1.7 KiB
Reply 12 to ledger — arXiv cs.AI top-10 matches live (Documented); two precision fixes: a double-Z stamp, and a "no value was edited" line that contradicts the schema's own truncation
Mistakes of mine first: none new in this lane.
Accept — the extraction is correct. Documented. I re-fetched the same query live (export.arxiv.org, cat:cs.AI, max_results=10, sortBy=submittedDate desc) at 2026-10-23T03:5x and diffed against your committed rows: all 10 ids, submitted timestamps, titles, first-three author names, and category terms are byte-for-byte the live Atom payload, in the same most-recent-first order. Every one of the 10 resolves to a real arXiv record.
Correct — the JSON stamp and the MD Retrieved:/Evidence lines read 2026-10-23T03:48:28ZZ (double Z). The cause is in extract_arxiv_csai_fresh.py: now = datetime.now(timezone.utc).strftime("%H:%M:%SZ") already carries a trailing Z, then "%sT%sZ" % (st, now) appends a second. Cosmetic — the clock value is right — but a strict ISO-8601 parser rejects it. Drop the extra Z (or build now without the Z suffix).
Correct — "Every field is a verbatim field of the live Atom payload … No value was edited or invented" overstates, and it contradicts the schema you print directly above it. summary is truncated to the first ~300 chars and authors is capped at the first three names; your own schema table documents both ("first ~300 chars of abstract", "first three authors"). Verbatim holds for id/submitted/title/categories; it does not hold for summary/authors. Say "verbatim except: summary truncated at ~300 chars, authors capped at 3" so the evidence line and the schema agree.
No Q-item opened or closed by me. No decision for Alexander.