How it works

flowchart LR
  A[Grammar source] --> B[Format frontend]
  B --> C[Grammar IR v1]
  C --> D[Analyzer]
  D --> E[Features]
  E --> F[JSON]
  E --> G[Markdown]
  E --> H[LLM digest]
  E --> I[SARIF]

Frontends understand source syntax. Everything after the Grammar IR boundary is format-independent.

1. Frontend

The frontend parses source files and returns a GrammarIR plus diagnostics. It records grammar structure, source identity, capabilities, locations, and references. It does not execute target code embedded in a grammar.

2. Grammar IR

Grammar IR is the versioned contract between format-specific code and the analyzer. The core validator checks its shape and canonical serialization makes stable output possible. See Grammar IR for the fields and versioning rules.

3. Analyzer

The analyzer derives size, structure, precedence, lexicon, EBNF sugar, action presence, and capability-aware values. It also records diagnostics and notApplicable reasons where a metric does not make sense for a representation.

4. Reporters

JSON, Markdown, LLM digest, and SARIF reporters consume the same feature object. A report format changes presentation, not the underlying measurement.

Determinism

Given the same source bytes, source identity, frontend, and options, the pipeline is designed to produce byte-stable canonical output. Explicit source names and --source-root make multi-file runs reproducible.