Verify behavioural equivalence against a live mailbox

[jbagsik]

Parent

#1

What to build

The acceptance gate for the whole extraction: prove the refactor changed no behaviour.

The unit suite cannot establish this, because the thing being verified is equivalence with a running system rather than conformance to a written expectation. Run the extracted stack against a real mailbox alongside the pre-extraction behaviour and compare.

This is a verification ticket, not a coding ticket. Its output is a written result, and — if anything diverges — a bug ticket rather than a quiet fix.

Acceptance criteria

• Delta synchronisation resumes identically: the same messages are picked up, in the same order, with no duplicates and no gaps across several polling cycles
• Folder movements are identical for every outcome, including the skipped case
• Status transitions on messages and attachments match the pre-extraction values for the same inputs
• Deduplication behaves identically for a message delivered twice
• Stored external identifiers are stable across folder moves — the immutable-identifier regression is specifically checked, not assumed
• An initial catch-up spanning several runs completes with the same result as before
• The comparison and its result are written up; any divergence is filed as its own ticket instead of being fixed inline

Blocked by

• Remove Graph code from mail-inbox and verify the dependency is gone #7 — the old code must be gone, otherwise it is unclear which path was exercised