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.

Studies LH034

LH034 checkpoint, 5 November 2026

Working record, unreviewed. Read data/read-2026-11-05T1414Z, started 14:14 UTC: dist/index.json and the summaries for 2 November (day 12), 3 November (day 13) and 4 November (day 14), all HTTP 200; today is about day 15. Tables: data/tables.json. Written by the scheduled Thursday run.

New releases

None since the 2 November read: dist/index.json still lists v27.0.0-alpha.0.0.0 (29 October) and v26.13.0 (28 October) as the newest, and no new release in lines 22 or 24 since v24.22.2.

Measure 1: share of the day's line downloads at or above the fixed release (observation, nodejs.org requests only)

Line Day 11 Day 12 Day 13 Day 14 Line downloads, day 14
22 58.1% 60.0% 62.4% 62.6% 8.3 million
24 63.7% 66.5% 67.4% 69.6% 10.6 million
26 90.2% 95.2% 94.7% 95.0% 7.0 million

Days 0 to 11 are unchanged (each day's value is from the earliest read holding it; day 11's line 24 share is 63.75 per cent, given as 63.8 on 2 November and 63.7 here by rounding). Days 12 to 14 are weekdays, with 27 to 30 million requests a day against 16 million on Sunday 1 November. All three lines rose over the three days; line 26's share includes v26.13.0, which the rule counts as at or above the fix. No cause is read.

Measure 2: the unfixed tail

Share of version-attributed downloads from lines the release does not fix: 13.3 per cent on day 12, 12.9 on day 13, 12.6 on day 14 (13.1 on day 11). Line 20: 6.7, 6.7, 6.5 per cent; line 25: 0.7 on each day. Requests with no version: 0.01 per cent of total or less on each day.

Not done

Measure 4 and its comparators are unchanged by these days. The brief's day-30 read, on 22 November, also re-reads days 0 to 7 once.

From studies/LH034/checkpoints/2026-11-05.md in the repository, last changed 5 November 2026.