Where Investor-Firm Rosters Live

The problem in one sentence

A team roster for an investor firm (e.g., AIX Ventures) is a fact about the firm, but it’s filed today next to the deal that prompted us to fetch it — so the same roster will end up duplicated under every deal that firm participates in, and the copies will drift the moment one of them gets re-crawled.

Where things stand right now (2026-05-08)

The first roster generated by [[Crawl-for-Better-Team-Structured-Output]] / [[Team-and-People-Metadata-Ingestion]] was for AIX Ventures, fetched as part of due diligence on ChromaDB. It currently lives at:

ai-labs/investment-memo-orchestrator/io/alpha-partners/deals/ChromaDB/outputs/aix-ventures-roster/
├── team-roster.json
├── team-roster.md
└── team-roster.raw.json

That works for one deal. It will not work for the second one.

The shapes the answer could take

Three options, each with a real downside:

1. Per-deal (current): deals/<deal>/outputs/<firm-slug>-roster/. Simple, deal-bundle is self-contained. Duplicates every shared firm.
2. Per-firm global: firms/<firm-slug>/team-roster.json. No duplication, single source of truth. Loses deal context (which deal triggered the fetch, what was the snapshot at memo-time).
3. Per-firm canonical + per-deal pointer: firms/<firm-slug>/team-roster.json is canonical; deals reference it via a small JSON pointer file or symlink. Best of both, but adds a layer of indirection that has to be respected by every consumer (memo writer, MemoPop app).

Open questions to resolve before committing

• Do investor firm rosters change fast enough that a per-deal snapshot is worth keeping? (My instinct: they change quarterly at most.)
• When a memo is finalized and shipped, does it embed the roster as it was at memo-time, or always reflect the current firm roster? (Almost certainly the former — memos are point-in-time documents.)
• Does this affect the orchestrator’s tenant model (io/<firm>/deals/<deal>/)? Could firms/ live alongside deals/ under each tenant, or does it need to be cross-tenant?
• The same question applies to portfolio-company team rosters cited in a memo (e.g., AIX’s portfolio cards mentioning Hugging Face, Perplexity). Single-write, multi-cite is the pattern.

What this stub doesn’t do

This is just a placeholder so the question doesn’t get lost. Promote to a real exploration the next time we’re about to file a second investor roster — that’s the moment the duplication problem becomes concrete and forces a decision.