Divergence

Divergence here is the engine's median against a random walk — the price at the forecast's origin, carried forward. A large positive number means the model expects a rise relative to standing still; a large negative one, a fall.
Both columns are as of the issue date shown, not today. Comparing a forecast made last week against today's price would measure elapsed time, not disagreement.

It is not divergence from analyst consensus. That is the more valuable comparison and it needs an estimates feed; the panel beside this one names it.

Read the skill column alongside. A large disagreement on a series the model has no measured skill on is a claim, not a signal.