Ce Logging Observability

CE Logging and Observability

You are managing the logging and observability surface of the library. All changes must comply with the architectural rules in ADR-028 and the contributor guidelines in Standard-005.

Logger Domains (Mandatory)

All loggers must live under the calibrated_explanations root and use one of the four reserved domains below:

import logging
# Pattern 1: Module-level (preferred if path matches domain)
logger = logging.getLogger(__name__)

# Pattern 2: Explicit domain (if path does not match)
logger = logging.getLogger("calibrated_explanations.governance.checkpoints")

Operational vs. Governance Separation

• Governance/Audit events MUST use the governance.* domain. These include checkpoint creation, rollbacks, binary trust decisions (accept/deny), and policy changes.
• Operational events (fallbacks, cache hits, performance data) MUST use core.* or plugins.*.
• Fallbacks MUST use the WARNING level with a clear explanation of why the fallback was triggered.

Logging Context (ADR-028 §3)

Use the structured context helpers from calibrated_explanations.logging to propagate identifiers like request_id, tenant_id, and explainer_id.

from calibrated_explanations.logging import update_logging_context, logging_context

# 1. Update global (thread-local) context
update_logging_context(explainer_id="exp-123", tenant_id="acme")

# 2. Scoped context (recommended for local blocks)
with logging_context(request_id="req-456", mode="factual"):
    logger.info("Starting explanation generation")

Supported Context Keys

Must be one of: request_id, tenant_id, explainer_id, checkpoint_id, plugin_identifier, mode.

Contributor Checklist

1. No basicConfig — never call logging.basicConfig() in library code.
2. No PII — ensure no sensitive data (e.g., feature values, training data) is included in log messages.
3. Structured fields — prefer passing IDs via context rather than string concatenation.
4. JSON compatibility — ensure logs are compatible with structured formatters (standard fields like extra are supported).
5. Level check — use DEBUG for noise, INFO for normal state, WARNING for fallbacks/corrections.
6. Exception context — use logger.exception("message") inside except blocks to capture stack traces automatically.

Testing Logging

Verify logging output in tests using the caplog fixture:

def test_should_log_warning_when_plugin_fails(caplog):
    with caplog.at_level(logging.WARNING, logger="calibrated_explanations.plugins"):
        # ... trigger fallback ...
        assert "fallback triggered" in caplog.text
        assert any(r.explainer_id == "exp-1" for r in caplog.records)