baseline-quality-assessment

Baseline Quality Assessment

Scope

Captures current-state evidence BEFORE any change so after-change validation can cite improvement or non-regression. Optional — only needed when a plan modifies existing behavior.

When to Use

• Plan modifies an existing route, endpoint, or screen
• Plan changes DB schema on an existing table
• Plan refactors existing code paths
• Plan claims "improves X" or "fixes Y bug"

When to Skip

• Greenfield plan (nothing to compare against)
• Pure additive work with no user-facing overlap
• Plan explicitly covers only new behavior

Capture Pattern

mkdir -p e2e-evidence/baseline/

# Web: screenshot the current state
agent-browser open http://localhost:3000/features/target
agent-browser screenshot e2e-evidence/baseline/before-ui.png

# API: current response
curl -s http://localhost:3000/api/target | jq . \
  > e2e-evidence/baseline/before-api.json

# Performance: current latency
for i in {1..20}; do
  curl -s -o /dev/null -w "%{time_total}\n" http://localhost:3000/api/target
done > e2e-evidence/baseline/before-latency.txt

# Accessibility: current axe score
npx @axe-core/cli http://localhost:3000/features/target \
  --save e2e-evidence/baseline/before-axe.json

Compare Pattern (in post-change gates)

Every gate that claims "improves" MUST cite both files:

Before: e2e-evidence/baseline/before-latency.txt p95 = 340ms
After:  e2e-evidence/vgN-after-latency.txt   p95 = 180ms
Improvement: 47% reduction

Pass Criteria for "Improvement" Claims

Good:

• "p95 latency reduced from 340ms to < 200ms (see baseline/before vs vgN-after)"
• "axe violations reduced from 4 serious → 0 (cite both JSON files)"
• "Bundle size reduced from 482KB → 310KB"

Bad:

• "Faster" → how much? measured how?
• "Better" → on what dimension?

Non-Regression Rules

Even a plan that is only adding features should capture baseline for:

• Page load time of the parent route
• Bundle size of the shared entry
• Key integration test (synthetic check existing behavior still works)

Post-change gates assert non-regression:

• "Parent route load time did not increase by > 10%"
• "Bundle size did not grow by > 50KB (new code should fit under budget)"

Security Policy

Baseline captures may contain authenticated UI state. Store under e2e-evidence/baseline/ which is part of the gitignored evidence dir.