AI collapsed "when?" to "now." Quality still hangs on "why?" and "how?"

QA automation engineer. Most of my work is test automation; lately it's tooling for AI coding agents. They turn out to be the same problem: neither gives you a deterministic system to assert against.

Python · one line below contains a real defect · +10 points for a confirmed report

 1 | def dedupe_keep_newest(events):
 2 |     """Keep only the newest event per id."""
 3 |     if not events:
 4 |         return []
 5 |     seen = {}
 6 |     for e in events:
 7 |         prev = seen.get(e["id"])
 8 |         if prev is None or e["ts"] > prev["ts"]:
 9 |             seen[e["id"]] = e
10 |     ordered = sorted(seen.values(), key=lambda x: x["ts"])
11 |     return ordered[: len(seen) - 1]

File a bug report against the line you suspect:

L1 · L2 · L3 · L4 · L5 · L6 · L7 · L8 · L9 · L10 · L11

Each link opens a pre-filled GitHub issue. The triage bot judges it, comments, and deploys the next round. Struck-out lines were already reported and closed works-as-designed.

No confirmed reports yet. The board is yours to open.

_{Powered by GitHub Issues: every guess is a bug report and every verdict is a bot comment. Answers are encrypted in the repo, so reading the source won't help you. The maintainer plays for free and scores nothing.}