DOCS/Explainability
⌘K
v1
Concepts

Explainability

How drivers and reasons are attached to every recommendation.

Handing price control to a system is only acceptable if you can see why it decided what it did. "No black box" is a core requirement in Elastly, not a tagline — every recommendation carries its full provenance.

What travels with a recommendation

  • Drivers — the human-readable factors behind the price: the strategy used, the competitor median and how many sellers it covers, and each guardrail that moved the price.
  • Binding rule — if a guardrail held the price off its target, which one.
  • Rules applied — the ordered trail of every clamp, with the before/after value at each step.
  • Confidence — a transparent heuristic (higher with a competitor signal, lower when a guardrail held the price or the move is large).

The optional LLM layer

When configured, an LLM turns those deterministic drivers into a one or two sentence natural-language summary — strictly grounded in the provided facts, never inventing a number. It's an explanation layer on top of the math, not the decision-maker. Without it, the recommendation still shows its drivers; the language is just terser.