Commit graph

2 commits

Author SHA1 Message Date
Shay
417f71917c feat(adr-0041): core chat --show-verdicts + FanOutSink
Two thin layers closing the audit story end-to-end:

- core chat --show-verdicts prints format_verdict_summary(verdicts)
  to stderr after each turn.  Stdout stays clean for piped
  consumers.  Format is dense and terse; designed to skim, not
  machine-parseable (the JSONL sink owns that contract).

- FanOutSink forwards every emitted line to N sinks in declaration
  order.  Fail-fast on first error — consistent with ADR-0040's
  single-sink contract (audit failures surface).  Composes with
  any combination of JsonlFileSink / JsonlBufferSink / future
  sinks.

Two formatters, one bundle: format_turn_event_jsonl (machine,
ADR-0040) and format_verdict_summary (operator, ADR-0041) both
consume the same TurnVerdicts.  No risk of drift.

Summary format:
  [identity=0.83 safety=ok ethics=VIOLATED:foo refusal=- hedge=YES]

Audit story now reads end-to-end:
  - TurnVerdicts bundle (ADR-0039)
  - Machine JSONL sink (ADR-0040)
  - Fan-out + operator CLI (ADR-0041)

Files:
- chat/telemetry.py — FanOutSink dataclass, format_verdict_summary,
  _format_verdict_short helper
- core/cli.py — --show-verdicts on chat subparser; cmd_chat prints
  summary to stderr when set
- tests/test_telemetry_fanout_and_summary.py (new) — 13 tests
- docs/decisions/ADR-0041-cli-verdicts-and-fanout.md (new)

Verification:
- Combined pack-layer + telemetry suite: 212 green (was 199; +13)
- CLI suites unchanged: smoke 67, runtime 19, cognition 121
- core eval cognition: intent 100%, versor_closure 100% (baseline)
- Manual smoke: echo "light is" | core chat --show-verdicts prints
  expected bracketed audit line to stderr alongside response.
2026-05-17 21:47:47 -07:00
Shay
226f14a941 feat(adr-0040): structured-logging sink for turn-event audit
Adds the canonical JSONL sink surface consuming TurnEvent records
that ADR-0039 made uniform across main and stub paths.  One
deterministic line per turn; redact-by-default trust boundary;
opt-in content emission; runtime auto-emits on attached sink.

Trust boundary (CLAUDE.md):
- Metadata-only by default — no surfaces or input tokens emitted.
  include_content=True opt-in at attachment time.
- Path fixed at construction for JsonlFileSink; no user-controlled
  paths interpreted at emit time.
- Sink errors propagate — telemetry failures should surface, not
  silently drop audit signal.

Determinism:
- sort_keys=True; compact separators. Same event → byte-identical line.
- No implicit wall-clock; timestamps caller-provided.
- Field set fixed; missing TurnEvent attrs fall back to safe defaults.

API:
- serialize_turn_event(event, **kwargs) -> dict  (pure)
- format_turn_event_jsonl(event, **kwargs) -> str (pure, deterministic)
- TurnEventSink Protocol; JsonlBufferSink; JsonlFileSink
- ChatRuntime.attach_telemetry_sink(sink, *, include_content=False)
- _emit_turn_event invoked after both turn_log.append sites

Wire format (alphabetised, always present): cycle_cost_total,
dialogue_role, ethics_pack_id, ethics_runtime_checkable_count,
ethics_upheld, ethics_violated, flagged, hedge_injected,
identity_pack_id, refusal_emitted, safety_pack_id,
safety_runtime_checkable_count, safety_upheld, safety_violated,
stub_path, turn, vault_hits, versor_condition.

Conditional: identity_* (when score present), surface /
walk_surface / articulation_surface / input_tokens (when
include_content=True), timestamp (when provided).

Files:
- chat/telemetry.py (new) — serializer, formatter, sinks
- chat/runtime.py — attach + emit + post-append calls
- tests/test_telemetry_sink.py (new) — 29 tests
- docs/decisions/ADR-0040-telemetry-sink.md (new)

Verification:
- Combined pack-layer + telemetry suite: 199 green (was 170 after
  ADR-0039; +29)
- CLI suites unchanged: smoke 67, runtime 19, cognition 121
- core eval cognition: intent 100%, versor_closure 100% (baseline)
2026-05-17 21:39:58 -07:00