Matrix admission paradox loop — standing admission gate

HALTED2026-07-02

Built and sealed a standing lab that decides, one candidate at a time, whether a new statistical instrument earns a place in the trusted measurement stack; across about a hundred sittings it admitted 16 instruments, rejected 45, and then stopped and handed a list of decisions back to the operator.

Why it carries this status

Lifecycle, not result. This says where the audit sits in its process — never whether what it found was good.

**HANDOVER PUBLISHED — LOOP PARKED (operator directive row 112 executed in full).** ... **The loop now runs NOTHING until operator direction** (rule 7 posture: parked, not idle-spinning).

findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/LEDGER.md

2026-08-17. Derived from folder evidence, then adversarially challenged; the challenge pass was upheld.

Blocked on Operator (nasimubd / Nasim). The decision queue in HANDOVER-PROMPT.md must be steered before the loop resumes: the Frontier-#4 review that would unGATE batch B-05 and the ADR's L3 sentinel, ratification of ADR-2026-07-14 (4 open questions) and its hardened successor ADR-2026-07-15 (6 open questions), the tri-tier BAN/WATCH/ORTHOGONAL ratification (LEDGER row 105), the 73 cycle-1 draft statuses, and the PR #599 merge.

The verdict

The audit's own conclusion, reproduced in full from the source below. Not a summary — this is the document, rendered. Links inside it that point at unpublished files are shown as plain text rather than as links that would 404 here.

verdict.md — plain-English bottom line

Status (2026-07-03, after 13 working iterations in one day): THE LAB WORKS AND HAS SHIPPED ITS FIRST VERDICTS.

Seal: all four real-data self-test controls PASS (after the operator's no-synthetic directive forced a redesign mid-seal — the redesign made the controls stronger).
Admissions (2): verdict-stability k-of-N ADMITTED on crypto (row 15) and forex (row 18) — the FRAGILE↔CONDITIONAL boundary is measurable on both substrates for the first time.
Kills (3, all wins): parametric Chow k-of-N (own null caught it); regression-invariance axis (inverted-signal discovery: stable coupling ANTI-predicts orthogonality, ρ=−0.29); F014/ICP-family identification (BLIND-GAUGE — fails a perfect real-data ground-truth case; every "nothing is invariant" it ever produced was blindness).
Repairs (1): forex ICP twin-column defect — 163/167 forex features graded (was 0).
Draft declarations awaiting operator review: crypto — 4 CONDITIONAL (all 10/10, STABLE-flagged); forex — 60 CONDITIONAL + 9 FRAGILE (the campaign's first); FAIL cells were already assignable.
Open instrument gap: STABLE on both substrates — the ICP identification family is blind; an identification-capable framework is the standing SOTA-sweep target (blocked this firing by search availability; re-queued).
Correction chain of "ICP 0/91": not rejection (iter 7) → not evidence at all (iter 12). The ceiling is an instrument question, not a settled fact.

Next actions: operator review of both proposals; SOTA sweep retry; Frontier #4 precondition review; full-grid k_v extension.

source: findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/verdict.md

Still owed 9

Batch B-05 (11 candidates on the EXPIRY boundary) is still GATED on the operator's Frontier-#4 review; until it is unGATEd the expiry design's L3 regime-turn sentinel cannot be built.
Neither live-expiry ADR is ratified: v1 carries 4 blocking questions (§5) and the hardened v2 carries 6 (§9).
The expiry system's E6 replay validation (known-expiring positive control plus known-stable negative control) has never been run.
The tri-tier BAN/WATCH/ORTHOGONAL collapse proposed under operator directive row 105 awaits ratification.
73 cycle-1 draft feature statuses and the STABLE-declaration proposal await operator review.
PR #599 (68 commits) is open and awaits the operator's merge — the loop never merges its own PRs.
Standing pattern reviews left with the operator: the 5-sighting tail-invariance pattern (rows 104–109), the IID-certificate law (rows 99/111/113/115), the dcor×SKIT-null matched-pair proposal, the 10-sighting inverted-mechanism trail, the RW+MCS pipeline-adoption proposal, the H-005 full-window re-proposal, the 03c wild-bootstrap gray zone, the M4-bar question, and the H-062 W-DRO park.
The boneyard counter reconciliation is listed as unresolved in the operator decision queue.
The root verdict.md was never refreshed after 2026-07-03 and understates the campaign by 90 iterations.

What it claims, and what backs each claim 23

Every row pairs a claim with the file it came from and the verbatim text in that file. The sources sit above the deploy root, so the quote is embedded and the path is printed as text rather than linked — a link would resolve on a laptop and 404 here.

Claim	Evidence
The audit is a standing, continuously-running admission procedure that decides per candidate matrix whether it enters the trusted stack, is scoped, or is rejected. ASSERTED	a continuously-running, append-only admission procedure that decides, for every candidate matrix (measurement instrument), whether it is ADMITTED into the trusted stack, EXCLUDED into the boneyard, or INCLUDE-IF (scoped) findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/CLAUDE.md
The rulebook was frozen at its landing commit, and that commit's hash is the pre-registration hash — no threshold, gate, or verdict grammar may move afterwards. CONFIRMED pre-registration hash f959424012d0b54a8984c7fbba00cdd4c2c31615	Status: FROZEN at the commit that lands this file (operator decision 2026-07-03). That commit's SHA IS the pre-registration hash. Every rule below is fixed: thresholds, label, gates, verdict grammar, and the stop predicate may not be moved after seeing any candidate's results. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/M0-MATRIX-ADMISSION-PRE-REGISTRATION.md
The stop condition is deliberately unreachable, so the loop can never finish — its product is the ledger, not a finish line. ASSERTED	New bars and new papers arrive continuously → clauses 2–3 cannot hold simultaneously. The loop is immortal; its value is the ledger, not the finish. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/M0-MATRIX-ADMISSION-PRE-REGISTRATION.md
All four self-test controls passed on real data before any real candidate was evaluated, opening the lab. CONFIRMED C1′ ρ = 1.000000 over 40 cells; C3′ ΔAUC +0.6626, CI95 [+0.4275, +0.8358], n_eval = 256, n_boot = 400; C4 9/9 verdict match	Status: SEAL COMPLETE (2026-07-03, no-synthetic designs per amendment 03c). C1′ PASS (row 3 — exact clone ρ=1.000000 EXCLUDEd at M3) · C2′ PASS (rows 4+6 findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/SELF-TEST-CONTROLS.md
The regression control reproduced the earlier one-shot gate's verdicts exactly, with one candidate dying at an earlier gate — a deviation explained in writing rather than smoothed over. CONFIRMED n = 9 historical candidates; 9/9 verdict match, 8/9 gate match (DSR-proper killed at M0 instead of M3)	7 EXCLUDE / 2 INCLUDE-IF reproduced — 9/9 verdict match. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/C4-REGRESSION-FIXTURE.md
Two mid-flight operator directives were recorded verbatim and folded in as superseding amendments: per-run permission for spike runs was lifted, and all synthetic data was banned. CONFIRMED 2 amendments (03b operator-go lifted, 03c no synthetic data); LEDGER rows 0c and 0d	> Going forward, you are not allowed to use any synthetic data at all. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/M0-AMENDMENT-2026-07-03c-NO-SYNTHETIC-DATA.md
The forex invariance engine had been silently producing an empty result for every feature; the repair took graded forex features from zero to 163 of 167. CONFIRMED 163 of 167 forex features graded (was 0 of 167); 50 constant-in-env skips + 4 features with <3 environments	Forex ICP map non-empty: 163/167 features graded (was 0). findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/FOREX-ICP-WIRING-DEFECT.md
A second, deeper cause of the forex failure was two byte-identical columns feeding the same design matrix, producing a singular-matrix error swallowed by a bare exception handler. CONFIRMED countIf(!=) = 0 over 701,424 rows; LinAlgError on 5 of 5 features in the reproduction probe	on fxview EURUSD@5, `duration_us` is byte-identical to `intra_duration_us` (`countIf(!=) = 0` over 701,424 eur_parity rows; stddev identical in all 7 windows). The v3 anchor set fed both twins into ICP → rank-deficient design matrix → `LinAlgError: Singular matrix` for every feature → swallowed by the bare `except` → silent `{}`. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/FOREX-ICP-WIRING-DEFECT.md
With the instrument repaired, forex mirrors crypto: no feature reads as regime-invariant on either substrate. MEASURED forex 0 of 163 features invariant at α = 0.05; crypto 0 of 92	forex Tier-A mirrors crypto — 0/163 features regime-invariant at α=0.05 (crypto committed: 0/92). The instrument now grades forex instead of being blind findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/FOREX-ICP-WIRING-DEFECT.md
The invariance-identification family was excluded as a blind gauge, which retroactively converts its historical 'nothing is invariant' result from evidence into blindness. REFUTED historical readout 0 of 91 features invariant, now void as evidence	ICP-family identification excluded as BLIND-GAUGE (row 17): fails the committed perfect-duplicate ground-truth case, so all its negatives (incl. the historical "0/91 invariant") are blindness, not evidence. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/STATUS-EVIDENCE-MAP.md
The FRAGILE-versus-CONDITIONAL boundary became measurable for the first time via a verdict-stability counter admitted separately on each substrate. MEASURED crypto M3 −0.753, M5′ ρ = +0.585; forex M4 ΔAUC +0.0318 (CI entirely > 0), hardening +0.016; thresholds crypto k=6/10 (p_base 0.264), forex k=7/8 (p_base 0.513)	k_v ADMIT #1 crypto (row 15: anchor exact 0-of-10 on the committed duplicate; M3 −0.753; M5′ ρ=+0.585 vs ~0 null) · ADMIT #2 forex (row 18: M4 ΔAUC +0.0318 CI entirely > 0; hardened row 22, +0.016 on top of `mred`). findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/STATUS-EVIDENCE-MAP.md
As of the 2026-07-04 consolidation, no admitted instrument could support the STABLE status at all — an openly declared instrument gap. CONFIRMED 0 instruments in the STABLE decision rule; 9 certified gate slots at that date, 4 of them inside status rules	NO ADMITTED INSTRUMENT — open gap. Interim: k_v = N/N earns a STABLE-candidate flag only, never a declaration findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/STATUS-EVIDENCE-MAP.md
That STABLE gap was later closed on both substrates by an admitted e-value invariance instrument. CONFIRMED e-ICP-fp admitted on forex + crypto (LEDGER rows 34/35)	STABLE slot filled both substrates (e-ICP-fp) · FAIL boundary certified beyond G0 (PID-red #15, O-info #16 WATCH). findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/HANDOVER-PROMPT.md
A recurring law emerged from the kills: any monitor whose null assumes independent samples false-alarms wildly on autocorrelated bars, and eight instruments died of exactly this. CONFIRMED 8 instruments killed; false-alarm rates 11–57% against a 5% nominal (worst on record 56.6%)	L-IID — the IID-CERTIFICATE LAW (8 members, LEDGER rows 99/111/113/115). Any monitor whose "how big is chance?" null assumes independent samples false-alarms at 11–57% instead of 5% on autocorrelated bars. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/ADR-2026-07-15-LIVE-EXPIRY-HARDENED.md
An inverted-mechanism pattern was discovered and then repeatedly re-observed: instruments that should predict orthogonality anti-predict it. MEASURED first sighting ρ = −0.29; later distributional meters calgap −0.159, MMD −0.155, W1 −0.143 (CI [−0.178, −0.106])	regression-invariance axis (inverted-signal discovery: stable coupling ANTI-predicts orthogonality, ρ=−0.29) findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/verdict.md
At the park point the campaign had admitted 16 instruments, killed 45, and consumed 57 of the 68 fed candidates, with one batch left gated. MEASURED 16 ADMITs · 45 kills · 60 boneyard rows · 23 certified gate slots · 57 of 68 candidates consumed · 11 candidates still gated · 117 ledger rows	All actionable feed evaluation COMPLETE: B-01/B-02/B-03/B-04/B-06 CLOSED — 57/68 candidates; B-05 (11, EXPIRY boundary) remains GATED on the operator's Frontier-#4 review. Campaign at park: 16 ADMITs · 45 kills · 60 boneyard rows · 23 certified gate slots · PR #599 OPEN (68 commits). findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/LEDGER.md
Certificates are scoped to the exact grids they were earned on and expire on the next completed regime slice; a crypto certificate never transfers to forex. CONFIRMED crypto grid = BTCUSDT@250 × 10 slices; forex grid = EURUSD@5 × 8 windows	Scope: certificates cover the grids they were admitted on — BTCUSDT@250 (10-slice grid) and EURUSD@5 (8-window grid). Other combos have report-only corroboration (row 21), not certification. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/STATUS-EVIDENCE-MAP.md
Nothing currently detects, live, that a regime slice has completed — so issued certificates can go silently stale, which is the failure class the campaign exists to prevent. CONFIRMED 23 certified gate slots exposed to silent staleness	Every ADMIT verdict carries the expiry field `next completed forex regime slice` — but nothing currently detects, live, that a regime slice has completed. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/ADR-2026-07-14-LIVE-EXPIRY.md
Two live-expiry designs were written and both remain proposals — neither is ratified, and the hardened version found five gaps in the first. OPEN v1: 3 layers, 4 open ratification questions; v2: 5 gaps identified, 6 open ratification questions, 8 [MAGIC] knobs listed	Status: PROPOSED — operator ratification required. Nothing here is self-ratified. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/ADR-2026-07-15-LIVE-EXPIRY-HARDENED.md
The expiry system has never been validated end to end: its positive/negative replay control has never been run. OPEN 0 E6 replay runs executed	E6 has never been run — this is why the probe is EVALUATING, not OPERATIONAL. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/ADR-2026-07-15-LIVE-EXPIRY-HARDENED.md
The 2026-dated literature citations underpinning the hardened expiry design are unverified and must pass a re-verification gate before any of them can be wired into a live gate. OPEN prior verification pass 15/15 title matches (2026-06-23)	The 2026-dated frontier citations (the `26xx.xxxxx` IDs) are workflow-surfaced and MUST pass the campaign's standing arXiv re-verification gate (the same 15/15 title-match pass performed on 2026-06-23) before ANY is wired into a gate. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/ADR-2026-07-15-LIVE-EXPIRY-HARDENED.md
The loop stopped on an explicit operator instruction after delivering the handover, and runs nothing until steered. CONFIRMED parked 2026-07-14 at iteration 103	Parked per the directive: "Once you have done that, you can stop until I give you further directions." findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/HANDOVER-PROMPT.md
The root verdict file is stale relative to the ledger: it reports the state after 13 iterations while the campaign ran to iteration 103. CONFIRMED verdict.md covers 13 iterations / 2 admissions / 3 kills; LEDGER row 117 records 16 admits / 45 kills at iteration 103	Status (2026-07-03, after 13 working iterations in one day): THE LAB WORKS AND HAS SHIPPED ITS FIRST VERDICTS. findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/verdict.md

The audit folder 17 markdown files

Source of record: findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/ — not published, so these are listed rather than linked.

File	Role
`ADR-2026-07-14-LIVE-EXPIRY.md`	Proposed live-expiry architecture v1: three concentric layers, expiry semantics, implementation sketch, and four blocking ratification questions.
`ADR-2026-07-15-LIVE-EXPIRY-HARDENED.md`	Hardened successor to the expiry ADR: five gaps in v1, a fail-closed anytime-valid stack, adopt/reject register, magic-knob table, pre-launch validation gate, and six ratification questions.
`BONEYARD.md`	Append-only do-not-re-propose list: seeded exclusions, grounded negatives, killed questions, and every new kill with its recorded reason and re-entry clause.
`C4-REGRESSION-FIXTURE.md`	The nine historical candidates replayed through the frozen ladder candidate by candidate, with the score table and the one gate-level deviation explained.
`CLAUDE.md`	Hub: what the loop is, the two-loop architecture, the seal status, the hub-and-spoke map, and the standing conventions.
`FOREX-ICP-WIRING-DEFECT.md`	Iteration-1 spoke: code-level root cause of the empty forex invariance map, fix plan, acceptance criteria, and the 2026-07-03 repair scorecard.
`FRONTIER.md`	The self-refilling priority queue: pinned opening items, SOTA candidates, the imported six-batch harvest feed with per-candidate consumption status, and standing refills.
`GATE-LADDER-M0-M5.md`	Per-gate specification M0–M5: the question each gate asks, its pass criterion, the exclusion reason it produces, and what each gate kills.
`HANDOVER-PROMPT.md`	The verbatim resume prompt for a future session: state at park, the full operator decision queue, and the deliverable index.
`LEDGER.md`	Append-only verdict ledger, rows 0a–117: one row per iteration with candidate, substrate, gate reached, verdict, scope, expiry, and evidence links.
`LOOP-PROMPT.md`	Canonical loop prompt SSoT plus the operator resource directive and the bigblack tmux launch instructions.
`M0-AMENDMENT-2026-07-03-OPERATOR-GO-LIFTED.md`	Superseding amendment recording the operator directive that lifted per-run permission for capped read-only spike runs.
`M0-AMENDMENT-2026-07-03c-NO-SYNTHETIC-DATA.md`	Superseding amendment banning all synthetic data and replacing the four self-test controls and the power gate with real-data-only designs.
`M0-MATRIX-ADMISSION-PRE-REGISTRATION.md`	The frozen rulebook: governing principle, frozen label, scope, gate ladder, verdict grammar, standing rules, seal, stop predicate, pinned opening iterations.
`SELF-TEST-CONTROLS.md`	The four blocking controls that validate the gate itself, with the seal-complete status and per-control outcomes.
`STATUS-EVIDENCE-MAP.md`	One-page map of which instruments back each feature status, the nine-slot certified stack as of 2026-07-04, what is not counted, and certificate validity conditions.
`verdict.md`	Plain-English bottom line — but frozen at 2026-07-03 / 13 iterations, so it is stale relative to the ledger's 117 rows.

Dashboard twin

findings/dashboard/campaigns/2026-07-02-matrix-admission-paradox-loop/index.html

Generated by findings/dashboard/build_audits.py from findings/evolution/audits/2026-07-02-matrix-admission-paradox-loop/AUDIT_LEDGER.json — never hand-edited. Each quote was verified to occur in the file named beside it when the ledger was written.