trace-foundations

Foundation tracing uses a two-tier cache with heuristic matching and LLM verification to identify open-source foundation affiliations for a list of repositories. It processes expanded GitHub repository data and outputs foundation names with supporting evidence.

Automating foundation tracing involves a layered workflow matching repositories against a foundation-project cache, applying org heuristics, and using LLM web research for unknowns. This approach enforces deterministic scripting with human-in-the-loop checks to ensure accuracy.

LLM verification for foundation mapping handles unknowns by conducting web research when cache and heuristic matching fail. The LLM evaluates repository evidence to make foundation assignment decisions, which are then routed to human review for final confirmation.

You need an expanded repository dataset named repo_exp.csv to start tracing foundation affiliations. The workflow consumes this structured CSV input and outputs foundation.csv containing foundation_name and supporting evidence for each repository.

A key limitation of using LLM web research for governance identification is potential inaccuracy, which necessitates a human-in-the-loop check. The workflow mitigates this by enforcing deterministic scripting and routing uncertain LLM decisions to human review.