Trace Foundations

To merge new projects with an existing cache (preserving LLM-discovered entries):

python3 scripts/build_foundation_cache.py --merge -o output/.cache/foundation_projects.json --summary

Step 2: Run the matching script

python3 scripts/trace_foundations.py output/repo_exp.csv -o output/foundation.csv --summary

The script applies 3 layers:

1. Cache match: Check each repo against foundation_projects.json (by owner/repo or project name)
2. Org heuristics: Map known GitHub orgs to foundations (e.g., apache/* → Apache, kubernetes/* → CNCF)
3. Mark unmatched: Remaining repos are marked foundation_name=unknown for LLM processing

Step 3: LLM Web Search for unmatched repos

For repos still marked unknown:

1. Batch by GitHub org — repos in the same org likely share the same foundation
2. Web Search for each unmatched org/repo: "{project_name}" foundation OR governance
3. Determine: which foundation (if any) does this project belong to?
4. Record:
   • foundation_name — foundation name, or none if not affiliated
   • evidence — actual URLs and facts from search
   • confidence — S/A/B/C

IMPORTANT: If LLM discovers a previously unknown foundation with projects:

1. Add the foundation and its projects to output/.cache/foundation_projects.json
2. Consider adding the org mapping to scripts/trace_foundations.py's ORG_FOUNDATION_MAP
3. Re-run the script to batch-match remaining repos

Step 4: Present to user

Show results grouped by foundation:

• Foundation name
• Repos under this foundation
• Evidence and confidence for each

User reviews and may correct assignments. Update output/foundation.csv with corrections.

Step 5: Output summary

Print:

• Total repos processed
• Repos matched by cache / heuristics / LLM
• Repos with foundation affiliation (count by foundation)
• Repos with no foundation (none)
• Count by confidence level

Next step

→ Can run in parallel with /trace-companies