The reading surface doubles its own error prefix: the CLI layer prepends 'reading: ' and the locator's errors already carry it, so an operator sees 'abcd: reading: reading: agents/...' on a refusal. The path this appears on is the definition-refusal path spc-62 introduced as its own proof of wiring, so the first thing the new surface shows a human on failure is a stutter.

The reading surface doubles its own error prefix: the CLI layer prepends 'reading: ' and the locator's errors already carry it, so an operator sees 'abcd: reading: reading: agents/...' on a refusal. The path this appears on is the definition-refusal path spc-62 introduced as its own proof of wiring, so the first thing the new surface shows a human on failure is a stutter.

Half closed on 2026-08-31 by itd-185, and the record stays open for the half that is not. The ingest verb's refusals no longer stutter: its surface trims the prefix the core already carries. The assemble verb still prepends unconditionally at internal/surface/cli/reading.go:49 while the core's errors carry the same prefix, so abcd reading assemble still renders abcd: reading: reading: ... on a refusal. It was left alone deliberately: it is a shipped verb's output rather than the new one's, and changing it belongs with whoever is next in that file.

Grounds

• pursued: one rule for the whole sub-tree removes the stutter everywhere rather than verb by verb; a surface test per sub-verb, text and --json alike, asserting exactly one tag on a CORE refusal would show it wrong — a row that still renders 'reading: reading: ', or a future sub-verb added without going through readingRefusal()