Trendline and prediction analysis is for a single line, so it would be found in a 1D1M story or in a drilldown section. It calculates a linear best fit line. It also calculates the confidence level to determine how good of a fit the data is with the best fit line. So a line with a lot of variability would have a lower confidence level than a very straight one. Whether or not the content prints is based on an adjustable threshold in the analytics section of the settings modal.
The language changes a bit depending on verbosity, but the core insight is communicating the slope of the line and the absolute change over a given period. For higher verbosities, it also comments on the R-squared value, which is a statistical concept that quantifies how well the data fits to the line.
Finally, there’s a prediction piece, which is also configurable in the analytics section of the modal. You can choose to write or not write the prediction, as well as the number of periods used in the prediction. The prediction essentially extends the best fit into future periods, and it uses the slope and the intercept of the line to calculate a predicted value. The confidence piece of the prediction adds additional “bounds” to this prediction by using the upper and lower bounds of the confidence interval that was selected (same configurable threshold).
Therefore, there are some pre-requisites for this analytic to write, so it will not be included in every story. It needs to have at least 30 data points and it also needs to be “linear” with a certain level of confidence.