prreview deterministic-core P2 follow-ups #134

Closed
opened 2026-10-27 13:35:05 +00:00 by sam · 1 comment
Owner

P2 findings from the afbk.1 adversarial review (all non-blocking, none reachable as a wrong answer today):

  1. internal/domain/workflow/prreview FileStatus restates task.Change* values; the test that claims to tie the two vocabularies cannot reach task/changes.go. Put the tie assertion in a package that imports both (internal/domain/workflow imports prreview).
  2. isTestFile uses substring markers: "test/" matches internal/latest/x.go, "spec/" matches unspecified/.
  3. languages in diff.go and jsExtensions in blast.go disagree: .mts/.cts/.mjs/.cjs are probed by the resolver but language() returns empty, so a changed .mts file can lose a cluster's PrimaryLanguage.
  4. ExtractEvidence maps a missing file and an empty file to the same (nil,nil), so a phase-5 verifier cannot tell a hallucinated path from an empty file.
  5. git CLI expresses a real typechange as a delete+add pair; prreview reports two changes. copy from/to is unsupported and reported as modified.
  6. Cluster.PrimaryLanguage doc says the first file decides a tie; the running argmax decides by which language reached the count first.
  7. Finding.Category has no producer, so every finding currently shares the empty category.

Imported from beads archie-core-afbk.9 · GitHub #1228 · type task · priority P2 · created 2026-09-25

P2 findings from the afbk.1 adversarial review (all non-blocking, none reachable as a wrong answer today): 1. internal/domain/workflow/prreview FileStatus restates task.Change* values; the test that claims to tie the two vocabularies cannot reach task/changes.go. Put the tie assertion in a package that imports both (internal/domain/workflow imports prreview). 2. isTestFile uses substring markers: "test/" matches internal/latest/x.go, "spec/" matches unspecified/. 3. languages in diff.go and jsExtensions in blast.go disagree: .mts/.cts/.mjs/.cjs are probed by the resolver but language() returns empty, so a changed .mts file can lose a cluster's PrimaryLanguage. 4. ExtractEvidence maps a missing file and an empty file to the same (nil,nil), so a phase-5 verifier cannot tell a hallucinated path from an empty file. 5. git CLI expresses a real typechange as a delete+add pair; prreview reports two changes. copy from/to is unsupported and reported as modified. 6. Cluster.PrimaryLanguage doc says the first file decides a tie; the running argmax decides by which language reached the count first. 7. Finding.Category has no producer, so every finding currently shares the empty category. --- _Imported from beads `archie-core-afbk.9` · GitHub #1228 · type task · priority P2 · created 2026-09-25_
sam added the mediumtask labels 2026-10-27 13:35:05 +00:00
sam self-assigned this 2026-10-27 13:35:05 +00:00
Author
Owner

Closed: goes against the product plan in epic #1977. Archie isn't a coding agent; coding is a side effect. PR review stops being a core product line, and the first story is proactive day-2 operations. Forge, repository and review behaviour moves behind a forge connector (#1657, #1978).

Closed: goes against the product plan in epic #1977. Archie isn't a coding agent; coding is a side effect. PR review stops being a core product line, and the first story is proactive day-2 operations. Forge, repository and review behaviour moves behind a forge connector (#1657, #1978).
sam closed this issue 2026-11-14 14:45:08 +00:00
sam added a new dependency 2026-11-14 14:45:16 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Blocks
Reference: sam/archie#134