What a good summary does
States what is unusual relative to the asset's own recent history, names the window, and attributes each figure to its source. "Volume is three times its thirty-day median" is checkable; "volume is surging" is not.
It also says when nothing is unusual, which is the most common honest answer and the one marketing-driven summaries never give.
The fabrication risk
A model generating text about markets will produce plausible numbers whether or not it has them. This is the central failure mode and it is invisible at a glance, because fabricated figures look exactly like real ones.
The only defence is architectural: the model summarises data passed to it and is not permitted to supply figures of its own.
Why sources and windows matter
A figure without a window is not a measurement. "Up fifteen percent" over an hour, a day and a month are three different statements, and the one chosen is usually the most dramatic.
Naming the source lets you check it, and it reveals when two sources disagree — which happens often and is itself informative.
What a summary cannot become
A recommendation. Describing conditions and advising action are different acts, and sliding from one to the other is what turns a useful tool into a liability for the person reading it.
FBT Swap generates market summaries from data with the source and window named, states when nothing unusual is present, shows an unavailable state rather than inventing a number, and never converts a summary into a buy or sell instruction.