Persistence Honesty: Guests, Firebase, Cache, and User-Owned Sync
“Local-first” in our prose means interaction and ingestion default to the device — not “no server for signed-in users.” This part is the mode matrix every writer and growth campaign must internalise. Deeper guest/auth detail also lives in content/coderlegion-sovereign-engineering-serial/05-local-first-browser-vault.md; here we align with the 2026 calibration ledger.
Ground truth table
| User mode |
Authoritative trades |
Code touchpoints |
| Guest |
Browser localStorage (namespaced keys, quota handling) |
localPortfolioStore patterns |
| Authenticated |
Firebase cloud — authoritative history |
useTrades → TradeService.getTrades |
| UI preferences |
Zustand persist with partialize — chart/filters only |
portfolioStore |
| IndexedDB |
Primarily Firestore client SDK cache; cleared on logout where implemented |
useAuth teardown |
| Google Drive |
User-owned file (pocket_portfolio_db.json) — dumb sync lane |
Sovereign Sync docs / architecture page |
Why “IndexedDB is our database” is wrong
Guests do not use IndexedDB as the primary trade vault — they use localStorage. Authenticated users’ authority is Firebase, with IndexedDB as offline/cache for the SDK. Claiming IndexedDB is “the portfolio vault for everyone” fails procurement diligence in one grep.
Local-first ≠ no cloud
Approved phrases (from docs/command/claims-vs-codebase-calibration.md):
- Client-side aggregation by construction for LLM context.
- Stateless inference handler for portfolio text on
/api/ai/chat.
Prohibited without sign-off:
- “Zero server footprint” / “fully local portfolio” for auth workflows.
- “No cloud” — Firebase, Vercel, LLM APIs, optional Redis/KV are all cloud-adjacent.
Google Drive: sovereign sync, not app logic
Drive holds a user-owned JSON file. The app does not execute portfolio business rules on Google’s servers — it treats Drive as dumb storage for export/sync. That is a credible “user-owned lane” story without claiming we eliminated Firebase for signed-in users.
Engineering takeaway for Part 12
Honest persistence makes the inference boundary (Parts 2–3) believable: we are not pretending the full ledger never existed; we are proving the LLM hop does not require re-uploading it row-by-row.
Part 4 of Sovereign Ingestion & Stateless Inference.
Read the full Sovereign Intelligence book, explore Open Portfolio, or try Pocket Portfolio.