Skip to content

product.md

Latest commit

 

History

History
78 lines (64 loc) · 3.67 KB

product.md

File metadata and controls

78 lines (64 loc) · 3.67 KB
Raw

voiage - Product Definition

Vision

voiage aims to be the premier cross-domain, high-performance Value of Information (VOI) library for researchers and decision-makers across health economics, clinical research, finance, environmental policy, engineering, and adjacent uncertainty-analysis domains.

Mission

Provide a comprehensive, open-source VOI analysis toolkit that is rigorous, reproducible, contract-first, and usable from Python while keeping the path open for stable polyglot bindings and native execution-core expansion.

Target Users

  • Health Economists and HTA Agencies: cost-effectiveness and reimbursement analyses.
  • Clinical Researchers: adaptive trial design, sample-size optimization, and calibration-study planning.
  • Policy Analysts: uncertainty-reduction decisions in public, environmental, and implementation policy.
  • Financial Analysts: research portfolio and investment decisions under uncertainty.
  • Academic Researchers: VOI teaching, method validation, and methodology development.

Current Product Shape

  • Core VOI Surface: EVPI, EVPPI, EVSI, ENBS, structural VOI, NMA VOI, calibration VOI, adaptive trial VOI, portfolio VOI, sequential VOI, CEAF/dominance, heterogeneity, and plotting are implemented in-repo.
  • Frontier Surface: perspective, preference, validation, threshold, distributional/equity, implementation-adjusted, and adjacent frontier contracts are fixture-backed and explicitly maturity-labelled.
  • Cross-Domain Support: healthcare, financial, environmental, engineering, and domain-agnostic APIs share the same core data contracts.
  • Polyglot Readiness: R, Julia, TypeScript, Go, Rust, and .NET bindings are aligned around shared conformance fixtures and registry-aware release paths.
  • HPC Enablement: CPU parallelism, scheduler adapters, Apple Metal, discrete GPU, TPU, FPGA, and ASIC lanes have contracts and evidence gates. Production speedup claims remain gated by benchmark and hardware evidence.
  • Documentation: Astro and Starlight provide the authoritative documentation site, local docs gate, and GitHub Pages deployment path.

Architecture

  • Python Reference Facade: Python remains the primary user-facing package.
  • Contract-First Core: schemas, deterministic fixtures, result envelopes, and diagnostics define compatibility before binding expansion.
  • Backend Abstraction: NumPy, JAX, parallel/distributed adapters, and future native kernels must preserve public API behavior.
  • Rust-Core Boundary: Rust foundations and deterministic kernels exist, but broader production-kernel migration remains evidence-gated.
  • CLI-First Workflow: core and frontier methods expose command-line entry points for reproducible batch analysis.

Completion Boundaries

  • In-repo roadmap and Conductor tracks can be repository-complete while registry approvals, external indexing, fabricated hardware, and accelerator speedup evidence remain external gates.
  • FPGA and ASIC support is currently pre-silicon evidence only unless real board, shuttle, or fabricated-silicon evidence is added later.
  • Binding publication status should distinguish generated artifacts and automated release workflows from registry-side approval or indexing.

Success Metrics

  • Reliable >90% local test coverage with CI-enforced lint, type, docs, and test gates.
  • Stable conformance fixtures across Python and supported external bindings.
  • Clear external-gate reporting for registries, HPC distributions, and hardware acceleration.
  • Method coverage competitive with BCEA, dampack, voi, and commercial VOI tools.
  • Review-ready documentation, examples, and reproducible evidence packets.