A Mission Engineering technician at Acquisitio hires this to triage inbound capability requests in <2 hours instead of the current 14-day queue, because the legacy ticket-routing system buries requests behind manual classification — and the queue depth is rising as DoD program tempo accelerates.
Engagement constitutionCost AVOIDANCE only — no displacementCTR target ≥ 0.40AI-risk gate must clear 80%Transfer-Posture verified before ScaleITAR-aware throughout
⬤ Perseveremet ✓
Accuracy ≥ 85% AND latency < 2h on held-out eval
current88%
threshold mark at 85%+3 pts above
committed 2026-05-03 · before MVP ran
◐ Pivotnot triggered
Accuracy 60–85% OR HITL load > 40% — pivot on bucket re-diagnosis
HITL load · current8%
HITL threshold at 40%far below
committed 2026-05-03 · before MVP ran
○ Killnot triggered
Accuracy < 60% OR domain-sparsity probe fails — out-of-regime
Marginal returns to prompt complexity collapse after exemplar-rich few-shot. This is Bucket 1 · Prompting failure — the model has the capability, the deployment was under-exemplar'd. Tier (b) lift confirms.
Failure-mode bucket · 6-way classification
1
Prompting failure
⬤ this case
2
Context / RAG failure
eliminated
3
Tool-design failure
eliminated
4
Harness / orchestration failure
eliminated
5
Capability gap · domain sparsity
eliminated
6
Evaluation methodology failure
eliminated
Asset Library fix · prompting-pattern-v2 (exemplar harvest)Pivot if curve doesn't shift after fix · trace lineage available
4-doc PLAN · Business Case / Market Validation / Development / Scaled GTM
75% authored · 1 expanded
Business CaseFY27 modeled cost avoidance $450k/yr · 3 engineers not hired · payback 4 months12 sections updated 2d ago▸