A discount at exactly the configured ceiling was routed to a manager instead of auto-approving. The cause was 0.00004 of a penny, and the textbook boundary pair walks straight past it.
Scoping full regression for a managed-package uninstall across three teams and 600+ stories — how one triage question turned an unbounded problem into 161 defensible variants.
Six conclusions reached in one day, then withdrawn. Five came from reasoning about the UI instead of reading the source. A simple evidence grading system that separates confident findings from correct ones.
A satirical manifesto of seven delivery anti-patterns — formalism over common sense, testing in production, people as replaceable parts. Written from real projects. If one sounds familiar, you have a problem.
We planned to build a controllable clock to assert selectability at 08:59 versus 09:00 — then dropped it. Deciding which boundaries you own is a test design skill, not laziness.
An approval control checked an undocumented metadata list before it loaded any thresholds — and anything on that list skipped the manager entirely. Why "undocumented but working" is handover debt, not a defect.
The integrations worked. Around 90% of what was logged as "integration errors" turned out to be test-data state nobody on the team controlled. How to tell the difference before you waste a day.
A technical design was rewritten between two reads — new title, new authors, a deleted section, and a change log still claiming v0.1. What that does to your test cases, and how to catch it.
A waived fee usually reaches the billing system as the full value plus an offsetting discount. A test asserting "fee = 0" passes against the broken build. How to test waivers properly.
An order submits cleanly, then fails silently inside the Orchestration Plan — a missing notification, an invalid catalog mapping. Why every work item is its own assertion.