Crucible · SR 11-7 export preview acme-fraud · spec 9f2a4c7b · 2026-Q2
download .pdf manifest.json
Model Risk Report

acme-fraud · Q2 2026

SR 11-7 / OCC 2011-12 alignment · sealed spec 9f2a4c7b
Report id · MR-2026Q2-0118
Generated · 2026-06-22 15:42Z
Author · m.chen (model-risk)
Signatures · 2 of 2 affixed

1 · Executive summary

The acme-fraud detection system has converged on its sealed held-out corpus in Q2 2026 with measured recall 0.93 and attack-success rate 0.07, against a regulatory red line of recall ≥ 0.90. One halt event (h_19c4) was triggered intra-quarter and is currently active; lift conditions are partially satisfied and four downstream runs are blocked. No untracked overrides occurred.

recall (sealed)
0.93
≥ 0.90 · pass
ASR
0.07
peak 0.41 at r.18
val ↔ heldout gap
0.02
≤ 0.05 · no overfit
white-box recall
0.92
CI [0.881, 0.946]

2 · Governance & controls

ControlPolicyState
sealed spec, append-onlyMRG-1.0in force
held-out corpus never trained onMRG-4.2attested · monthly diff
two-person rule on patch applyMRG-11.1 (4-eyes)11 / 11 patches
halt on recall < 0.90 (3-round debounce)MRG-12.41 active halt
cost ceiling · $25/runMRG-12.70 breaches Q2

3 · Halt event h_19c4

Halt h_19c4 fired at 2026-06-22 14:14:02Z when held-out recall fell to 0.86 across three consecutive rounds (r.32, r.33, r.34). Cause: META-INV-04 (case-folding) regression. Four runs are blocked. Blue patch p_2a17 is proposed and awaiting verification; 0 of 2 reviewer acks affixed. No override was issued.

4 · Model lineage

SpecSealedRangePatchesFinal recall
9f2a4c7b2026-04-01r.00 → r.46110.93
8d11e3a42026-01-12r.00 → r.3890.91
7a09b2f12025-10-04r.00 → r.2970.88

5 · Sign-off

Model owner
s. lydon
2026-06-22 15:39Z · key 0xA3F2…
Model risk
m. chen
2026-06-22 15:41Z · key 0x14B0…
MR-2026Q2-0118 · crucible · acme-fraudpage 1 of 1