Skip to main content

Investigations and lessons

Each record summarizes what was established during a dated window: what happened, what we learned, and what stayed uncertain. Records describe the past; they do not certify current health. Detailed evidence and unfinished work are kept privately.

RecordPeriodAbout
Storage reliability, capacity, and backups2026-10Space reclamation, two-copy placement, the stock runtime cutover
Lakehouse and workflow experiments2026-10Retiring Airflow, Spark, Trino, and Temporal
Operation contracts and agent workflows2026-08 to 10Shared safety contracts and workflow lessons
Metric and resource experiments2026-08 to 10Measurement contracts and request right-sizing
Security and credential delivery2026-07 to 09Secret-provider quota and review findings
Flux and platform reliability2026-05 to 10Probe-blocking policies and dependency ordering
Observability and logging2026-04 to 10Monitoring-path failures, log loss, and the delivery canary
Reboot and thermal investigations2026-04 to 10Silent reboots, idle-state tuning, an open cooling fault
Learning experiments and scope2026The old playground and the rule for new experiments
DNS ownership and monitoring capacity2026Stale public DNS and a full Prometheus volume