Questionable is a branch. Inactive is a fact.
A Friday injury designation describes a tree of possible Sundays. A final inactive list cuts off the branches. That distinction sounds obvious until a familiar name tempts the forecast to treat hope as roster data.
Before the London kickoff, the official lists made Keenan Allen and Terry McLaurin unavailable. The forecast did not infer a surprise recovery, invent a decoy role or quietly preserve their player outcomes. Both names came off the card.
Revise only what the evidence touches
The winner stayed Indianapolis 24, Washington 20 at an explicitly uncalibrated 65%. The final lists changed the receiving map, not the whole game model. Josh Downs entered at 40 or more receiving yards because he led Indianapolis in receiving yards and Allen was inactive. Stefon Diggs entered as an anytime touchdown outcome because he led Washington in receiving scores and McLaurin was inactive.
Jonathan Taylor, Tyler Warren and sleeper Jacory Croskey-Merritt remained. Their roles survived the new facts. A bounded revision should be boring in exactly this way. If one inactive receiver causes every score, probability and threshold to perform interpretive dance, the method has become a weather vane wearing a lab coat.
The checks have different jobs
Official injury reports define known availability before the final list. Final inactives settle who is eligible. Weather describes the playing environment. Operator publications show how one displayed market changed, but they do not guarantee an executable price and they do not manufacture Atlas probabilities.
Those sources can disagree or arrive at different times. The correct response is not to average them into soup. It is to let each source answer the question it is qualified to answer and keep unresolved items visibly unresolved.
Freeze the receipt before kickoff
The final revision was published before kickoff and then action locked. That lock captured the winner, score, five player outcomes, probabilities, article state and evidence timestamps. After kickoff, the application and database reject revisions even if somebody discovers a very persuasive adjective.
The depth chart gets one final chance to ruin breakfast, then the database takes away the keys. A forecast is cheap. An immutable receipt is the part that can be graded.
Grade after FINAL, not during the autopsy
The locked card now waits for a verified final game result and complete player statistics. The winner estimate will receive a Brier error. The five player events will be graded once each. Yardage projections will receive absolute errors. Voids require documented participation evidence instead of a convenient shrug.
That result will not prove calibration from one game, train a language model or justify rewriting history. It will add one honest observation to a growing record. The corpse stays where we found it.
