1.A mis-selling call, summarized and flagged
Open the call →DoTap the line link next to the risk flag. Then pick Today's model (stand-in) and Family floater, mother's diabetes.
Check"Pakka 15%" is flagged and linked to its line. On the family floater, today's model returns no summary: a critical miss.
2.Ask the Advisor Copilot
Ask it →DoRead the answer, tap a clause tag such as WP 2.1, then tap Continue.
CheckThe small model refuses to promise 12% and cites the wording. The language check fails, but it is not critical, so nothing is rerouted.
3.Scan a claim file
Run it →DoRead the Diagnosis row, then open Aadhaar (masked) and press Run the harness.
CheckThe diagnosis is misspelled at 100% confidence: confidence alone is not enough. On the Aadhaar, the large model takes over.
4.See which model ships
Open Compare →DoRead Compare, then open the Scorecard tab.
CheckOn 180 calls: no candidate passes all four gates at the 85% bar and 15 s, so nothing ships yet. On the 18-call scorecard, closed-source model and open model + examples pass the release gate; open model is blocked.
5.Share it
Open the Brief →DoPress Print or save as PDF.
CheckOne page to forward: what each module proves and what we need to start.