Skip to content

AIGP Dialect: Autonomous Intelligence Governance — The AI Observable Surface — 2. The Observable Surface of Autonomous Intelligence

AIGP SpecificationAIGP Dialect: Autonomous Intelligence Governance — The AI Observable Surface › 2. The Observable Surface of Autonomous Intelligence

← 1. Abstract · Section index · 3. Domains of Concern (RFC-034 Instantiation) →

2. The Observable Surface of Autonomous Intelligence

2.1 Completeness Claim

An AI system that senses, reasons, acts, communicates, delegates, and remembers produces observable events in exactly six categories. No seventh category has been identified; if one is, this dialect must be versioned.

Category Definition Cybernetic ground
Action The system does something in the world Actuation (Wiener §4)
Interaction The system exchanges messages with another entity Communication (Shannon, Watzlawick)
Perception The system receives or constructs input Perceptual Control (Powers §8)
Decision The system selects among alternatives Regulation (Ashby §6)
Non-action The system does NOT do something expected Silence as signal (Laouris A1)
Self-observation The system observes its own state Second-order (von Foerster §7)

2.2 Exhaustive Event Enumeration

Within these six categories, the following observable events constitute the complete event grammar for AI governance:

Actions (the system does)

Event ID Event Stage Observable
A1 Model invocation 9 model_id, input_tokens, output_tokens, duration_ms
A2 Tool execution 10, 21 tool_name, params, result, duration_ms
A3 Helper call 24 helper_id, endpoint, token_present, response_status
A4 Output generation 12 output_length, format, classification
A5 Memory write 25 key, classification, retention_policy
A6 Delegation 19, 24 target_agent, scope_envelope, depth

Interactions (the system communicates)

Event ID Event Stage Observable
I1 Receive instruction 4 source, content_length, intent_classification
I2 Send response 12, 13 recipient, content_length, status
I3 Agent-to-agent message 24 sender, receiver, protocol, delegation_token
I4 Governance report (RECORD) 14 app_id, timestamp, summary
I5 Governance trace (TRACE) 14 trace_id, spans, stages_traversed
I6 Heartbeat agent_id, timestamp, status

Perceptions (the system receives/constructs)

Event ID Event Stage Observable
P1 Context assembly 5, 6 sources, total_tokens, retrieval_method
P2 Document retrieval 7 query, results_count, source_types
P3 Model selection signal 8 policy_model, actual_model, source
P4 Policy receipt policy_version, policy_id, received_at
P5 Environment reading env_vars, config_source, staleness

Decisions (the system chooses)

Event ID Event Stage Observable
D1 Model selection 8 chosen_model, alternatives_available, policy_constraint
D2 Tool selection 10, 21 chosen_tool, alternatives_considered, policy_allowed
D3 Delegation decision 19 delegate_to, scope_narrowing, depth_remaining
D4 Response strategy 11 reasoning_steps, intermediate_results
D5 Autonomy boundary 23 tokens_remaining, budget_pct, escalation_needed

Non-actions (the system does NOT do)

Event ID Event Stage Observable
N1 Missing RECORD 14 expected_at, not_received_by, agent_id
N2 Missing heartbeat last_seen, expected_interval, current_gap
N3 Policy not refreshed policy_version, latest_available, staleness_hours
N4 Tool not called 21 available_tools, invocation_count_zero, expected_usage
N5 Delegation refused 22, 23 reason, budget_exhausted, scope_exceeded

Self-observations (the system observes itself)

Event ID Event Stage Observable
S1 Error detection 26 error_type, consecutive_count, circuit_state
S2 Budget tracking 23 daily_tokens_used, daily_max, remaining_pct
S3 Policy version awareness current_version, latest_version, drift_detected
S4 Health self-report status, uptime, last_error
S5 Performance self-assessment avg_latency, p95_latency, success_rate


← 1. Abstract · Section index · 3. Domains of Concern (RFC-034 Instantiation) →