Skip to content
Impact Gate Impact Gate Impact Gate

Impact Analysis

Impact analysis answers a fundamental question: given a code change, which E2E tests need to run?

How It Works

The impact engine takes a git diff and maps each changed file to the route families it belongs to. A route family represents a product feature (e.g., “channels”, “messaging”, “settings”) with associated source paths, test directories, and user flows.

channel.go changed
  -> belongs to "channels" family
    -> specs live in specs/functional/channels/
      -> run those tests, flag coverage gaps

Three Mapping Strategies

  1. Route families — the route-families.json manifest maps source file patterns to features
  2. Dependency graph — static reverse-dependency analysis catches transitive impacts (a utility changed, so all consumers are affected)
  3. Traceability — file-to-test mappings from CI execution history provide empirical evidence of which tests exercise which files

Running Impact Analysis

The recommended way to run impact analysis is through the review command, which combines impact analysis, behavior analysis, coverage planning, and defect prediction:

Terminal window
# Unified review: impact + coverage gaps + defect risk + recommendations
npx impact-gate review --path . --since origin/main


# Review + generate test files for uncovered flows
npx impact-gate review --path . --since origin/main --generate

For lower-level access, the impact command outputs just the impact analysis:

Terminal window
npx impact-gate impact --path . --since origin/main

Use plan when you want .e2e-ai-agents/plan.json and .e2e-ai-agents/ci-summary.md artifacts with impactedFamilies, runSet, confidence, and decision.

Building the Route Families Manifest

For accurate results, build a manifest that maps your codebase to features:

Terminal window
# Offline scan (free, no API key)
npx impact-gate train --no-enrich --path .


# With LLM enrichment for better metadata
npx impact-gate train --path .

The scanner uses directory matching, test-derived discovery, server-derived grouping, and name matching to build file-to-family mappings. LLM enrichment adds URL routes, priority levels, and human-readable flow descriptions.

Interpreting Results

Each impacted family in the output includes:

  • confidence — how certain the mapping is (heuristic, traceability, or AI-confirmed)
  • impactLevel — direct change vs. transitive dependency
  • runSet — the specific spec files to execute
  • gaps — user flows with no test coverage for the changed code