LegacyDiff runs reviewed Python implementations on identical inputs and separates output differences from execution errors and unmet requirements. A React workspace brings the evidence, NVIDIA Nemotron investigations, collaborative review, and version-bound approvals together over SpacetimeDB.
Stack, status, evidence, and public actions are rendered from the typed project record.
Category
Developer Tooling
Type
Platform
Priority
Strong
Overview
What this project is
LegacyDiff runs reviewed Python implementations on identical inputs and separates output differences from execution errors and unmet requirements. A React workspace brings the evidence, NVIDIA Nemotron investigations, collaborative review, and version-bound approvals together over SpacetimeDB.
Problem
Why it matters
A rewrite can introduce a defect or faithfully copy an old one. Matching outputs alone cannot establish which behavior is correct.
Solution
Approach
Compare measured outputs with independent requirements and pinned evidence, investigate differences with bounded AI tools, then let a reviewer approve the intended behavior and export its checks.
Architecture
System shape and stack
Execution
Measure before explaining
A Python runner executes hash-verified fixtures for returns, pricing, accounts payable, and permissions. Output differences, execution failures, and independent requirement failures remain separate results.
Review
Approve an exact version
SpacetimeDB enforces membership and reviewer permissions. Decisions preserve the reviewed proposal and candidate; stale approvals and stale worker results are rejected. Downloadable bundles carry the selected sources, decision, and runnable checks.
AI
Investigations remain advisory
Nemotron specialists inspect behavior, policy, and regression evidence through bounded tools. Recorded calls, citations, partial results, and failures remain visible. Models cannot approve decisions or automatically repair the implementation.
- React
- TypeScript
- Vite
- Python
- SpacetimeDB
- NVIDIA Nemotron
Technical Highlights
Visible technical signal
- Actual execution across four reviewed synthetic scenarios
- Server-enforced membership and approvals tied to exact versions
- Exportable acceptance tests with source and decision integrity checks
What It Proves
Builder signal
Connecting execution, evidence, AI tooling, collaborative state, and human decisions into one inspectable developer workflow.
Boundaries
Context that should stay visible
Local application using reviewed synthetic fixtures. The runner has reduced isolation and is not an arbitrary-code sandbox. Public hosting, arbitrary repository ingestion, and production readiness are outside the current build.