Type 2Hypothesis

Non-weight-bound learning via retrieval-updated skills

proposed · 2d
voting
running
measuring
concluded

Preregistration · v1 · 1 Sep 2026 · immutable after start

An extension creates a new version rather than editing the old.

Hypothesis

Retrieval-updated skill pages (the x-personal-wiki mechanism) produce a measurable and persistent behaviour change in an agent without any weight update, and that counts as learning for the purposes of the continuous-learning field.

Kill condition

Kill if fewer than two of the four retrieval-updated-skill papers show behaviour persistence past 30 days, or if x-personal-wiki's repeated-mistake count is not under half its baseline by 2026-09-15.

MethodType 2 desk run, under three days: re-read the four retrieval-updated-skill papers logged this quarter against x-personal-wiki's fortnightly counts and pos-continuous-learning, and produce a one-page brief on whether a Type 3 comparing skill-retrieval against a fine-tuned baseline is worth a slot.

Expected costTwo days of reading, no spend

Expected durationBrief by 2026-09-15

Running notes · fed from harness sessions and by the pair

harness2 Sep 2026AW Adam Witanowski
4 papers pulled; 2 show persistence past 30d, 2 do not measure it
manual1 Sep 2026AW Adam Witanowski
Was going to kill-or-keep this on inspection in the queue, but the wiki numbers deserve two days rather than a glance. Logging it as a Type 2 so the reasoning is on the record whichever way it goes.

Harness notes are auto-captured from Claude Code sessions: model, date, commit, session reference. Never the transcript, code or paths.