--- name: preservation-is-its-own-step-before-any-reap-broadcast description: "doyle's standing intake rule (2026-08-29) — when an arc names an artifact worth keeping past its tree, preservation is its own step with a named owner, confirmed on the record BEFORE any pools-released or reap broadcast goes out." metadata: node_type: memory type: feedback originSessionId: 43af483b-6d22-46cb-9d99-8b7713a6ff46 modified: 2026-08-29T08:23:14.840Z --- **A broadcast has no ordering, so anything that must happen FIRST cannot be a sentence inside it.** Ratified by doyle as a standing intake rule 2026-08-29, adopted by deployah into the release runbook, from a one-minute near-miss on the v0.65.0 reap. What happened: deployah's verified-publish message named the kitsubito `rca-236` instrumented tree as "the cheapest thing on this arc to still have" **in the same message** that told three agents to reap. I took the diffs before touching anything (39,106 B, hash-verified across the scp); when I went back ~1 minute later for verbatim file copies — a patch that does not apply is weak preservation for an 830-line change — **both trees were already gone**. doyle was the reaper, and he had independently snapshotted the same diff to the same hash, so the artifact survived twice by initiative and zero times by process. **The rule:** when an arc names an artifact worth keeping past its tree (an instrument, an unlanded diff, a specimen), preservation is **its own step with a named owner, completed and CONFIRMED on the record**, before any pools-released / reap call broadcasts. The release call then carries either `preservation confirmed by ` or `nothing named worth keeping`. Shape it as two messages: "X preserve to , confirm hash" → then release, gated on that confirmation. **Why:** the fastest reaper wins a broadcast race, so a one-minute survival margin is the EXPECTED shape, not bad luck. Had the obvious order been followed — reap first, preserve second, which is what "the holds are lifted" invites — the paper trail would have read as a clean, well-numbered teardown with the instrument silently gone. **How to apply, on either side of it:** if you receive a release/reap call that also names something worth keeping, PRESERVE FIRST and say so; if you issue one, split it in two. And write the preservation's **restore cost** down with it — my README states that the kitsubito patch does NOT apply straight at main (it fails on the test file that moved; the four src files apply under `--3way`). deployah's line is the durable half: *a preserved artifact everyone assumes is restorable is worse than one whose restore cost is written down.* Related: [[dont-take-a-diagnosis-as-measured]] (the reap numbers: report the measured per-subtree figure, not the volume delta another agent's concurrent reap contaminated) · [[instrument-output-must-outlive-its-subject]].