FileOperationRecord

Whether the tool itself reported success.

This is separate from status, which also becomes error when a successful write or edit cannot be read back for diffing.

Change counts for diff, taken before it was truncated for display.

The single provenance for what a DiffMessage shows. None is the only way this says "unknown" — never DiffStats(0, 0), which means a verified zero. Unset until a diff is computed, and left unset for both outcomes that have nothing to count: untrusted_before, where any count would be fiction, and unreadable_after, which returns before a diff is ever computed.

Deliberately not metrics.lines_added/lines_removed, which are session accounting and do not always mean diff lines: a new-file write_file sets lines_added from the whole file rather than from a diff.

What this operation can honestly say about what it changed.

See DiffOutcome. Consumers should branch on this rather than infer the state from status, diff, or diff_stats.

Why the post-operation read failed, as reported by the backend or OS.

The payload for diff_outcome == "unreadable_after", and set only with it. Carried separately from error, which on that path holds a fixed caller-facing summary (elsewhere it holds the tool's own output). Without this the user is told only that the content could not be read, which restates the problem instead of explaining it.