feat(F035): Observability — event bus stats, causal tracing, drift detection, context visibility

[tfatykhov]

Summary

Implements F035 Observability with all 4 sub-features, giving Nous full visibility into what its autonomous systems are doing and what the LLM actually sees.

F035.1 — Event Bus Observability

• EventBusStats with in-memory counters, per-handler success/error/timing, ring buffer of recent events
• Wired into EventBus._dispatch and _safe_handle (zero-allocation hot path)
• get_stats() on SessionTimeoutMonitor, SleepHandler, HeartbeatRunner
• GET /events/stats, GET /events/recent endpoints
• format_event_bus_status() for Telegram /status

F035.2 — Causal Chain Tracing

• event_id (12-char hex), trace_id, caused_by on Event dataclass + ORM model
• DB migration with indexed columns
• Root event emission: heartbeat_tick, session_ended, turn_completed, sleep_started
• Child propagation across all handlers (episode_summarizer, fact_extractor, sleep_handler, outcome_detector, subtask_worker)
• State-modifying events tagged with modifies in data
• GET /events/trace/{id}, GET /events/recent-traces (CTE query), GET /events/modifications

F035.3 — Behavioral Drift Detection

• BehaviorSnapshot (27 metrics), DriftDetector (z-score analysis, per-metric thresholds)
• BehaviorDriftCheck as heartbeat check — captures snapshots, stores to DB, detects anomalies
• Warning (2σ) vs Alert (3σ) severity tiers
• GET /behavior/snapshot/latest, /behavior/trends, /behavior/anomalies, /behavior/drift-report

F035.4 — Context Visibility

• ContextLogger with section parser, token estimation, ring buffer for full payloads
• Hooks into _build_api_payload() — logs every API call with token breakdown by section
• update_response() wired to capture actual vs estimated tokens after API returns
• Memory leak prevention (_sync_entries_index)
• DB writer for persistent context_log table
• GET /context/log, /context/log/{id}, /context/log/{id}/payload, /context/log/{id}/sections, /context/diff

Review Process

• 3-agent plan review (architect + DB specialist + devil's advocate): 11 P1s identified and fixed
• Implementation review: 1 new P1 + 3 P2s identified and fixed
• All agent_id scoping enforced on new tables
• 76 tests passing (68 new + 8 existing EventBus)

New Files

• nous/observability/__init__.py, context_logger.py, snapshots.py, drift.py
• sql/migrations/026_observability.sql
• tests/test_event_bus_observability.py, test_causal_tracing.py, test_context_logger.py, test_drift_detection.py

Modified Files (15)

• nous/events.py — EventBusStats + Event causal fields
• nous/storage/models.py — ORM Event columns
• nous/brain/brain.py — emit_event passthrough
• nous/api/rest.py — 15 new endpoints
• nous/api/runner.py — context logger hook + update_response
• nous/config.py — 7 new settings
• nous/main.py — wiring for all components
• nous/telegram_bot.py — 3 formatting functions
• nous/heartbeat/runner.py, checks.py — stats + drift check
• nous/handlers/* — trace propagation
• nous/cognitive/layer.py — root event trace IDs

Test plan

• 76 unit tests pass across 4 new test files
• 8 existing EventBus tests pass (no regressions)
• All review P1 fixes verified applied
• Integration test with running Postgres + migration 026
• Verify Telegram formatting renders correctly
• Verify dashboard endpoints return correct data shapes

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.