Optional "positive direction" on metrics

[siljamardla]

This is especially relevant for experimentation and reporting use cases. Metrics inherently have a direction of change that indicates improvement. Revenue going up is good, while cost going up is bad. We should be able to define this positive direction for a metric once (in OSI) and then we're able to reuse it everywhere.

For example, we can automatically colour a week-over-week change in a BI tool based on whether the change was good or bad.

[nikitastrezhnev]

While at it, we could give a thought to the broader set of behaviours here, and how to encapsulate them semantically.

A. While period-on-period revenue ⬆️ = Good, and PoP cost ⬆️ = Bad, there are metrics where "sentiment" is non-linear (i.e. higher is good, but much higher is bad). Think body temperature or, in business context – metrics like Inventory Days of Supply (how many days your current on-hand inventory will last at the current sales/usage rate), which you want to keep "in check".

• I reckon this would start to reek too much semantically of the concept of "targets" which should be an altogether different animal (or should it?). Food for thought :)

B. We should keep in mind that this "direction sentiment" concept would need to apply not only for temporal changes, but also across any subsets of the data. E.g. if in my analysis I compare average-revenue-per-order of New customers and Existing customers, we should be able to highlight that delta based on the same logic (higher = better).