# Investigations

Time-stamped incident reports and experiments. One entry per incident; the entry is the canonical source. Per-node changelog entries link back here.

## Index

| Date         | Title                                                              | Status        |
| ------------ | ------------------------------------------------------------------ | ------------- |
| 2026-04-21   | [Hubris crash loop — thermal + USB drive](2026-04-21-hubris-crash-loop.md) | Drive removal A/B test passing as of 2026-04-28 (3+ days uptime) |

## Conventions

- File name: `YYYY-MM-DD-<slug>.md`. Use the *first* date if the incident spans multiple days.
- Mandatory sections: Summary, Timeline, Root cause, Mitigations applied, Open questions.
- Update the entry as the situation evolves; never rewrite history. Add new dated sections at the bottom.
- Link back from every node's changelog that's affected.