); the probe stays EVALUATING (Q4 monitor specified but unbuilt).wold_R reported-not-gated; forex PBO 0.20 parked (out of scope).