Interlock

Results: an explained escalation against real Stripe (test mode)

Generated 2026-09-14T01:00:20+00:00 by experiments/escalation_live.py. Every step made real Stripe test-mode API calls on one PaymentIntent, pi_3UFOFi88KhIqqdFL1ueAaUx9. No mock data. The journal is results/escalation_live.jsonl; the receipts, verify output and scoreboard are in results/escalation_live.json.

Workflow time (at, SLA, decisions) is an injected clock advanced by the script, so a 4h SLA passes without waiting 4h. Stripe calls, refund ids and webhook deliveries are real and wall clock.

What happened

step wall clock (UTC) detail
payment 2026-09-14T01:00:11+00:00 payment_intent=pi_3UFOFi88KhIqqdFL1ueAaUx9, amount=10000
listener 2026-09-14T01:00:13+00:00 ready=True
request 2026-09-14T01:00:13+00:00 request_id=refund-50/pi_3UFOFi88KhIqqdFL1ueAaUx9, status=QUEUED, reason=needs_judgment, detail=[‘amount at most 1000 cents’], group=ap-leads
approved 2026-09-14T01:00:13+00:00 by=ana, status=APPROVED
hand refund 2026-09-14T01:00:15+00:00 refund=re_3UFOFi88KhIqqdFL1Qpt3qv8, amount=3000, status=succeeded
send 2026-09-14T01:00:16+00:00 status=REFUSED:stale_premise, reason=stale_premise, changes=[{‘field’: ‘refunded_by_others’, ‘was’: 0, ‘now’: 3000}], repairs=[‘still_fits’]
repair accepted 2026-09-14T01:00:17+00:00 by=ana, repair=still_fits, status=COMMITTED
second request 2026-09-14T01:00:18+00:00 request_id=refund-15/pi_3UFOFi88KhIqqdFL1ueAaUx9, status=QUEUED, group=ap-leads
sla tick 2026-09-14T01:00:18+00:00 moved={‘refund-15/pi_3UFOFi88KhIqqdFL1ueAaUx9’: ‘finance-manager’}, group=finance-manager, level=1, breach=True
ap-leads after breach 2026-09-14T01:00:18+00:00 by=ana, status=REFUSED:lease
approved up the chain 2026-09-14T01:00:19+00:00 by=fm, status=COMMITTED
confirmation 2026-09-14T01:00:20+00:00 path=webhook

The reviewer saw, on the re-escalation: “a fact this action depends on changed since it was decided. Changed: refunded_by_others: was 0, now 3000. Suggested, needs rules or a person: 5000 still fits: 7000 of 10000 is left to refund, if the other refund was not this one”.

Refunded on the payment: 9500 cents ($30 by hand, $50 by the agent once, $15 after the SLA escalation). Refund ids: re_3UFOFi88KhIqqdFL1Qpt3qv8 (hand), re_3UFOFi88KhIqqdFL1ebviQSB, re_3UFOFi88KhIqqdFL1obdhASH.

Confirmation by the target

Path used: webhook. stripe listen forwarded refund events to a local server that passed the raw body and Stripe-Signature header to confirm.confirm_event. Deliveries:

Events for refunds this journal does not know (the hand refund, other test traffic on the account) are ignored.

verify()

request valid happened once authorized assumptions held approved by approval verified confirmed by target escalations
refund-50/pi_3UFOFi88KhIqqdFL1ueAaUx9 True True True True True ana True True 2
refund-15/pi_3UFOFi88KhIqqdFL1ueAaUx9 True True True True True fm True True 2

Problems: none and none. Receipts are unsigned (no key), so they prove internal consistency, not that the journal was never rebuilt.

Scoreboard (derived from the journal)

field value
requests 2
cleared_no_person 0
cleared_verified 0
sent_after_person 2
rejected 0
closed_by_repair 0
open 0
escalated 2
stale_approvals_caught 1
crash_to_person 0
repairs_suggested 1
repairs_accepted 1
sla_breaches 1
confirmed_by_target 2
no_person_share 0.0
escalated_by_reason {‘needs_judgment’: 2, ‘stale_premise’: 1}
time_to_decision {‘n’: 3, ‘median’: 1200.0, ‘p90’: 15360.0, ‘max’: 15360.0}

Both requests were over the auto-approve limit by design, so no request cleared without a person here. This run shows one story working end to end; it is not a measure of how many approvals Interlock removes. Goal: cut two thirds of manual agent approvals.

This site is open source. Improve this page.