# `Monty.Frameworks.Models.EvidenceMatching.FeatureEvidence.Scorer.FeatureEvidenceScorer`
[🔗](https://github.com/d-led/ex.tbp.monty/blob/v0.1.0/packages/ex_tbp_monty_core/lib/monty/frameworks/models/evidence_matching/feature_evidence/scorer/feature_evidence_scorer.ex#L1)

What a feature-evidence scorer has to be able to do.

Mirrors
`tbp/monty/frameworks/models/evidence_matching/feature_evidence/scorer.py::FeatureEvidenceScorer`,
which is a `Protocol` and, here, a behaviour: given a graph, a channel and an
observation, answer with one evidence value per node of that channel's graph.

Monty's own note is that the name is a placeholder from a refactor; it is kept
because an experiment that names a different scorer names this shape.

# `evidence`

```elixir
@callback evidence(
  scorer :: term(),
  graph_id :: String.t(),
  input_channel :: String.t(),
  query_features :: %{optional(String.t()) => term()}
) :: [float() | nil]
```

Evidence in `[0, 1]` per node, or `nil` where a node has no stored value.

The first argument is the scorer's own configuration — Monty's `self`, which
holds the graph memory, the weights and the tolerances. A channel that is not
matched on its features answers all zeros: it still contributes pose evidence,
which is computed elsewhere.

---

*Consult [api-reference.md](api-reference.md) for complete listing*