mirror of
https://github.com/hanzoai/idv.git
synced 2026-10-30 18:33:39 +00:00
LexisNexis reported "approved" out of InitiateVerification for any non-4xx reply and again from CheckStatus with no request at all; Intellicheck issued its request and then ignored the response, including its status code, so a 500 and a rejected document both read as approved. None of the three read what the provider decided. A caller gating on any of them would admit an unverified subject. An integration that does not parse the response now refuses instead of answering, and the refusal names the provider so an operator can tell it apart from a network fault. LexisNexis loses the query builder nothing called; implementing it means parsing the FlexID verdict and mapping it to a status. Two properties are now table-driven over every constructible provider, so one added later inherits both: a freshly created verification is never already approved, and an unread status read refuses. The tests point at a server that ANSWERS — an unreachable base URL would pass for the wrong reason, leaving the synthesized approval downstream of a return no test reaches. scripts/mutants.py drives cloud's strict-scored engine over both properties and the webhook guards: 6 rows, 6 KILLED on a green baseline. LLM.md now states per provider whether it can read a verdict and whether it can prove a webhook, and carries the retirement runbook — sequenced on the route, since an orphaned Ingress publishes idv.hanzo.ai today despite its CR disabling ingress. Co-authored-by: Hanzo Dev <[email protected]>