Privacy-safe logs and performance diagnostics

Inputs

Define the failure claim, candidate/build identity, workload, process, time window, diagnostic tools, privacy scope, and baseline evidence.

Rules

• OBS-001 — Log contracts, not payloads. Prefer state transitions, operation categories, and bounded opaque correlation. Never log credentials or protected content for convenience.
• OBS-002 — Measure an interval. Use paired operation boundaries and consistent time units. Instrument success, failure, cancellation, and timeout without double-ending intervals.
• OBS-003 — Match diagnostic to failure. Distinguish CPU work, blocking, UI invalidation, allocation growth, and retained objects before proposing fixes.
• OBS-004 — Preserve raw provenance. Keep original traces/crashes privately with tool version and capture conditions. A parser summary is an interpretation, not the complete evidence.
• OBS-005 — Verify without the harness. Retest the final candidate without temporary fixtures or debugger effects; compare the same workload and report unmeasured claims.

Workflow

1. Capture the smallest reproducible scenario and privacy-reviewed telemetry.
2. Inspect symbolication and timeline coverage before trusting a summary.
3. Form one ownership or performance hypothesis and add a focused regression test.
4. Measure again and remove or bound temporary diagnostics.

Verify

Check cancelled intervals, private interpolations, missing symbols, empty traces, repeated presentations, and retained task/observer ownership. Inspect actual artifact contents before sharing.

Output

Return the candidate, capture contract, evidence-supported diagnosis, change, comparison, and unresolved hypotheses.

References

• Logging, signposts, and trace interpretation.
• Memory graphs and crash investigation.