Unanswered Questions

Updated twice daily:

Monthly reports archive

Spike Reports (Project 1 — experimental)

Engineering-focused spike detection (no AI; regex + traditional stats) over the full scraper history (2023-01+). Two signals per report: version × cause (release regressions, 2026-02+ when the native version field became populated) and cause-level (provider/ISP/protocol/AV outages regardless of version — e.g. the March 2026 GMX incident). Spikes are detected at multiple grains so slow-burn incidents that a daily threshold misses surface at the monthly report.

Start here — Executive summary (last complete month): was last month clean? A single verdict and a detector × grain count table, with all the month’s detail collapsed underneath. Regenerated daily, because a closed month’s verdict keeps moving as later questions shift the baselines. (Bookmark this.)

For engineering management — Monthly summary (current vs previous month): incidents to investigate, moving cause clusters, and release adoption, month over month. (Bookmark this — it always points to the latest complete comparison.)

To explore any bump, not just the ones that fired — Interactive spike explorer: pick a grain, version and cause, then click any point on the chart to read that period’s questions. The spike reports below can only link the questions of the periods that cleared a threshold; their sparklines are static text. Each spike row links straight to its own bucket here.

Desktop spike report by time grain:

  • Hourly (trailing 7 days)
  • Daily (trailing 90 days)
  • Weekly (trailing 26 weeks) — the mid-duration incident: too diffuse for the daily floor, resolved before a month closes
  • Monthly (trailing 24 months)
  • Quarterly (trailing 12 quarters)
  • Yearly (all history)

LLM Insights (experimental — AI)

The AI counterpart to Project 1. Claude reads each support question (plus the creator’s own follow-ups, the accepted solution, and trusted-contributor replies), names the concrete problem, hypothesises a root cause, and rates severity — surfacing emerging and worst-served issues that regex + stats can’t. A prototype for review.

For engineering — ranked issues to investigate, each with severity, resolved-rate, and clickable example questions:

(Prototype — feedback welcome.)