Skip to main content

Module edge_confidence

Module edge_confidence 

Source
Expand description

Confidence for call-graph edges (Plan competitive-harvest t-013).

Static call edges are stored as bare symbol_id lists with no confidence. This adds a sidecar layer: each edge accumulates runtime confirmations (the asd tracer observing the caller actually invoke the callee) on top of a static prior, and a confidence is derived the same way effect confidence is in t-001 — reusing RuntimeEvidence::derive_confidence.

Two provenances:

  • static-known — the edge is in the static call graph (prior STATIC_PRIOR); runtime confirmation raises it, absence never lowers it (a path simply not exercised is not evidence the edge is wrong).
  • runtime-only — the tracer observed a call the static walker missed (dynamic dispatch, reflection, callbacks). It definitely happened, so it starts at RUNTIME_ONLY_PRIOR and confirmations push it up fast. These are the high-value rows: real edges absent from the static graph.

The sidecar lives at /asd/v1/index/edge-evidence/<from>/<to>, leaving the hot-path callees/callers lists untouched.

Structs§

EdgeEvidence
Accumulated evidence for one directed call edge.

Constants§

RUNTIME_ONLY_PRIOR
Prior for an edge seen only at runtime (missed by static resolution).
STATIC_PRIOR
Prior confidence for an edge present in the static call graph.