9,900+ Tests. All TDD-First.

pytest · pytest-asyncio · mypy strict · bandit · pip-audit · GitHub Actions

• Inter-agent handoff verification, per-agent model routing validation, autonomous task delegation correctness
• Cache layer testing: L1 memory / L2 Redis / L3 PostgreSQL hit/miss/eviction edge cases, TTL behavior, 88% aggregate hit rate validation
• Security testing: Ed25519 webhook signature verification, parameterized SQL injection resistance, rate limiting enforcement, OWASP control validation
• CI gates block merge on: SAST failure (bandit), CVE detection (pip-audit), type errors (mypy strict, 15 modules), lint violations (ruff)

pytest · RAGAS · LLM-as-judge · Brier score calibration

• Golden evaluation set: 24 fixtures including 8 adversarial cases (contradictory documents, truncated inputs, multi-language, high-noise OCR)
• RAGAS pipeline: context recall (0.35), faithfulness (0.40), answer relevancy (0.25) scored by LLM judge with rubric and evidence extraction
• CI regression gate: any accuracy drop below 94.6% baseline blocks merge before code review
• Brier score calibration: tracks predicted vs actual extraction success across 24 fixture types

pytest · pytest-asyncio · Pact v3 · Locust · GitHub Actions

• Multi-turn conversation flow verification: 3 bots, bilingual routing (EN/ES), qualification logic, temperature scoring, admin handoff edge cases
• Webhook security: dual-scheme signature verification (HMAC + RSA), request replay prevention, malformed payload handling
• Integration: Redis rate limiting (60 req/min enforcement), atomic session locks, conversation state persistence and recovery
• Consumer-driven contract tests (Pact v3): 5 contracts on the GoHighLevel API v2 boundary — contact lookup, SMS dispatch, opportunity creation, tag management, 404 error handling. Pact JSON generated and versioned.
• Load tested against production: 20 concurrent users (Locust) — p50=130ms, p95=250ms, p99=400ms at 5.54 req/s. Rate limiting and auth enforcement verified under concurrent load. Zero 5xx errors.
• Zero test failures across the full 3-month production run (January – March 2026)

Playwright · TypeScript · 4 viewport breakpoints

• Business content integrity: heading hierarchy, CTA visibility, pricing presence, contact info, footer structure
• Link integrity: internal anchor target validation, external link rel=noopener enforcement, mailto format validation
• Visual regression: screenshot comparisons at 3 viewports (375px mobile, 768px tablet, 1024px desktop) with per-pixel threshold
• Performance: DOM load under 3s, above-the-fold content in viewport, zero JS runtime errors, image dimension attributes (CLS prevention)
• Viewport responsiveness: 4 breakpoints, horizontal overflow detection with 5px tolerance