CounterSignal / Judge Console

Try to kill the hypothesis.

A frozen customer-discovery protocol where negative evidence is a first-class outcome. This deterministic reviewer mode uses the exact pre-registered SmallBet decision rule and never places a phone call.

Deterministic reviewer mode · NO CALL

PRE-REGISTERED EXPERIMENT

Permit-status ambiguity recurs often enough that contractors already spend operator time on a manual workaround.

smallbet-permit-ops-v1protocol a7229d00…b47a565 questions locked

Tell me about the last time a permit status was unclear or did not match what your team expected.
What did your team do to resolve it?
Roughly how often has that kind of follow-up happened in the last month?
What happens operationally if nobody follows up?
Who or what currently keeps track of those exceptions?

Experiment authority: a call can contribute evidence. It cannot change the questions, rewrite a contradiction, or declare product-market fit.

8answered denominator

5supporting

0disconfirming

0nonresponse

0invalid / ungrounded

EVIDENCE LEDGER

CALL-E BOUNDARY

metadata.workflow_type = falsifiable_customer_discovery metadata.experiment_id = smallbet-permit-ops-v1 metadata.protocol_hash = a7229d00ec935e760d5764572b142a33a062095db0df8c5f3f32c18b88b47a56 result must bind to: ✓ exact accepted CALL-E call id ✓ exact reviewed recipient ✓ exact protocol hash ✓ recipient-side transcript quote refusal / voicemail / unreachable → nonresponse low-confidence / ungrounded / mismatched → invalid ambiguous provider outcome → outcome_unknown / NO BLIND REDIAL

CLAIM BOUNDARY

The decision is a pre-registered operational rule, not a population-level statistical estimate. “Would take a follow-up” is never treated as demand. Silence never enters the answered denominator.

Permission provenance: the real dogfood run only dials respondents who affirmatively opt in to the AI interview. A public business number is not treated as permission.