Theater Mission AI Confidence Degradation Early Warning Cell

Theater Mission AI Confidence Degradation Early Warning Cell

Mission Scope

• Treat this skill as an advisory planning and decision-support aid for U.S. warfighter missions in its domain.
• Confirm mission systems in scope, authority boundaries, model dependencies, and acceptable confidence floors before producing recommendations.
• Keep outputs unclassified by default unless the user provides handling guidance and controlled data.

Workflow

1. Frame the AI-enabled mission threads, critical model dependencies, and commander decision points.
2. Detect drift, poisoning, bias, stale training assumptions, or sensor-distribution change that could degrade output trust.
3. Build recommended hold, rollback, monitor, and alternate-workflow branches with explicit operational tradeoffs.
4. Bind each recommendation to tool telemetry, packetized evidence, human approval gates, and degraded manual workflows.

Required Output Format

1. Situation snapshot.
2. Recommended confidence posture and rationale.
3. Alternative branches with degradation triggers.
4. Decision points now/next/pre-delegated.
5. Staff tasking by owner and suspense.
6. Model trust packet, protocol bindings, and confidence notes.

Domain Products

Primary products for this skill: model trust watchlist, confidence degradation trigger table, and rollback-or-retain decision board.

Domain Toolchain Defaults

• Primary: tool_suite_id=ts-theater-mission-ai-confidence-early-warning-v1 with protocol_stack_id=ps-theater-mission-ai-confidence-early-warning-stack-v1.
• Alternate: commander-approved baseline-model comparison board plus manual review queue.
• Degraded: approved-baseline-only posture with human-only release decisions.

External Tools and Protocol Integration

• Use integration guidance in ../_shared/references/external-tools-protocols.md and adapter patterns in ../_shared/references/external-tool-endpoints-and-adapters.md.
• Include packet_id=DPL-MISSION-AI-CONFIDENCE-001 for critical recommendations.
• Prioritize these protocol families for this domain: signed model attestations, API/JSON, STIX/TAXII, and USMTF.
• Include source system, refresh UTC, confidence, drift indicators, and unresolved model-governance gaps in each recommendation.

Authority and Assurance Gates

• Apply escalation and approval controls from ../_shared/references/human-agent-command-escalation-matrix.md and ../_shared/references/warfighter-tool-authority-gates.md.
• Run assurance checks from ../_shared/references/us-joint-protocol-assurance-drill.md and ../_shared/references/mission-assurance-checklist.md.
• If model provenance, validation data, or override authority is uncertain, downgrade to advisory-only and require human command review.

Guardrails

• Do not fabricate model performance, retraining status, or authority to deploy or rollback models.
• Separate detected indicators from inferred causes.
• Prefer conservative recommendations when AI outputs influence fires, life-safety, or strategic posture.