Loop Governance Policy Engine
PromptKing doesn't just track AI usage. It makes AI execution auditable, governable, and defensible.
AI started as a cost center. With loops, it becomes a risk center. v3.71.0 is where that shift becomes a product.
What changed between v3.69 and v3.71
- v3.69 detected that loops exist and cost too much.
- v3.70 gave you the controls to govern them.
- v3.71 makes the governance defensible to a regulator, an auditor, and a board.
The regulatory moment
Article 14 of the EU AI Act becomes enforceable August 2, 2026. These are not optional and they are not theoretical. Penalties reach 7% of global annual turnover or €35 million. Every ungoverned loop running in a high-risk system is a potential Article 14 gap. Most IT teams have no way to know which of their pipelines qualify.
What v3.71.0 ships
- Loop Governance Policy Engine — per-pipeline MAX_ITER, session cost ceiling, HITL checkpoint, verification gate, comprehension review gate, log retention. Three policy templates. Auto-pause on breach.
- Simulate before you enforce — before any policy activates, show exactly what last month's loops would have cost. How many paused. How much saved. Enforce only after review.
- Regulatory Compliance Scorecard — every loop scored 0-100 against EU AI Act Articles 12, 14, 19, 50, California SB-833, NIST AI RMF, SOC2 Type II. Specific gap labels per article.
- Context Health Score — context window utilization, compaction events, session restart count combined into 0-100. CRITICAL below 40 means the agent was operating on degraded context.
- Comprehension Debt Tracker — percentage of loop-generated PRs with no mandatory human review gate. The metric no engineering org is measuring.
The multi-buyer moment
CFO: Sees loop spend split into verified output vs ungoverned waste — the number that moves budget conversations from usage to accountability.
CIO / IT: Gets per-pipeline policies with MAX_ITER, cost ceilings, and auto-pause before the billing cycle ends.
Legal: Article 14 human oversight, Article 12 logging, and Article 50 disclosure scored per loop with specific gap labels for audit.
Security: Training data risk flags, external data access signals, and intent drift detection on every loop session.
Engineering leadership: Comprehension review gates and context health scores surface code quality decay before it reaches production.
v3.68 makes people aware. v3.69 makes them nervous. v3.70 makes you necessary. v3.71 makes you unavoidable.
Govern your loops now
Simulate before you enforce. Score every loop against Article 14.