---
name: parked-thin-lane-delivery-emit-residue
description: "PARKED work owed by todlando — delivery.rs:149,313 → emit_line_err! PLUS BRAIN_PHASE breadcrumbs in brainproc.rs, held until doyle confirms the v0.66.0 source-confirmed cut"
metadata: 
  node_type: memory
  type: project
  originSessionId: 880d1285-2246-4429-9b5e-eb4331af141f
  modified: 2026-08-29T21:31:29.121Z
---

Parked 2026-08-29, waiting on an external ping, so none of it is obvious from the code.

## Item 1 — my own emitter residue (the original lane)

Convert two sites in `crates/spt/src/api/delivery.rs` from bare `eprintln!` to
`spt_proto::emit_line_err!`: line 149 (`IO_SINK_WARN:`) and line 313
(`DISPATCH_RESULTS_WARN:`). Behaviourally inert — both write the same line to stderr.

**Why they are mine:** I minted both by copying a sibling that was still pre-conversion at the
time — `9ec181e9` (W3/#239, `publish_boundary_io` copied `publish_state_io`) and `2e71652a`
(W2/#238). The #241 emitter conversion (`cfb9d9f8`) converted the three sites its snapshot held
and could not know about the two added on top. Full attribution:
`EMISSION-RESIDUAL-CENSUS-RECONCILIATION.md` at repo root.

**Do NOT widen:** `main.rs:144,193` are #243's, not mine — byte-identical at the census cut
`4d6007ac`, so census-truncation misses. Tree-wide split: 340 census-mechanism / 2 mine /
20 test-dir files / 1 inside a test module.

## Item 2 — BRAIN_PHASE breadcrumbs (scope +1, shape APPROVED)

Source: `RCA-242-R2-LINUX.md` at repo root (golden r2 Linux red, ruled a TEST defect). The
`BRAIN_UP` → `brain.ready` window in `crates/spt-daemon/src/brainproc.rs::run_brain` is silent
on the happy path, which cost that RCA two round-trips to bound. Three silent stretches,
confirmed at source:

1. `announce_coordinator_image` — `Ok(true) => {}`, a successful round-trip prints nothing;
2. `resume_session_cursors` — `Ok(_) => {}` is the ALWAYS-TAKEN arm today (no daemon-hosted
   PTY sessions), so the round-trip that happens is the one that never speaks;
3. `write_ready` → `current_exe_hash()` — a `OnceLock` doing `fs::read(exe)` + SHA-256 INLINE
   on first call, before the atomic write.

**Approved shape** (doyle, 2026-08-29): one token family, `emit_line_err!` throughout —
`BRAIN_PHASE:announce ok in {ms}ms` / `BRAIN_PHASE:resume ok ({n} cursors) in {ms}ms` /
`BRAIN_PHASE:ready written in {ms}ms (exe-hash {ms}ms, computed={bool})`. Per-phase ELAPSED ms,
never wall stamps (`CONN_LIFECYCLE` already carries wall_ms; two clocks in one log start
disagreeing). The `computed` flag is load-bearing: the `OnceLock` means only the first brain in
a process pays the hash, so a reader seeing 4ms must be able to tell *fast* from *not this
process's turn*. The three numbers sum to the gap — that is the one-grep requirement.

## Explicitly NOT in this lane

Moving `current_exe_hash()` off the ready path (the honest fix for stretch 3, so a slow
self-read cannot delay readiness). Ruled a separate board request: it touches ADR-0018 Q7's
ratified readiness semantic and `REQ-BRAIN-HASH-ONCE`'s resident-bytes truthfulness contract.
doyle mints it BACKLOG in the #243 filing batch, cited to the RCA. **Do not fold it in.**

## Coordination

hertz owns the TEST half of the same RCA (the false `PRECONDITION` message + the 4 cells'
untuned 30s budget: `activity_link_push_e2e`, `endpoint_autostart_e2e`, `idle_edge_seal_e2e`,
`idle_edge_drain_e2e`). No file overlap. Steer already sent and endorsed by doyle: the fix is
progress-bounded, not a bigger clock (the #235 shape) — r1 passed at 21.5–43.2s against a 30s
budget, i.e. deadline-adjacent even when green.

**Related artifact:** `EMISSION-RESIDUAL-CENSUS-METER.py` at repo root — re-runnable meter,
`python … <sha>`, moved out of a session scratchpad on purpose; #243 wants it when minted.