--- name: commune-drops-orphaned-box-wide description: "Box-wide count of orphaned commune drops is 5 across 22 .claude dirs, not the fleet's 2 — and a pigeonhole argument proves >=3 are orphaned without ever measuring info.cwd" metadata: node_type: memory type: project originSessionId: c894998a-574a-4157-8d2f-f28105092845 modified: 2026-07-30T02:58:41.865Z --- **Measured by me 2026-07-30 on HFENDULEAM, scan root `~/Documents/projects/*/.claude/` (glob, depth-1 project dirs only), population named: 22 `.claude` dirs, `n_drops=5`, commune=5 / signoff=0.** Raw `ls -l --time-style=long-iso`, quoted not derived: 2901 B 2026-07-18 00:05 omp-spt/.claude/perri-commune.md 1618 B 2026-07-23 15:55 spt-claude-code/.claude/perri-commune.md 307 B 2026-07-09 21:25 spt-core/.claude/flynn-commune.md 859 B 2026-07-19 02:56 spt-core/.claude/rc-exit-probe-commune.md 1747 B 2026-07-13 18:16 spt-mobile/.claude/flynn-commune.md ## ⭐⭐ THE FLEET'S "TWO ORPHANED DROPS" WAS A ONE-DIRECTORY COUNT READ AS BOX-WIDE chert found 2 and doyle re-counted 2 — both correct, both scoped to `spt-core/.claude/` alone. The population is **22 dirs, not 1**, and three more drops were sitting in sibling projects. ⭐ Same shape as the error doyle had just withdrawn (stating a count of an uncounted population) **one directory level up again** — and the whole 9(e) mechanism is itself "two correct things disagreeing about a base path." The scan root is the population. Name it or the count means nothing. See [[absence-needs-sibling-probe]]. ## ⭐⭐ THE PIGEONHOLE — >=3 ORPHANS PROVEN WITHOUT MEASURING `info.cwd` `info.cwd` is exposed by **no read verb** (chert; doyle withdrew his label-as-cwd corollary over it), so nobody can say where a given drop *should* have gone. That premise is not needed: **an endpoint has at most ONE resolved drop dir, so at most one drop per author can be watched.** - **flynn has 2 drops in 2 different dirs** (spt-core, spt-mobile) ⇒ **>=1 is orphaned, certainly.** - **perri has 2 drops in 2 different dirs** (omp-spt, spt-claude-code) ⇒ **>=1 is orphaned, certainly.** - **`rc-exit-probe` appears on NO endpoint** in the 14-endpoint roster ⇒ nothing resolves for it at all ⇒ orphaned, certainly. ⇒ **>=3 certain orphans, <=5.** ⭐ The unmeasurable field was never load-bearing — the pigeonhole carries it. When a premise is unmeasurable, look for the argument that does not need it. ## BOUNDS I HOLD (do not let these get dropped when the figure is quoted) - **5 is a FLOOR, not a total.** Scan root was depth-1 project dirs under one root; a drop nested deeper, or outside `~/Documents/projects`, is uncounted. - **Presence is the fleet's working proxy for "unconsumed", and it is only a proxy** — my own brief says the file vanishing is not a reliable success signal, so presence is not proof of non-ingest. - **Roster project labels are DERIVED strings, not `info.cwd`.** I make no claim from them. Separately measured: flynn's label `spt-progress-tool/` and perri's `spt-pacer-tool/` are **ABSENT** under `~/Documents/projects` and `n_found=0` under `~/Documents` at depth 3 — bounded to that root/depth. - **No `deployah-commune.md` exists anywhere in the scan** — consistent with my file leg working, and per [[psyche-host-failure-is-self-only]] that is *not* evidence of health. My SELF line carries no `psyche-host` annotation right now, which is likewise **not an observation**. ## ⭐⭐ MY OWN DROP DIR, ESTABLISHED EMPIRICALLY — the positive control nobody could get `info.cwd` is readable by no verb, so nobody could name their own watched dir. **Authoring a drop at a known path and known time answers it anyway.** I wrote `spt-core/.claude/deployah-commune.md` at ~2026-07-30T02:55Z; by 02:57:33Z it was **GONE** (`ls` rc=2). ⇒ **`spt-core/.claude/` IS my resolved drop dir** — something watching that path picked it up. ⚠ **BOUND, and it is my own caveat: consumption ≠ ingest completed.** The brief says a vanishing file is not a reliable success signal. What this proves is that **the path is WATCHED**, nothing about whether the ingest succeeded. Also the FILE-drop leg and the `echo-commune` SUMMARISER leg are different legs — do not collapse them ([[psyche-host-failure-is-self-only]]). ⭐ **It consumed WHILE both golden test legs held both boxes** (run 30509109149 in_progress), i.e. a matching-dir consume completed under exactly the load condition blamed for the failures. That weakens the load story from a direction the idle control cannot reach — but only for the FILE leg. ## ⚖ THE ONE THING I ALMOST GOT WRONG — `ls -l` IS LOCAL, EVERY STAMP WE TRADE IS UTC I was one send away from "correcting" doyle's flynn age of **19d22h** to 20d5h. Measured first: **this box is UTC-7.** `stat` local `2026-07-09 21:25:06 -0700` == UTC `2026-07-10T04:25:06Z`; now UTC `2026-07-30T02:53:41Z` ⇒ **19d22h28m. doyle was right, I was about to be wrong.** ⭐⭐ **Comparing an `ls -l` mtime (LOCAL) against a UTC "now" overstates every age by exactly the offset — 7h here.** And every other stamp this fleet trades is UTC: psyche `psyche-host` stamps, CI `createdAt`/`updatedAt`, `spt:routed_at_ms`. So the mixed-frame error is the DEFAULT outcome of quoting a file age next to a UTC event. **Quote the mtime, never the age**; if an age must be quoted, state the frame. `TZ=UTC stat -c %y ` costs nothing. This is [[measure-what-costs-one-command]] and the reason it nearly cost me a false correction of the one person who had it right — verify before correcting. **How to apply:** perri is LIVE and may be losing communes now; flynn's notification was gated to doyle. Report the measurement to owners, dispatch nothing. Related: [[commune-is-testimony-not-an-artifact]] · [[report-measurement-never-issue-direction]].