[imgoose]

Problem and change

The Codex runtime SQLite query already loads at most 200 matching rows, but collection stopped after 30 admitted entries. Thirty newer normal notices could therefore hide an older readable failure inside that same bounded result, before the failure-priority selector ran.

Process the existing bounded candidates before applying the existing failure-priority output selector in both shipped helpers. Keep the 30-record / 128 KiB complete-message output budget, source/thread/content-echo filters, redaction and original-attempt uncertainty. Report the SQLite source limit only when the 200-row query bound is reached; output omissions remain reported by the shared selector.

Validation

• New regression: before 2/18 passed, after 18/18 passed. Covers 29/30/31/199/200 newer notices, the unchanged SQL bound, privacy filters, complete-error byte limits, offline saving, simulated certificate upload failure, and UTF-8 fragment SHA-256 reconstruction for both packaged helpers. External networking is blocked during the regression.
• Windows complete suite: 130/135 passed. The unchanged main baseline reproduces the same five failures: one CRLF assertion and four pre-existing local-media filename fixtures containing ?, which cannot be created on Windows.
• Read-only composition with PR fix: preserve structured MCP failure evidence and macOS App logs #35 head 003c0eaf93cf1ed3f5de45bc360774ce0caf4c5a: 132/136 Windows full-suite checks passed; its existing CRLF fix removes that baseline failure, leaving only the four media filename fixtures. All 94 existing collection checks and all 18 new regressions pass in this composition.
• Ubuntu Node 24 full suite: 199/199 on both push run 40705335492 and pull_request run 40705790812, exact head 80e4a96a8f8d7f304aad76155777aba8c8fe7358; all CodeQL checks passed.

Release scope

Three files only: the two helper mirrors and the regression file. No package/version bump, backend deployment, Desktop change, host configuration change or Doctor work. PR #35 remains unchanged; this fix can ship in the same final collector batch after normal review/merge. Controlled validation is not evidence of natural adoption, original-user recovery or final Axiom storage. No production diagnostic report was posted during this development validation.