A percentage can wear a very convincing suit
Put 60% beside a football forecast and the sentence suddenly looks as if it has passed an audit. It has not. At Who Fed the Goblin, that number is a subjective research judgment about one defined event. It is not a demonstrated success rate, a sportsbook price or an immunity certificate. The Goblin's legal department is a folding chair.
Our research starts with roles, statistics, availability, matchup conditions and a plausible game scenario. The percentage records how we weigh that evidence. Calling it uncalibrated is not decorative small print. We have not demonstrated that the group of events assigned 60% will happen 60% of the time. The distinction matters most when the number looks reassuring.
Read the event before the estimate
A player projected for 70 receiving yards and assigned a 60-yard target has two separate numbers. Seventy is our central yardage scenario. Sixty is the boundary of the event being forecast. Finishing on 60 clears an at-least-60 target; finishing on 59 does not. Neither tells you what a sportsbook will offer, and our threshold is not automatically a quoted betting line.
The projection is not a floor. A receiver can retain his starting job and still miss the target because the offense runs fewer plays, protection fails or targets go elsewhere. A running back's useful receiving work does not rescue a rushing-yard forecast. A touchdown forecast means the player personally scores; throwing a touchdown pass does not satisfy it.
This is why we keep both event results and yardage error. A forecast can clear a modest threshold while its central yardage projection is badly wrong. Counting only wins would hide that difference. Counting only projection error would hide whether the exact event happened.
The average is not the odds
Consider a deliberately invented example, not a real player's record: four receiving games of 20, 20, 100 and 100 yards. The average is 60, yet only two of the four games reach 60. Change the sequence to 60, 60, 60 and 60 and the average stays identical while all four reach the target. An average alone cannot tell you an event's hit rate.
We use season totals to anchor what a player has done and whether a workload assumption is plausible. We do not divide a season touchdown total by games and call the answer the chance of scoring next week. Multiple touchdowns can occur in one game, and a new team, injury or different role can change the underlying opportunity. More decimal places do not repair a missing argument.
Five cards are not one probability
Our five player outcomes describe five individual events. They are not a combined parlay recommendation. Adding their percentages makes no sense, and multiplying them assumes independence that has not been established. A quarterback struggling under pressure can damage multiple receiving forecasts together. A comfortable lead can help a rushing target while reducing opportunities for a receiver on the same team.
For arithmetic illustration only, five independent events each assigned 60% would all occur with probability 0.6 to the fifth power, or about 7.8%. That is not the probability of any Goblin package. The example exists to show why five individually plausible events do not become a likely five-for-five result. Real shared-game dependence needs its own analysis.
We also do not infer betting value from a forecast percentage without a relevant price, or from an attractive price without a credible event estimate. A publisher-displayed quote may be useful context, but it is not proof that the same wager is currently executable. One quote is a snapshot. Calling it movement requires comparable observations at different times.
The small labels carry the heavy equipment
A final injury report and a game-day inactive list answer different questions. Full practice does not guarantee an unrestricted workload. Questionable does not mean out. A blank report is not a clean bill of health. When those checks are missing, the page should say so, even when uncertainty makes the design less tidy.
Weather has the same problem. A city's Sunday high is not the temperature at kickoff. A regional thunderstorm discussion is not proof that the stadium will have a delay. A retractable roof is not a confirmed closed roof. We distinguish the kind of evidence retrieved from the condition we would like to know.
Before relying on a forecast, check its timestamp, the latest revision, the event definition, the assumed role and the unresolved risks. A sleeper label means a less established or more volatile opportunity. It is not a secret synonym for safe.
The receipts must survive the result
Our application appends revisions rather than quietly rewriting an earlier call. A newer report points to the one it supersedes. The database blocks new game forecasts at kickoff, and a completed verified pregame scan can freeze them earlier. The original article text, sources and timestamps remain part of the record. This is an application audit trail, not independent notarization or a claim that an administrator could never change the software.
After a verified final result, we grade the last eligible report. Probability error, event hit rates and yardage error answer different questions, and each needs a sample size. Earlier editorial v1 calls published no probabilities, so their historical probability fields remain empty. Filling them after the result would create a very impressive model of the past and a fairly embarrassing research process.
A bad outcome deserves an honest result entry. A method change needs more: a specific failure hypothesis, a versioned experiment and evidence held apart from the examples that inspired the change. We do not claim calibration, automatic language-model training or improved forecasting simply because we have started keeping score. The percentage is the beginning of an argument. The record is where that argument has to live.
