GPUDECK · COST PER FINISHED RENDER · INPUT TOO STALE
Cost per finished render divides a price by a measured goodput rate. Today those two observations are 18 days apart — the price side observes 2026-09-22, the goodput side 2026-09-04 — and GPUDeck's published limit is 7 days.
A cost figure built from mismatched vintages would look exactly as confident as a good one. So it is withheld rather than printed with a caveat. This is a pre-committed rule, applied against GPUDeck's own headline metric.
Nothing measured is hidden: the goodput ratios remain on the observatory and in the open data, and the price records remain on prices. Only the division of one by the other is withheld, and only while the gap persists.