Escalation SLA / Supervised-Autonomy Receipt

Note (dated 2026-08-04): a chain composes independent computations performed in an analytical order. No step's output is carried into the next step by the execution model; each step computes from its own declared inputs. This is true of every chain on this site, including chains whose per-step copy below reads as a data handoff.

"Agent halted → human X answered in T → decision D" as protocol-native evidence. A work mandate compiler establishes the supervision context, an agentic AI autonomy/risk classifier grades the agent, and a gate routes: a failing grade escalates to human review (OCG §22.8 HALT, open record); any passing grade records the decision directly into an EU AI Act Art 12(2)-conformant decision log. Built for an AI-Act Art.12 obligated deployer or a supervised-autonomy desk needing replayable escalation-SLA evidence, not for asserting a broader certification.

OpenChainGraph · 3 Steps governance mandate 1 Escalation Gate chain_depth:3 Client-Side · Zero PII

Chain Stages · 3 Steps (1 escalation gate)

1 ROOT node

Work Mandate Compiler art-274-compile-work-mandate

the compiled work mandate (escalation_triggers, scope, conditions) establishes the supervision context for Stage 2 autonomy/risk classification

MCP Call · compile_work_mandate

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "compile_work_mandate",
    "arguments": {
      "mandate": {
        "mandate_type": "work_mandate",
        "scope": {
          "tool_ids": ["art-67-agentic-ai-risk-classifier", "art-236-build-ai-decision-log-record"],
          "chains": []
        },
        "conditions": [],
        "escalation_triggers": [
          {"pointer": "/overall_grade", "op": "eq", "value": "F"}
        ]
      }
    }
  },
  "id": 1
}

↓

2 STEP · ESCALATION GATE node gate

Agentic AI Risk & GPAI Governance Classifier art-67-agentic-ai-risk-classifier

overall_grade and governance_tier → GATE: overall_grade='F' escalates (HALT, open OCG §22.8 escalation record, closed via verify_escalation_closure on the anchor worker, closing decision available as an AuthZEN 1.0-mapped evaluation at anchor.ainumbers.co); any passing grade continues to Stage 3 Art 12 decision-log recording.

§22.8.1 Decision Gate

rule[0]: if output.overall_grade == "F"  →  next: escalate   (§22.8.2 HALT + open record)
default:                                →  next: art-236-build-ai-decision-log-record

Default continues to: art-236-build-ai-decision-log-record

MCP Call · classify_agentic_ai_risk

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "classify_agentic_ai_risk",
    "arguments": {
      "autonomy_level": "supervised",
      "financial_use_case": "other",
      "human_oversight": "review-before-action",
      "model_type": "GPAI",
      "systemic_designation": "no",
      "training_compute": 0,
      "transparency": "none",
      "gpai_documentation": "none",
      "systemic_risk_eval": "n/a",
      "downstream_highrisk": "false"
    }
  },
  "id": 2
}

↓

3 TERMINAL node

AI Decision Log Record Builder (EU AI Act Art 12) art-236-build-ai-decision-log-record

art12_completeness_score, chain_position, and anchor_surface instructions complete the supervised-autonomy escalation-SLA receipt. Final stage.

MCP Call · build_ai_decision_log_record

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "build_ai_decision_log_record",
    "arguments": {
      "model_id": "supervised-agent-v1",
      "model_version": "1.0.0",
      "input_digest": "abc123def456abc123def456abc123def456abc123def456abc123def456abcd",
      "output_digest": "def456abc123def456abc123def456abc123def456abc123def456abc123def4",
      "decision_label": "AUTONOMY_TIER_LOGGED",
      "confidence": 0.9,
      "override_flag": false,
      "subject_ref": "SYNTHETIC-AGENT-SESSION-001",
      "retention_months": 12,
      "operator_id": "supervised-autonomy-desk"
    }
  },
  "id": 3
}

Question: Our agent amends orders with no human in the loop. Which oversight tier and GPAI duties apply to it under the AI Act? Run the AINumbers MCP tool `run_chain` with {"chain":"escalation-sla-supervised-autonomy-receipt"} and read `art-67-agentic-ai-risk-classifier.autonomy_oversight_verdict`. Verify: call `verify_execution_hash` (https://mcp.ainumbers.co/mcp) with `claimed_hash` set to `composite_execution_hash` and the full `composite_artifact`. Ledger, for a human re-check: https://ledger.ainumbers.co/ PII rule: send synthetic or anonymised inputs only. The MCP server runs these kernels and logs no payloads. Chain page: https://ainumbers.co/chaingraph/chains/escalation-sla-supervised-autonomy-receipt.html

Escalation SLA / Supervised-Autonomy Receipt

Ask your agent