Plan: Data model, event log, and indexing :: Vault

Problem statement

Vision sketches layers (event log, document store, graph index, credential anchors) but the repository does not yet implement them as schemas, migrations, or query APIs shared across desktop, web, and extension.

Current state (as-is)

Desktop: demo graph visualization without a committed persistence model.
Web: Veramo JSON stores in browser localStorage for DIDs/VCs; merge append file for provider JSONL.
Specs: OpenAPI + JSON Schema for interchange exist separately from a unified internal vault schema.

Target state (to-be)

Internal canonical model: events referencing blobs by hash; explicit supersedes / revokes for corrections (per vision).
Derived indexes: full-text or embedding-backed search with provenance (which model, which file region, when).
Versioned migrations for any on-disk vault format.

Phase 0 — Canonical event schema

Define event types (enum or closed vocabulary), JSON Schema, and example fixtures.
Align field names with interchange bundle where overlap makes sense; document mapping where not.

Acceptance: validators run in CI on fixtures; desktop or site can append/read a minimal event stream.

Phase 1 — Blob store

Content-addressed storage for large files; streaming hash API; dedupe.
Garbage collection policy for unreferenced staged blobs.

Phase 2 — Graph index

Derive edges from events (e.g. “document A supports claim B”) for navigation views.
Optional export to standard graph serialization for tooling.

Phase 3 — Search and multi-modal proposals

OCR / speech / video outputs as proposals with review queue (per vision); never auto-commit high-stakes fields without explicit user action.

Dependencies

Desktop and Web for where the event log is authoritative.
GPU/CPU capability detection for model workloads.

Risks

Schema churn without migrations corrupts user vaults — treat migrations as first-class with tests.
Privacy of embeddings and cloud model calls — default local-only unless user opts in.