Round 3: the last two places that render "cannot tell" as "fine"

[sylvesterdamgaard]

The scripted fake dropped the off-session guard — third instance

ScriptedPaymentGateway extends FakePaymentGateway, whose entire purpose after this campaign is refusing what production refuses — then overrode charge() and threw it away.

None of the five dunning suites vaulted an instrument, so every intent they produced was non-off-session. The whole adaptive-dunning, smart-retry and card-updater surface was green against a pending result the real Stripe and Mollie gateways cannot return — both refuse such an intent with FAILED, which opens dunning. The tests asserted the opposite.

Fixed on both sides: the fake defers to the parent's refusal before consulting its script, and the five suites vault a default method, as any account actually being dunned necessarily has. Twelve tests failed when the guard was restored — that is the measure of how much was being asserted about a call production would never make.

ReportNexus printed "nothing to do" for a board that measured nothing

It read only triggered() and approaching(), so a deployment whose dataset location is unreachable — every state unknown — got "No US states have triggered or are approaching" and a SUCCESS exit.

The alert emitter and the console were fixed in earlier rounds; this scheduled command was the last one left, and it's the one an operator reads in a cron log. It now names the unmeasured states, says plainly this is not "below", points at the env var, and exits non-zero.

Gate

pint · PHPStan level max (0 errors) · 1034 tests — green.