Iteration 10 — Verdict-Stability k-of-N: THE CAMPAIGN'S FIRST ADMISSION

Iteration 10 — Verdict-Stability k-of-N FIRST ADMISSION

What happened in this iteration, in plain words: third time's the charm — and this time it's the charm because of the two kills that preceded it. The new counter asks the literal status question per regime ("was this feature orthogonal here, yes or no?") and counts the yeses. It walked the full six-gate ladder and passed everything that matters: the known cheat (a feature that's a committed duplicate of a core signal) reads exactly 0-of-10, as ground truth demands; four engineered features read a perfect 10-of-10 — orthogonal in every regime the campaign has ever sliced; the counter is demonstrably NOT just a repackaging of existing instruments; and on held-out data its history strongly predicts who keeps their orthogonal status (correlation +0.59, with the shuffled control reading ~0). The lab issued its first certificate: ADMIT — with a scope, a role, and an expiry date, per the frozen grammar. And a satisfying scientific rhyme: the volume family that iteration 9 found maximally coupling-stable reads 0-of-10 orthogonal here — the two instruments confirm each other's story from opposite sides. The FRAGILE-vs-CONDITIONAL boundary is now assignable on crypto for the first time in the campaign's history.

Preflight (A0)

2026-07-03 13:19 UTC — load1 6.27 / 32 cores (≤24, transient) · no competing nasimubd jobs · 21 GiB available, si/so≈0 · clickhouse-server active · sidecar active + /health healthy → ALL PASS. Run: hermetic (committed sha-pinned cells, ZERO ClickHouse), 0.25 min inside the 5c/5G scope, zero generated values (03c).

The ladder walk (all six gates, one bounded slice)

INSTRUMENT: k_v(f) = # of the 10 regime environments where feature f was ORTHOGONAL (admitted G0 verdict: redundancy ≤ TAU 0.85) M0 status-linkage PASS — FRAGILE↔CONDITIONAL is DEFINED on such counts M1 trust anchor PASS — ground truth (c), committed known answers: turnover_imbalance (committed duplicate, |ρ|=1.0 vs ofi) → k_v = 0 EXACT ✓ curated instruments → k_v = 10 of 10: aggression_ratio · bar_petrosian_fd · bar_katz_fd · bar_dispersion_entropy ✓ M2 substrate PASS — verdict counting adds no assumption beyond admitted G0 M3 redundancy attack CLEAR — not a repackaging: ρ(k_v, mean redundancy) = −0.753 (kill bar 0.95); new content = cross-regime verdict patterns no single-cell gauge reports M4 incremental value narrow miss on criterion 1: ΔAUC +0.0135, CI [−0.0025, +0.0351] ∋ 0 (observed above the pairing-null band +0.004) M5' power DECISIVE — (a) known-answer anchor exact (above); (b) direction on held-out among-orth rows (n=256): ρ(expanding orth-history, stays-orthogonal) = +0.585 95% CI [+0.418, +0.825] pairing-null band [−0.065, +0.149] → detects its target at our N; detection collapses when pairing destroyed VERDICT (frozen grammar): ADMIT( crypto · 10-window regime grid · role: FRAGILE/CONDITIONAL boundary readout of G0 · expiry: next completed crypto regime slice ) — the campaign's FIRST admission (LEDGER row 15) CROSS-INSTRUMENT CONSISTENCY (iter 9 ↔ iter 10): volume family: max coupling-stability (iter 9) ↔ k_v = 0 orthogonal (here) curated bar_*: breaking coupling (iter 9) ↔ k_v = 10 orthogonal (here) → two independent readouts, one coherent mechanism

Technical record

Item	Value
Ledger	row 15 — ADMIT (first of the campaign); certified stack grows to 8 slots (G0–G11 seven + k_v readout)
Consequence	Loop-2 may now assign FRAGILE and CONDITIONAL on crypto using k_v + unexpired G0 (the boundary was starved since campaign start)
Journey	parametric variant killed (iter 7) → regression-invariance axis killed with inverted discovery (iter 9) → verdict-counting variant ADMITTED (iter 10) — the kills steered the design
Expiry	next completed crypto regime slice (interim §5.2); block-permutation refinement inherited as an expiry-review item
Numbers	anchor k_v(turnover_imbalance)=0 exact · 4 features at 10/10 · M3 ρ=−0.753 · M4 ΔAUC +0.0135 CI ∋ 0 (narrow miss) · M5′ ρ=+0.585 CI [+0.418,+0.825] · wall 0.25 min
Harness	`verdict_stability_kofn.py` · `verdict_stability_results.json` — hermetic, sha-pinned panel
Compute / prod impact	0.25 min capped · zero ClickHouse · zero writes outside audit folder + dashboard

Item

Value

Ledger

row 15 — ADMIT (first of the campaign); certified stack grows to 8 slots (G0–G11 seven + k_v readout)

Consequence

Loop-2 may now assign FRAGILE and CONDITIONAL on crypto using k_v + unexpired G0 (the boundary was starved since campaign start)

Journey

parametric variant killed (iter 7) → regression-invariance axis killed with inverted discovery (iter 9) → verdict-counting variant ADMITTED (iter 10) — the kills steered the design

Expiry

next completed crypto regime slice (interim §5.2); block-permutation refinement inherited as an expiry-review item

Numbers

anchor k_v(turnover_imbalance)=0 exact · 4 features at 10/10 · M3 ρ=−0.753 · M4 ΔAUC +0.0135 CI ∋ 0 (narrow miss) · M5′ ρ=+0.585 CI [+0.418,+0.825] · wall 0.25 min

Harness

verdict_stability_kofn.py · verdict_stability_results.json — hermetic, sha-pinned panel

Compute / prod impact

0.25 min capped · zero ClickHouse · zero writes outside audit folder + dashboard

▶ Next iteration

Iteration 11 = Loop-2's first consumption: draft the first crypto FRAGILE/CONDITIONAL status declarations. In plain words: with the counter certified, the campaign can finally do the thing it was built for — assign statuses. Next firing drafts the first declaration table: per feature, k_v + the admitted-stack evidence → FRAGILE (low k_v) or CONDITIONAL (high k_v, with its regime scope named), each with evidence links, published as a PROPOSAL page for operator review before any status becomes official. The k_v=10 quartet (aggression_ratio, petrosian_fd, katz_fd, dispersion_entropy) are the first STABLE-candidate flags too — pending Frontier #3's identifiability-sharpened F014 question. Steering alternatives: full forex grid (fix-plan step 6) + forex verdict-stability, or F014 first.

Iteration 10 · 2026-07-03 · Frontier #2 closed on crypto: first ADMIT of the campaign · hermetic · capped · zero synthetic data · append-only