You asked for notes on non-Swift runs, so here is one. Two real runs of bug-echo (v1.7.0, 61ec226) on a private Python project — ~48 files, nine parallel "adapter" modules that each pull the same kind of data from a different third-party source, so it is a good sibling-hunting target.
The fix that seeded it. A sort key return (1, item.order, item.title) was removed because item.order was not what it claimed to be: it came from order=position assigned while enumerating a listing that arrived newest-first, so "order" silently meant "reverse chronological index".
What bug-echo did.
- The removed line was not self-sufficient, and the precondition walk resolved it in one step (
item.order ← enumeration index ← unsorted listing). Primary ingredient: order=<enumeration index>; precondition: "listing not sorted by a meaningful key first". That is exactly the split the skill describes, and it worked unchanged on Python.
- 12 candidates across the nine adapters → 0 BUG, 12 OK, 3 CANON (three adapters already sort by date or let the server order results, and were correctly reported as the reference implementations). Verdict: localized fix.
- A second, independent run of the same fix later that day (fresh session) agreed on 0 BUG / localized, but additionally rated one site WATCH (the shared data model's
order relies on the adapter flipping the listing; a one-line comment closes it) and one REVIEW (crawl-discovery order, stability unclear). Two runs agreeing on the BUG column while differing on WATCH/REVIEW seems like the expected spread for judgment labels.
What needed changing for Python. Nothing Swift-specific surfaced. Two small observations:
- With 12 candidates and zero findings, the report-sizing rule put a clean run into the full-report bucket. Locally I added "all candidates OK after classification → inline summary, no report file", which felt right for a zero-finding run.
- HEAD was a chore commit, so the fix had to be pointed at explicitly. That path worked fine — just worth keeping easy, since on non-Swift repos with CI-driven commit noise it will be common.
Thanks for building it. Happy to answer questions about the run; I cannot share the code itself.
You asked for notes on non-Swift runs, so here is one. Two real runs of bug-echo (v1.7.0,
61ec226) on a private Python project — ~48 files, nine parallel "adapter" modules that each pull the same kind of data from a different third-party source, so it is a good sibling-hunting target.The fix that seeded it. A sort key
return (1, item.order, item.title)was removed becauseitem.orderwas not what it claimed to be: it came fromorder=positionassigned while enumerating a listing that arrived newest-first, so "order" silently meant "reverse chronological index".What bug-echo did.
item.order← enumeration index ← unsorted listing). Primary ingredient:order=<enumeration index>; precondition: "listing not sorted by a meaningful key first". That is exactly the split the skill describes, and it worked unchanged on Python.orderrelies on the adapter flipping the listing; a one-line comment closes it) and one REVIEW (crawl-discovery order, stability unclear). Two runs agreeing on the BUG column while differing on WATCH/REVIEW seems like the expected spread for judgment labels.What needed changing for Python. Nothing Swift-specific surfaced. Two small observations:
Thanks for building it. Happy to answer questions about the run; I cannot share the code itself.