Treat truncated sanitized review output as unavailable

When a delegated native review requires complete sanitized output, a successful native exit is not sufficient: if the sanitized result exceeds the agent/tool return boundary and is truncated, the only safe verdict is unavailable. Prevent this by designing a lossless bounded transport (for example, a trusted size check plus chunked sanitized return) before launch; never reinterpret partial findings as clean. This applies to workflows whose contract requires complete reviewer output, and it does not imply the reviewed code is defective or that the native reviewer failed.

Evidence/example — 2026-09-10

During a read-only repository review, the pinned native command exited successfully, but its post-redaction output was truncated at the delegated return boundary. The wrapper correctly returned REVIEW_UNAVAILABLE, released its memory-only capture, and reported no findings. Source provenance: delegated reviewer execution status and the repository's review-output completeness contract.