What this demo does not do yet
Part of building this properly is being explicit about scope. This release focuses on stock-count discrepancies: unit conversion, evidence separation, and honest handling of missing or failed data. The items below are deliberately out of scope for this stage, not silently skipped.
Shipment reconciliation
Comparing dispatched vs received shipment quantities as an independent check is a separate module. This demo only surfaces a shipment as a possible, unconfirmed explanation for a stock difference.
Transaction ledger checks
Sales, returns, transfers and adjustments would need their own event feed and reconciliation logic. Not implemented in this release.
Raw RFID event integrity
Detecting duplicate tags, reader faults, or mapping conflicts requires raw scan event data, not just aggregate counts. Not implemented in this release.
Independent accuracy evaluation
No formal accuracy benchmark exists yet. Nothing on this site claims a percentage-accuracy figure for that reason.
Live AI narration
This demo shows deterministic facts only. No AI model is called at runtime, by design, so there is nothing to fail, hallucinate, or cost money on a public link.
Enterprise operations
Access control, backups, multi-tenant isolation, benchmarking at scale, and incident response are real requirements for a production deployment and are a separately scoped future stage.