Status: Accepted
Date: 2026-07-22
Decider: Chelsea Kelly-Reif
The product hypothesis requires actual alternate-system recovery, but no lawful design-partner export or target has yet been selected. Inventing a vendor-shaped connector would create misleading confidence and make an unvalidated abstraction look permanent.
Implement the denominator, normalization, integrity, neutral-restore, result, privacy, and receipt seams using invented synthetic contracts. Label every successful outcome as structural only.
A future strong restoration result requires:
- one documented native export;
- a separately captured baseline;
- load through a supported interface into an empty production-capable sandbox;
- independent target read-back; and
- declared workflow and allow/deny permission probes.
- Build generic connectors now: rejected because connector maintenance may be the reason to kill the product.
- Call SQLite a target system: rejected because storage representability is not operational recovery.
- Wait for a design partner before any code: rejected because the strict baseline, loss algebra, receipt privacy, and fail-closed boundaries are cheap technical risks worth testing independently.
- Structural evaluation alone cannot demonstrate the full product promise.
- Schemas may be replaced after the first native export.
- Product discovery remains the critical path.
- The implementation provides a safe test harness for silent-loss fixtures.