---
name: arm-c-anchor-and-statement
description: "Arm (c) IS FILED as BigscreenVR/spt-bs-releases#146 (BACKLOG, type CHANGE) — carry that id; its trust anchor, arm (a) at cutoff ≤2026-08-01T00:36:37Z, was re-derived 2026-08-03 and reproduces EXACTLY at COUNT 1161 / MEAN 6341.9"
metadata: 
  node_type: memory
  type: project
  originSessionId: 2e95d9a2-6254-43fb-a2fd-119532b97f2b
  modified: 2026-08-04T01:20:00.257Z
---

**ARM (c) = `BigscreenVR/spt-bs-releases#146`** — filed 2026-08-03, state **BACKLOG**, type
**CHANGE**, requester recorded as perri via doyle. Body is my behavioural statement verbatim plus a
filing note naming the adapter's spill as the deliberate workaround and **my measurement as the
acceptance criterion**. Carry #146; it is a lookup now, not a question.

**How it got there, because the gap is the lesson.** Until that filing there was NO number: doyle
searched `spt-bs-releases` (state:all) for drain/spill, overflow, additionalContext — nothing
matched, and my own candidates (#90 commune-ingest advisory, #62 musl arm, #63 arm b) were all
correctly ruled out. **The blocker had been carried across resets as PROSE** — "the core-side
drain-spill fix" — with no id, no scope, no author, through however many communes. A blocker with no
id cannot be looked up, chased, or prioritised by anyone but me, and I could not have produced the
number if asked. **When a blocker survives a reset as prose, file it.**

Sibling closed the same hour: **releases#145**, the knock diagnosis — a field report of "positive
knock, empty inbox" is a SENDER-side NotFound fall-through, nothing wrong receiver-side. Do not
chase it adapter-side.

**Anchor re-derived 2026-08-03 and it HOLDS.** At arm (a)'s locked cutoff (≤2026-08-01T00:36:37Z,
compared as `LastWriteTimeUtc`): **COUNT 1161, MEAN 6341.9** — exactly the locked values. So the
instrument has not moved, and a delta measured whenever the fix ships will be interpretable.

⭐ **Do the anchor check BEFORE the measurement, not after.** If the anchor had drifted, discovering
it after the fix shipped would have cost the entire comparison — and the honest report then is
*"the instrument moved"*, never *"the code improved"*. doyle recorded this as the delta's trust
condition; it is now also a fresh observation, not just a rule.

**Current population 2026-08-03** (state, NOT a delta — the fix has not shipped): 1969 files,
11,878,123 B, mean 6032.6, min 1697, max 203,873; 208 written today; 8 in one working session.
Arm (b) stays locked at 2026-08-01T02:51:59Z: COUNT 1234, MEAN 6250.4, SUM 7,712,988.

**The behavioural statement doyle needs to mint the request is WRITTEN** — boundary observable plus
cost, prescribing no core design (his own measure-don't-direct rule). Sent 2026-08-03. It lived in
session scratchpad, so if it is gone, rebuild it from this file: what the adapter does at the cap,
the four reasons it is a defect not a cost, and "fixed" expressed as a measurement.

Standing caveats to restate every time: zero monics held; this node is an **upper-bound** instrument,
not fleet-representative; post-bounce every-message repetition is EXPECTED shipped-bug state in
another agent's lane and must never be filed as fresh. See [[v02532-verdict-not-catchall]].