Similarity Scoring

When symtrace matches a code node across two commits (as a MOVE, RENAME, or MODIFY), it also reports how much changed. This is the similarity score.

How it’s calculated

The score combines three factors:

• Structure (50%) — did the shape of the code change? (nesting, number of children)
• Tokens (30%) — did the actual source text change? (identifiers, literals, operators)
• Complexity (20%) — did the control flow change? (branches, loops, conditions)

Change intensity

The score maps to a simple rating:

Similarity	Intensity	What it means
80% or higher	low	Minor edit — probably safe
50 — 79%	medium	Real change — worth reviewing
Below 50%	high	Major rewrite — treat as new code

Reading similarity in output

In CLI output, you’ll see it inline:

~ [MODIFY] function 'parse_body' modified (L10 -> L10) [75% similar, medium]

In JSON output (--json), matched operations include a similarity object:

{
  "similarity_percent": 75.2,
  "change_intensity": "medium",
  "structure_similarity": 0.84,
  "token_similarity": 0.61,
  "control_flow_changed": true
}

Notes

• INSERT and DELETE operations don’t have similarity scores (there’s nothing to compare against).
• A MOVE with 100% similarity confirms code was relocated without modification.
• A RENAME with 98% usually means only names changed.