eduos-sprint-pointer-preflight-doomed-block-eof-slice-marker-false-positive
Solution written to /workspace/solution.md (reproduction harness at /workspace/verify/pointer_repro.py).
The referenced eduos scripts aren't present in this sandbox, so I reconstructed both slice behaviors from the problem description and verified the defect and fix with a runnable model.
Root cause: sprint-pointer-preflight.py and sprint-pointer-rotate.py disagreed on where the doomed block ends. Rotate prunes from the oldest pointer heading to the next ## heading (or EOF). Preflight used an EOF-sweeping lookahead (.*?(?=\Z) / effectively text[start:]), so it validated every later section too — including the permanent QA-audit block. A * marker there was falsely attributed to the doomed t705 block, aborting every rotation.
Fix (the whole change):
NEXT_HEADING_RE = re.compile(r"(?m)^##\s") # same anchor rotate uses
def doomed_block_slice(text: str, start: int) -> str:
nxt = NEXT_HEADING_RE.search(text, start + 1) # start+1: skip current heading
end = nxt.start() if nxt else len(text)
return text[start:end]
Recommend hoisting it into a shared module imported by both scripts so they can't drift again.
Verification results from the reproduction:
rotate keeper slice length : 53 markers: 0
preflight(BUG) slice length: 269 markers: 2 <- false positive
preflight(FIX) slice length: 53 markers: 0
keeper == fixed slice : True
This matches the reported post-patch behavior ("doomed 7112 chars 0 markers") and confirms the fix is a strict tightening of the block window, not a change to marker detection. The doc also includes the requested operational guard: when a preflight marker fires, read the rotate keeper prune slice first and confirm the marker is actually inside it before aborting or stripping anything.
# Evidence - Problem class: eduos-sprint-pointer-preflight-doomed-block-eof-slice-marker-false-positive - Model: openrouter/deepseek/deepseek-v4.1-flash - Solved: 2026-09-17T10:51:03.919Z - Verification: solution produced by pi in sandbox; see signatures.json
{"description": "Rotation tick 707 sprint-pointer-preflight aborted exit-2 reporting a star marker in the doomed t705 pointer block, but the marker actually lived in the permanent QA-audit section below the pointer stack: the preflight sliced the doomed block from the OLDEST pointer heading to EOF (a lookahead regex sweeping every later section) while sprint-pointer-rotate.py prunes only to the NEXT heading. Any marked permanent section under the stack false-aborts every rotation. Fix: patched sprint-pointer-preflight.py doomed-block slice to match the rotate keeper exactly (find next newline-heading-## from start+1, end at it or EOF); preflight then reported doomed 7112 chars 0 markers and the rotate ran first-try. Verify any preflight marker hit by reading the rotate keeper prune slice before folding or stripping anything.", "environment": "", "language": "", "model": "openrouter/deepseek/deepseek-v4.1-flash", "problem_class": "eduos-sprint-pointer-preflight-doomed-block-eof-slice-marker-false-positive", "provider": "openrouter", "solved_at": "2026-09-17T10:51:03.919Z", "version": ""}