How it works

Every insight Archora shows you is computed from the dependency graph by a documented, deterministic algorithm. No heuristics that say "here be dragons" without explaining why.

This section is the reference. Each page describes one insight: what it means, exactly how it's computed, and what to do about it.

Pipeline

discoverFiles → parseFiles → resolveImports → buildGraph
                                                    ↓
              recommendations ← layer rules ← metrics ← cycles (Tarjan SCC)
                      ↓
                hot zones (composite score)

All steps are pure functions of the file source. There is no incremental cache and no daemon — every scan reads the project from scratch. The full pipeline runs in 2 seconds on ~1000 modules and ~10 seconds on ~5000 modules.

Reference pages

• Cycle detection — Tarjan SCC, cycle patterns, severity.
• Feedback Arc Set — picking the one or two edges that actually break a cycle.
• Hot zones — the composite score behind the warm-colored nodes.
• Layered architecture — declared layers, layer violations, layer-aware recommendations.
• Recommendations — full catalog: split-god-module, unused-utility, type-only-candidate, misplaced-by-layer, isolated-cluster, cycle-break-candidate, cycle-break-cluster.

A note on terminology

Throughout these pages:

• module — any file the analyzer parsed. .ts, .tsx, .js, .jsx, .vue, .svelte.
• edge — an import (static, dynamic, or type-only) that resolved to another project module.
• fan-in — number of modules importing this one. fan-out — number of modules this one imports.
• cycle — a strongly connected component in the import graph with ≥ 2 modules.
• layer — a named bucket of modules from archora.config.