# IR-69 close-out measurement — census blind spot LIVE; population NOT closed (doyle, 2026-08-29 ~20:30Z)

Measured at the assembled/landed sha `ec6da9b0` (== main), during the r2 golden window.
This is the close-out re-census IR-69 prescribes, run with an independent meter; it
FOUND what the entry's own hazard list predicted.

## Predicate (stated with every figure)

`eprintln!("<UPPERCASE-TOKEN><: or space>` at the format-literal start, shipping code
only — classification by cfg(test)-module CONTAINMENT (brace-tracked ranges, the
census's own corrected method), not by line grep alone. This is my meter, close to but
not identical to the sealed census's TOKEN-colon predicate; reconciliation with the
census authors invited below.

## Findings

1. **`crates/spt/src/cli.rs`: 338 TOKEN-shaped `eprintln!` sites remain, ALL SHIPPING**
   (0 of 338 inside any of the 8 module-level `#[cfg(test)]` ranges). First site line
   3945, last ~24251. Sample tokens: `INVALID_ID:`, `ENDPOINT_CREATE_RESERVED_ID:`,
   `ENDPOINT_AUTOSTART_CLEARED:`, `RC_FAIL:`, `ENDPOINT_RUN:`, `PURGED:`.
   RECONCILED (todlando independent meter, 2026-08-29): every figure confirmed;
   separator split is **332 colon-form + 6 space-form**, so 332 are misses under the
   sealed census's own TOKEN-colon predicate, not merely under my wider one. Range
   count corrected 7 → 8: this doc originally missed the trailing module (column-0
   `#[cfg(test)]` at 24391, running to EOF) — re-measured, 8 confirmed. Inert to the
   conclusion (all 338 sit below 24391), recorded so no live meter difference stands
   between the two records.
2. **Mechanism, confirmed structurally:** the census generator truncated cli.rs at its
   FIRST module-opening `#[cfg(test)]` — line 3007 (`mod reserved_id_tests`, closes at
   3164) — and never scanned lines 3164–39118. Proof: zero token-shaped sites exist
   before 3007; census credits the whole spt crate 194 sites (all in api/* etc.; the
   ~12 cli.rs sites it did see and the lane converted were pre-3007 — consistent).
   This is the census's OWN third-face truncation hazard ("truncating at the first
   `#[cfg(test)]` drops later shipping code") — the fix (module-opening check) narrowed
   the trigger but kept the blind spot for files where a test module CLOSES and
   shipping code resumes. IR-8 shape: a plausible census whose blind spot returns a
   clean answer.
3. **`crates/spt-daemon/src/daemon.rs:2768`** `eprintln!("SELECTION_PROBE {}", …)` —
   RETRACTED AS RESIDUAL (todlando 2026-08-29, re-verified on my meter): the site sits
   INSIDE `#[cfg(test)] mod tests` (opens 2291, brace-closes at EOF 3515). By the
   census's own containment method it is test code — out of scope for a shipping
   predicate entirely, not an unverified space-family member. This doc's original
   listing was the scope error; no artifact reconciliation owed.
4. **`crates/spt/src/api/delivery.rs:149,313`** (`IO_SINK_WARN:`, `DISPATCH_RESULTS_WARN:`)
   + `main.rs` (2 sites): SETTLED (todlando, 2026-08-29) — post-cut-minted read
   CONFIRMED, author todlando. At cut `4d6007ac` delivery.rs held three such sites
   (111, 182, 604); census saw them, `cfb9d9f8` converted them. His W3 `9ec181e9`
   added `IO_SINK_WARN:` and W2 `2e71652a` added `DISPATCH_RESULTS_WARN:`, each by
   copying a sibling that was still pre-conversion at copy time. Not census blindness,
   not general churn: two lanes copying a pre-conversion sibling — the same seam
   pick-audit flagged on the delivery.rs busy arm.
   ⚠ METHOD TRAP (todlando): `git log -S'<TOKEN>' -- <file>` here returns pre-cut
   ancestor commits — reading as "census miss" — because `-S` dates when the STRING
   first appeared in the file, not when the SITE did; a duplicated pattern returns the
   original every time. Settled by diffing the minting commits and reading the file at
   `cfb9d9f8` with both forms side by side. The "check `git log -S`" instruction this
   paragraph originally carried was itself the trap.

## What this does NOT change

- The shipped #241 tree converts everything the census saw: chain removed 495
  `eprintln!` ≈ census 486 + space family — the censused population IS closed, and the
  golden green stands. No stop-ship, no product implication for the landed conversions.
- The commit subject "closing the population" is true of the SEEN population only.

## Dispositions (doyle ruling, pending author reconciliation)

- IR-69 does NOT close at this sweep. Addendum: this finding; entry stays OPEN.
- Residual conversion = NEW board request (follow-on wave, operator triages), scoped
  **332 colon-form + 6 space-form** (todlando's split adopted), not a flat 338 — the
  332 are misses under the sealed census's own predicate and say so in the request.
  Generator fix owes a measurement of what module-close resumption it can now
  under-scan (the entry's own rule: every widening/narrowing owes a measurement of
  its new blind spot).
- Post-cut-minted sites (delivery.rs face) → the durable fix is ENFORCEMENT at the
  seam (clippy lint / xtask check denying bare `eprintln!` TOKEN-shapes in shipping
  code), not another snapshot census — §4's settled attribution is the argument: a
  snapshot cannot see a site that does not exist yet; a build-time deny would have
  caught both minted sites. Candidate rider alongside IR-58.
- **delivery.rs:149,313 conversion — RULED (doyle, 2026-08-29): todlando converts his
  own two sites as a THIN LANE, dispatched AFTER the v0.66.0 cut is source-confirmed**
  (main is frozen for r2 parity — tag == main == r2-tested `e4f52047`; the sites are
  behaviourally inert, `emit_line_err!` and `eprintln!` write the same stderr line, so
  nothing rides the release for them). Not folded into the residual wave: keeps #243
  scoped to the census-mechanism population, and the author converting his own residue
  while context is fresh is the cheap path. main.rs's 2 sites SETTLED as #243's
  (todlando tree-wide census: byte-identical at cut `4d6007ac`, behind main.rs's first
  module-opening cfg(test) at line 26 — same truncation mechanism as cli.rs, not his
  mint). His thin lane converts delivery.rs:149,313 only.
  **RESOLVED 2026-08-30 (todlando flag, doyle re-measured): the conversion ALREADY
  SHIPPED — it rode his breadcrumb branch (60a12056) into the r4 head under the
  "riding costs nothing" authorization and is in the v0.66.0 tag. Zero bare TOKEN
  `eprintln!` in delivery.rs at both d931dd63 and origin/main. No lane owed; the
  thin-lane dispatch this ruling scheduled never needed to fire.**

## Tree-wide resolution (todlando meter, origin/main, 2026-08-29 — no remainder)

362 shipping-classified TOKEN-shaped sites total:
- **340** census-truncation mechanism → #243 (cli.rs 338 = 332 colon + 6 space;
  main.rs 2). Exactly TWO files tree-wide have the triggering property (first
  module-opening cfg(test) precedes shipping code) — that is the generator fix's
  blind-spot measurement, answered.
- **2** post-cut mints (delivery.rs:149,313) → todlando thin lane, post-cut.
- **20** `crates/*/tests/**` integration files/fixtures — out of scope for a shipping
  predicate (his classifier reads whole tests/ files as shipping; meter blind spot
  DECLARED, stripped by path).
- **1** daemon.rs SELECTION_PROBE — test module, out of scope (finding §3 retraction).
Re-runnable meter: `EMISSION-RESIDUAL-CENSUS-METER.py` at repo root (untracked;
version-control rides #243's own lane if wanted). Usage:
`python EMISSION-RESIDUAL-CENSUS-METER.py <sha>` (default origin/main) — reads blobs
from the sha, never the worktree; reproduces the table (re-verified from that path).
Declared limits (also in its docstring): cfg(test) MODULES only (tests/-dir files read
as shipping — the 20); uppercase token at literal start + `:`/space only (lowercase or
mid-literal invisible); counts sites, not call paths (dead code counts).