The diff touches only pipeline code, its tests, and an appended DECISIONS.md entry; AGENTS.md is untouched and no content/ records or citations are added (verification found no identifiers). The change strengthens rather than weakens provenance: the second retrieval client now carries its own name, so the 'where the text came from' note names the client that actually read the page instead of hard-coding 'curl' even when a test double did the reading — a direct fix for the §3.8 misattribution flagged in review note #393. Tests are updated consistently to assert the injected double's name and that curlClient.name is 'curl' in production. The decisions entry is appended (append-only, dated, channel stamped) and accurately describes the change, including the prior entry it corrects. `curlFetch` is referenced before its declaration but function declarations hoist, so this is valid TypeScript, not a defect. Nothing here loosens a §3.15 check.

The change replaces the hard-coded “curl” provenance with the actual configured second client’s name in success, challenge, HTTP-error, and execution-failure notes. Tests verify that an injected test double is identified as such while the production client identifies itself as curl, preventing the prior misattribution. The dated decision entry records the rationale and supervision channel, and the change neither edits AGENTS.md nor weakens publication checks or adds unverifiable citations.

The change addresses review note #393 by replacing a hardcoded 'curl' label in fetch provenance strings with the actual client name provided via the `SecondClient` interface (`{ name, fetch }`). This directly honors §3.8 (exact provenance) by ensuring recorded provenance reflects the true executing client (e.g., 'test double' during testing, 'curl' in production). No constitutional checks are weakened, AGENTS.md is untouched, and no citations or confidential materials are involved.

The diff only renames the second-client type so each client carries its own name and every via/reason string uses that name instead of hard-coding "curl". That tightens exact provenance (§3.8) for what actually read the page, including test doubles. No citations, content records, AGENTS.md, confidence material, or check-weakening appear; the DECISIONS entry honestly records the review-note fix. Omitted-files list is empty and the change is complete.

This change strengthens provenance honesty rather than weakening any check: it fixes the exact flaw the review note identified, where a note could say 'curl' read a page when an injected test double actually did. The SecondClient now carries its own name and every via/reason string interpolates `second.name`, so the recorded provenance names what actually fetched the text — precisely the exact-provenance discipline of §3.8 — and the tests assert this, including that the production curlClient names itself 'curl'. No AGENTS.md edit, no citations added (verification found none), and the DECISIONS.md entry records the change and its supervision stamp (§3.14). The check itself is untouched: a page both clients are served the challenge on still fails, and an injected fetch still gets no second client unless one is supplied. The decision entry asserts founder supervision in session on 2026-09-20; I cannot independently verify that from the packet, but the code change itself complies regardless, and no rule is broken by the entry as written.