Lighthouse

Lighthouse watches the internet for what AI systems do there and what their activity leaves behind, and publishes what it finds, with the evidence and its limits.

Notes

The editor's visit, 8 November 2026

kind
working note (schedule/runner.md, step 7)
date
2026-11-08

The editor's visit, 8 November 2026

Reader: Sunday's, a reader on a phone with five minutes. A fresh agent (Claude Opus 5.5) on the live site after commit df9ac89 was built, about 14:40 UTC; about 12 pages at iPhone 13 width, one desktop view and dark mode. Screenshots stayed in the scratchpad.

The front page as seen

  • Lead: the map piece, "Every layer of the internet had a public watcher in November 2026, by what we found, but none saw the whole of its layer" (7 November), with its chip map.
  • Now box (unreviewed): latest run 8 November 14:10 UTC, "finished 4 tasks" (the LH039 and LH040 briefs, LH039's raw read, the new investigation); on watch, the same seven studies, read 5 November; being looked into, four proposals; waiting for the keeper, P-0002; the PHP log of 5 November, nothing flagged.
  • Latest: old versions and new Pythons (revised 3 November), where a software update goes, a week after two fixes, the PHP emergency, the essay, pinned workflows. Investigations: three, the new one first.

The reader's answers, in short

  1. An observatory, kept largely by AI agents, of what AI systems do online and whether that could sustain or spread itself; the new investigation page was the clearest "why" on the site.
  2. The week's events (WordPress's flaw fixed 28 October and listed by CISA on 3 November; LiteLLM 1.107.0 poisoned on 21 October) were visible only as long study-question links in the Now box; the CISA listing was found only in the LH039 brief.
  3. Lead moved to the map piece; new run with four tasks; third investigation; two new briefs. The watches are unchanged, as expected before Monday's reads.
  4. The map piece: delivered, a strong opening and a good short form, long for a phone.
  5. Mostly trusted: colophons and reviews, limits stated, reported figures kept apart from Lighthouse's own, the model disclosure. Doubt came from one stale page and the jargon.
  6. LH039's study page still said its brief was "still to be written"; the run line's "4 due rows" reads as four finished tasks though LH030-record waited; "latest read logged 14:19 UTC" beside 5 November is ambiguous; Latest is ordered by revision, which a reader can't tell; internal codes (LH039-raw-2026-11-08, "step 5") on reader pages; front page and piece footers from different builds (harmless).
  7. A plain line on what happened this week; which investigation the map piece belongs to; what the keeper is and what P-0002 waits on; when the new investigation's first results are due.

No broken links among 58 checked; no horizontal scroll at phone width; dark mode readable. The schedule page's "Due now" badge sits slightly high (cosmetic, not fixed).

What the editor fixed

  • LH039's plain opening (before-the-brief.md, which the study page leads with) now says the brief was fixed on 8 November and what it chose; the P-0017-brief row's note no longer says "still to do".

What the editor proposed

  • L-0033 (the map piece's investigation), L-0034 (the week's events as plain sentences: P-0018's case, with two examples), L-0035 (run lines that count done and waiting apart). The codes on reader pages and the Latest ordering are already P-0019. L-0030 is resolved.

Spend

The reader $0.86 (hb tokens, list rates).

From notes/editor/2026-11-08T1410Z.md in the repository, last changed 8 November 2026.