Iteration 12 — Feed batch B-03: STABLE's second handoff, 33 instruments on the lab's shelf

Iteration 12 — Feed batch B-03 published THIRD HANDOFF · 33 FED ALL SCREEN QUEUES EMPTY

2026-07-07 · twelfth firing · ledger row 17 · batch B-03 · campaign branch loop/matrix-harvest-campaign

In plain words: the third delivery is on the certification lab's shelf — thirteen more STABLE-boundary instruments, the boundary that decides the project's hardest question ("will this feature still be independent tomorrow?"). The ordering follows the campaign's standing economics: instruments whose papers ship exact statistical guarantees go first (a regularity-free test cheap enough to build in an afternoon leads), the three-piece anchor toolkit travels as a unit, the robustness diagnostics with hand-built trust anchors ride mid-pack, and the two entries carrying known real-data risks go last with those risks named in writing — one's guarantee assumes independent sampling our markets don't provide, the other struggles on noisy stochastic systems. Every flag from the screening travels with its instrument; the lab starts with our homework, not from scratch. The running total: thirty-three vetted instruments delivered across three batches, every screened queue now empty, and the pipeline turns back to the boundaries still waiting for their first interrogation.

Batch scoreboard

B-01 STABLE ×10 (iter 4) invariance testers — seqICP, IAS, IPP, HSIC-X… B-02 FAIL ×10 (iter 9) redundancy detectors — Reduced TE, Hoeffding D… B-03 STABLE ×13 (iter 12) universal inference, safe testing, anchor triad, causal-structure machines, robustness margins ───────────────────────────── 33 instruments fed · all screen queues empty · catalog 67/300

B-03 order (cheapest kill first)

#	id	Instrument	Kill-cost note
1	H-053	Universal inference	trivial construction, finite-sample bound — cheapest
2	H-058	Safe testing (safestats)	candidate-A-root flag — coordinate with the e-value track
3–5	H-065/066/067	Anchor triad (J · F · AR)	cheap econometric classics — run as a unit
6	H-027	LPCMCI	mature FOSS, autocorrelation-native; heavier compute
7	H-030	DYNOTEARS	M1-borderline: bootstrap anchor to wire
8–9	H-060/061	DRO cousins (GroupDRO · CVaR)	grouped attack recommended
10	H-063	V-REx risk-variance	M1-borderline: null collapse is the real test
11	H-064	Calibration gap	M1-borderline: ECE binning bias
12	H-062	W-DRO certificates	M2-risk: IID certificate — block recalibration required
13	H-028	Convergent Cross Mapping	M2-risk: stochastic substrate; heaviest machinery

Instrument

Kill-cost note

H-053

Universal inference

trivial construction, finite-sample bound — cheapest

H-058

Safe testing (safestats)

candidate-A-root flag — coordinate with the e-value track

3–5

H-065/066/067

Anchor triad (J · F · AR)

cheap econometric classics — run as a unit

H-027

LPCMCI

mature FOSS, autocorrelation-native; heavier compute

H-030

DYNOTEARS

M1-borderline: bootstrap anchor to wire

8–9

H-060/061

DRO cousins (GroupDRO · CVaR)

grouped attack recommended

H-063

V-REx risk-variance

M1-borderline: null collapse is the real test

H-064

Calibration gap

M1-borderline: ECE binning bias

H-062

W-DRO certificates

M2-risk: IID certificate — block recalibration required

H-028

Convergent Cross Mapping

M2-risk: stochastic substrate; heaviest machinery

Coverage counters after this firing

Boundary	Minimum	Verified	Screened-pass	Fed
STABLE	100	33	23	23 (B-01 + B-03)
FRAGILE↔CONDITIONAL	60	6	0	0
CONDITIONAL-scoping	40	3	0	0
FAIL	40	13	10	10 (B-02)
EXPIRY	40	10	0	0
cross-cutting	20	2	0	0
Total	300	67	33	33

Boundary

Minimum

Verified

Screened-pass

Fed

STABLE

100

23 (B-01 + B-03)

FRAGILE↔CONDITIONAL

CONDITIONAL-scoping

FAIL

10 (B-02)

EXPIRY

cross-cutting

Total

300

▶ Next iteration

Iteration 13 = EXPIRY screen S-04 (cohort of 10 ready) — the live-decay monitors face M0/M1: the e-detectors, confidence sequences, conformal test martingales, sequential quantile bands, the changepoint detectors, drift detectors, and adaptive conformal entries. Note: four EXPIRY entries are still provenance-pending from the scaffold (H-011..H-014) — they cannot be screened and will be named as such; the screenable cohort is the remainder. After S-04: FRAGILE↔COND harvesting (deficit 54 — family 9 stability indices, family 13 leads, local Gaussian correlation).

Steering point: no directive = EXPIRY screen next firing. Directives are ledger-recorded verbatim before taking effect.

Machine-readable record: ledger row 17 + FRONTIER-FEED.md §B-03 + catalog §B-03 transitions. WebSearch: blocked this firing (recorded). Pages are append-only.