An Auction Row Has No Consensus Field. So the Blackout You Put Around It Is Sized by Nothing.
The Treasury sold $25 billion of 30-year bonds at 18:00 UTC and four publishers graded the identical four numbers "in line" with no directional read, "demand disappoints", "weaker demand" and the weakest since June — one of them carrying "weak demand" in a headline over a body that applies no grade at all. We are not adjudicating, because the structural point is that an auction row has a time and a size and no consensus field, so there is nothing to score a result against and whatever your feed puts in the actual column is editorial judgement in a data field's clothes. Three publishers between them filled five fields and no two filled the same ones: the high yield of 5.441 per cent is absent from the publisher with the fullest bidder distribution, the when-issued 5.437 appears at one house with no tail, the 0.4 basis point tail appears at another with no when-issued level, and the cover is 2.36 at two and 2.401 at a third. The sequencing is worse. Dollar-yen touched 160 for the first time in about two and a half months, an index headline carried it at 16:56 UTC, the pair was 159.75 at 17:59, and the auction was at 18:00 — so whatever took it there, it was not a result that had not happened yet, and the headline claiming the auction sent the ten-year above 5.10 per cent rests on a session high its own body gives with no figure and no clock. The one reliable timestamp in the whole chain belongs to the scheduled event, which argues for anchoring an event log on the calendar and treating every price stamp as the softer field. Then a fourteen-minute problem: a presidential remark on Iran "briefly lifts the dollar" at 13:33 UTC and draws "limited reaction" from the same publisher at 13:47, which means a keyword filter would have fired on an effect shorter than its own detection latency. We cut two things and the cuts are the point — six article bodies at one publisher failed while its index carried nineteen timestamped headlines, and a translated morning note implies Fed pricing several times what a reproducible calculation gives, which we are not publishing and not resolving. There is no field anywhere in an ingest for "this number is inconsistent with one I can derive", and it is the most useful flag available at no extra data cost.