TESTING_DEFICIENCIES

Feedback filtering regressions (#17)

Missing boundary-condition coverage for --since

• We had tests that covered filtering strictly after a timestamp, but not the inclusive boundary case.
• Result: timestamp > threshold slipped through even though user-facing semantics needed >=.
• Missing test layer: end-to-end CLI behavior test for exact-boundary timestamps.
• Follow-up: add systematic boundary tests for all supported --since forms (unix ts, relative duration, RFC3339, last).

Missing end-to-end coverage for historical tree drift

• We did not test feedback export after the reviewed block stopped linking to the current worktree tree.
• Result: recent/historical reviews could silently disappear from trueflow feedback even though they were still in reviews.jsonl.
• Missing test layer: end-to-end behavior tests where a review is recorded on one revision and the file changes afterward.
• Follow-up: every history-sensitive export path should have at least one “tree drift after review” e2e test.

Missing cursor-granularity tests for --since last

• We did not test multiple records created at the same timestamp boundary.
• Result: a timestamp-only cursor could drop same-second records or force awkward semantics.
• Missing test layer: end-to-end cursor tests covering same-second additions and repeat exports.
• Follow-up: test cursor behavior at identical timestamps and across repeated invocations.

Missing record-centric range semantics tests for rev:A..B

• Existing tests exercised historical content loading, but not whether feedback records were filtered by repo_ref.revision membership in the requested range.
• Result: out-of-range records on matching paths could leak into exports, while in-range records could be omitted after later tree drift.
• We also missed the case where a review happened at an in-range revision on a file that was not itself changed by that range, which let diff-path filtering stay incorrectly coupled to revision-range feedback export.
• Missing test layer: end-to-end behavior tests asserting exact record inclusion/exclusion by revision range.
• Follow-up: for every revision-scoped export path, test both:
  • in-range vs out-of-range record membership
  • later-drift historical context resolution
  • in-range reviews on unchanged files

Synthetic test records were too unrealistic in key places

• Several tests used fabricated records without realistic path_hint, line_hint, or repo_ref.revision values.
• Result: path/revision-sensitive behavior was under-specified and easier to get wrong without tests failing.
• Missing test layer: fixture helpers that default to realistic record shapes, plus e2e tests that prefer real CLI-produced records where practical.
• Follow-up: improve test helpers so historical/export tests naturally include realistic metadata unless a test explicitly opts out.