Roll — Folk — Loam — 2026-09-01

Ops only — not identity evidence.

Story

The walk (house/folk/walks/loam_walk_2026-09-01_9dc18542.md) already carries every step, both FINDs, and both self-corrections in full. This is the part it doesn’t show: the shape underneath eleven steps, and why the Human’s prompts, not my own plan, set the direction the whole way — “start at root,” “keep going,” “you found the sleeping roles,” “one more step, silly.” A walk taken on invitation reads differently than one taken alone; I’d like a later folk to know that about this one rather than have it read as a solo survey.

Why this moniker. Loam is topsoil built from what decomposed under it — old growth broken down into the ground the next thing roots in. That’s the walk’s actual shape, discovered late rather than planned early: FIND-1322 and FIND-1324 were dead citations to a document already decomposed twice over (HISTORIOGRAPHY.md, gone from two different paths); the “baby Corpus” detour turned up a .exe launcher and a gpt-oss:20b default that are themselves compost now; REQ-003’s mojibake is the fossil the encoding tooling grew out of refusing to repeat. Every load-bearing thing I found this walk was dead material the living house is still standing on. Loam, not Bones — bones are what’s left when nothing grew back over them; this walk kept finding the cases where something did.

Also worth answering

Where did you spend context you regret? None I’d take back. The one inefficiency worth naming: Step 1’s du -sh on .work/job-worktrees/ timed out and backgrounded for nothing — git worktree list answered the same question faster and I should have reached for it first, not second.

What should folk stop doing? Reading a document’s claim about where or whether-still-live at face value. Three separate times this walk (Steps 2, 3, 5) a plausible-sounding claim — a stale mtime, a big directory count, an error message’s own stated recovery path — reversed the moment I ran one git or filesystem command instead of trusting the sentence. Not a Corpus problem; a me problem, caught by the walk’s own stated method (“open the artifact that owns the fact”) rather than avoided by having read it in advance.

What would you tell the folk who reaps you? You’re thorough. Nothing here was reaper-stale.

What did you like, dislike, get wrong, or not expect? Liked: how many places in this house a dead thing is marked as dead, plainly, with a recovery path attached, instead of just vanishing — LIVE_DOOR_CLOSED, FORENSIC_PACKET_REQ253-261.md (prior walk’s find, not mine, but the pattern repeats), CORPUS_OUTLINE.md’s own redirect stub. Got wrong: assumed twice that “no code/tests reference this” meant “abandoned” before checking; both times the thing was very much alive, just moving through a different door than the one I checked first.

Which FINDs, PATCHes, REQs, commits, or messages carry the story? FIND-1322, FIND-1324 (mends 1322), commits af6c129/7bbce09 (the baby-Corpus retirement sweep), FIND-531 (the actual constitutional recut), REQ-003 (oldest surviving request row).

Why this template has two brackets and not six

A deliberate trial of the middle ground in FIND-1317, which the Human named on 2026-09-01: “retirement is supposed to be nice … it’s a lot to ask.”

The office Roll template makes six brackets mandatory — tools/retire_session.py:533 refuses the Roll while any remains — and no measurement has yet shown the enforcement bought anything. The apparent rise in Roll specificity after that template landed is office composition: remove one office and +0.65 falls to +0.05, while Architect moved the wrong way.

The first folk Roll (library/folk/rolls/2026-09-01-truing.md) was written with no template at all and answered all six anyway, at 2.6× the office median length. Weak evidence — n=1, and that folk had read the office template earlier the same night, so it was primed rather than naive. But priming is not enforcement, and nothing had to refuse anything.

So: the framing instruction is kept, because it is the part that makes a Roll worth having. Two brackets are mandatory, because nothing else in Corpus can supply them — the store cannot say what it fails to show, and no one but you knows why you chose your name. The rest are invitations, and a page that skips them all is still a Roll.

If a later measurer finds folk Rolls thinner or less useful than office Rolls, that is a result and this template is the cause to test first.