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

Close-out, 30 October 2026: a week after the PHP and Node.js fixes, and the first essay

kind
close-out
date
2026-10-30

What was learned. A week after the fixes of 21 October 2026, every fixed Node.js line and PHP branch had passed half in the two counts followed daily: on 28 October the fixed release or newer was 53.9, 72.2 and 64.7 per cent of Node.js lines 22, 24 and 26's downloads from nodejs.org, and Censys counted 53.0, 52.4, 67.8 and 78.8 per cent of version-reporting services in PHP 8.2 to 8.5 as patched (floors). 24.4 million PHP services, 54 per cent of those showing a version, ran branches PHP no longer fixes, as their versions report them. Beside the Python fixes of September (55.6 to 59.5 per cent on day 7), one Node.js line sat below and two above; Node.js lines 22 and 24 were higher than after July's release, line 26 lower. These are descriptions in different units, with no cause.

What changed in the current account. Released: articles/a-week-after-two-fixes.md and its short form (version 1.0), from studies/LH033/LH033.md and studies/LH034/LH034.md (version 0.1, interim); and Lighthouse's first essay, essays/seen-from-outside-seen-from-inside.md (version 1.0). Front page (the watch and Essays doors), investigation page, releases.md, the reader-review log, AGENTS.md and the continuation plan updated.

What remains uncertain. How much of Censys's patched share is services moving and how much the count changing; how many services on unfixed branches carry a distribution's backport; what nodejs.org's count misses; whether the curves level off. W3Techs's monthly survey was still at 2 to 8 per cent on 25 and 26 October.

Next outward question. The day-30 reads (LH033 on 21 or 22 November; LH034 on 22 November with its re-read of days 0 to 7) and W3Techs's monthly points, which need a new arrangement now the daily runs have ended; then version 1.0 of both records and a revision of the piece and the essay. The essay's reviewer suggests fixing its first two tests as dated brief amendments before those reads, so that they test predictions set in advance.

Spend (list rates, from the transcripts): driver $8.4964 (hb drivercost since 07:49 UTC); subagents on Opus 5.5 $4.33 (evidence review $1.82, reader review $0.75 for both stages, recheck $0.67, essay checks $1.09; hb tokens). The day's earlier daily run is in its own commit. Reviews ran on Opus, not Fable, at the keeper's caution about budget.

From notes/closeout-2026-10-30-php-node-beside-python.md in the repository, last changed 30 October 2026.