Every fixture analysis ends with a Flags section. “Flags: None” means the checks behind the analysis passed and nothing needs qualifying. When a flag appears, it tells you which part of the output to treat with care, or why a recommendation has been withheld.
Divergence. Fires when the model’s probability and the market’s implied probability sit far apart on an outcome. Treat it as an instruction to look closer: either the model has found a mispriced market, or the market knows something the data does not show. The divergence flag has its own entry covering how to read it.
Data flags. These mark gaps in the inputs. If a figure the analysis needs is missing or unconfirmed, the output says so rather than guessing. An in-play analysis without live xG builds an estimate from shot data and labels it “xG Estimated” so you know the figure is constructed rather than measured. Where the data is too thin to analyse a market at all, it is labelled data-insufficient and no recommendation is made.
Reliability flags. These fire when the model’s own checks doubt its numbers on a market, for example an edge that looks inflated on one side of the match odds. Recommendations on that market are suppressed rather than published, and related markets that inherit the same probabilities are suppressed with it.
Assumption-sensitive. This flag means the conclusion leans on assumptions rather than confirmed data. The confidence tier is downgraded a step and the flag tells you why.
Read a flag before acting on the numbers it qualifies. Withheld or suppressed recommendations are the same system at work: where the checks fail, the platform holds the recommendation back instead of publishing it.