v1.0.0 — Policy-Driven A/B Governance Engine (MVP)

v1.0.0 — MVP Release

A policy-first A/B testing governance engine that turns experiment telemetry into deterministic decisions with audit-grade traces, plus an optional LLM copilot behind a secure proxy (no decision authority).

What’s included

• Deterministic decision engine (policy-as-code)

  • Decisions are driven by explicit policies (YAML) and structured signals (p-value, uplift, SRM, ROI, expected loss, power).
  • Produces stable outcomes like IMPLEMENT_TREATMENT, CONTINUE_TEST, STOP_TEST, with confidence + triggered rules.

• Governance-grade decision trace

  • Every decision includes: key drivers, risks, and concrete next actions.
  • Human-readable summary is generated from the same structured inputs (no hidden logic).

• LLM copilot via secure proxy (optional)

  • LLM is used for executive summaries only — never as the source of truth.
  • Built-in fallback behavior if the model is unavailable / forbidden / rate-limited.

• Caching

  • Repeat requests return instantly, reducing latency and cost.
  • Improves UX and stability under repeated interactive usage.

• UI improvements

  • Clean executive-style output (noise removed).
  • Copy-friendly summaries for quick sharing.

• Docker Compose dev stack

  • One command brings up the full system (web UI + LLM proxy):
    • docker compose up --build
  • Secrets are kept local via .env (never committed).

• GitHub Pages demo

  • Public UI available for quick product walkthrough (LLM features require local proxy/key).

Quick start (local dev)

1. Copy env:
   • cp .env.example .env
2. Add your proxy key in .env
3. Run:
   • docker compose up --build

Notes

• Public GitHub Pages build intentionally runs without LLM keys.
• The local proxy is the recommended path for the full “copilot + governance” experience.