Skip to content

Commit ad67aa2

Browse files
MiniMaxMiniMax
authored andcommitted
docs(backlog): §59 as watchlist-only (parked, needs ≥3 Mode-C runs)
§59 = 'Watchlist: Patch-mismatch hardening for the normal solve path'. Priority 4, labels kind/watchlist + theme/solver + area/prompt. Status: WATCHLIST ONLY — not an active backlog commitment. No code or architecture work until the activation trigger (≥3 Mode-C patch-mismatch runs on the normal solve path, ideally across different files) is met. Rationale: - The Mode-C failure (worker produces a patch that git apply rejects because it does not match the current working tree) is the same patch-mismatch symptom §56 addressed for --rework-pr. - Before §57, Mode-C on the normal solve path could produce regression-PRs (PR #441 vs. PR #439). After §57, Mode-C is a clean failure with no PR — acceptable behavior. - A §59 fix would turn these clean failures into clean successes, but the architectural cost (prompt anchoring, per-file git apply --check, optional re-prompt loop) is only worth it if the failure pattern is systematic. One data point (Issue #389 re-run, scripts/model_selection.py:52, 1 patch, 0 applied, no PR) is not enough to justify the work. - When a Mode-C failure appears on the normal solve path, log the file + issue + date here so we can see when the threshold is met. Non-goal while parked: no code changes. §57 + §58 are sufficient to keep the pipeline correct.
1 parent 198a007 commit ad67aa2

1 file changed

Lines changed: 62 additions & 0 deletions

File tree

docs/BACKLOG/open.md

Lines changed: 62 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -380,3 +380,65 @@ that would also reintroduce a recently-removed pattern should be
380380
caught earlier.
381381

382382
---
383+
384+
## 59. Watchlist: Patch-mismatch hardening for the normal solve path (2026-06-25)
385+
386+
Labels: `kind/watchlist`, `theme/solver`, `area/prompt`, `priority/4`
387+
388+
Priority: `4` (parked — **do not activate** without evidence)
389+
390+
**Status: WATCHLIST ONLY.** This item exists so we do not lose
391+
track of a possible quality follow-up, but it is **not** an active
392+
backlog commitment. Do not invest in a fix until the activation
393+
trigger below is met.
394+
395+
**The Mode-C failure mode on the normal solve path** (the same
396+
patch-mismatch symptom §56 addressed for `--rework-pr`):
397+
398+
- Worker produces a patch JSON
399+
- The patch references file content that does not match the current
400+
working tree (file moved, lines shifted, surrounding code changed
401+
since the model's training cutoff)
402+
- `git apply` rejects the patch
403+
- Worker reports failure → §57 now correctly stops the run before
404+
any PR is created
405+
406+
This used to be a silent regression (PR #441). After §57 + §58 it is
407+
a clean failure with no PR — acceptable behavior. A §59 fix would
408+
turn these clean failures into clean successes, but the bar is
409+
"is this worth the architecture work?", and a single data point is
410+
not enough to answer that.
411+
412+
**Current data point (1 of ≥3 needed):**
413+
414+
| Run | Date | Repo | Issue | Affected file | Failure | Result |
415+
|-----|------|------|-------|---------------|---------|--------|
416+
| #389 re-run | 2026-06-25 | ai-issue-solver | #389 | `scripts/model_selection.py:52` | `git apply` rejected | `nonzero_without_changes`, no PR ✅ |
417+
418+
**Activation trigger:** ≥3 Mode-C patch-mismatch runs on the
419+
normal solve path, ideally across **different** files (so we know
420+
it is a systematic prompt/model issue, not a one-off file-specific
421+
problem). Until that threshold is met, §59 stays parked.
422+
423+
**Non-goal (while parked):** no code changes, no architecture work.
424+
§57 + §58 are sufficient to keep the pipeline correct.
425+
426+
**Scope when activated:**
427+
428+
- prompt anchoring on the normal solve path (similar to §56's
429+
rework-pr fix: explicit file-version context, current branch tip
430+
SHA, recently-touched files in the issue scope)
431+
- per-file `git apply --check` before declaring application success
432+
- optional targeted re-prompting loop if `git apply --check` fails
433+
(model retries with the failure context instead of bailing out
434+
entirely)
435+
436+
**Touches (when activated):** `scripts/solve_issues.py`,
437+
`scripts/validation/rework.py`, `workers/openrouter_worker.py`,
438+
tests for the patch-mismatch path on the normal solve flow.
439+
440+
**Tracking note:** when a Mode-C failure appears on the normal
441+
solve path, log it here (file, issue, date, error message). Two
442+
more data points move this item from watchlist to active backlog.
443+
444+
---

0 commit comments

Comments
 (0)