Skip to content

README.md

Latest commit

 

History

History
25 lines (19 loc) · 2.66 KB

README.md

File metadata and controls

25 lines (19 loc) · 2.66 KB
Raw

ScenarioRank AI V2 documentation

These documents form the engineering baseline for the post-award V2 refinement.

Start here

  1. PROJECT_STATUS.md — read this first. The living source of truth for the project's objective, current state, decisions made, and next step. Every other document here goes deeper on one part of it.
  2. PHASE_0_BASELINE_AUDIT.md — what exists today and what Phase 0 establishes.
  3. architecture/CURRENT_ARCHITECTURE.md — components, responsibilities, and pipeline boundaries.
  4. architecture/DATA_FLOW.md — how requests and data move through the system.
  5. architecture/SCORING_AND_ASSUMPTIONS.md — formulas, AI-derived inputs, and unvalidated assumptions.
  6. architecture/KNOWN_LIMITATIONS.md — prioritized correctness, architecture, operational, and ethical risks.
  7. REPOSITORY_MAP.md — which files are active, support code, generated code, or likely legacy.
  8. BRANCH_STRATEGY.md — how the award snapshot is preserved while main becomes V2.
  9. V2_ROADMAP.md — staged modernization plan.
  10. LEARNING_CHECKPOINTS.md — conceptual questions the maintainer should be able to answer.

Architecture decisions

  • decisions/ADR-0001-main-is-v2.md — why main is the public V2 line and the award version is preserved separately.
  • decisions/ADR-0002-provider-abstraction.md — why Anthropic coupling was removed in favor of a provider-neutral contract, why Groq/Gemini were the first two providers built, and why not Google ADK. Its choice of providers is superseded by ADR-0004 — kept as the honest historical record of that real, tested work.
  • decisions/ADR-0003-runtime-provider-configuration.md — .env/.env.local precedence, one-provider-per-process-lifetime, and startup validation behavior.
  • decisions/ADR-0004-single-openai-provider.md — why ScenarioRank simplified to a single OpenAI provider after a real end-to-end test showed neither Groq nor Gemini could reliably complete a full run on its free tier, and why the provider-neutral contract is still worth keeping with one provider.

These documents describe the current implementation honestly. They do not imply that planned V2 capabilities already exist.