ML circle thread #13: bias testing that changed a launch
ML circle thread #13: bias testing that changed a launch
Ground this in something you measured or shipped — theory welcome, receipts preferred.
Thread index 13 — add your angle.
15 replies
Reading old tickets was archaeology that paid better than guessing anew. In ML practice (thread 13): The vendor demo lied by omission; our staging environment told the truth. We should have named a backup approver for production deploys before vacation season.
We learned that customers appreciate when you sunset features with timelines and export paths. In ML practice (thread 13): We learned that small trustworthy releases beat big risky bangs for morale. The vendor integration succeeded when we owned retries, not when we blamed latency.
The mentor who said 'write the decision and the rejected alternatives' improved future audits. In ML practice (thread 13): We learned that humour in retrospectives helps if it focuses on systems, not individuals' quirks cruelly. We learned that transparent promotion timelines reduce anxiety more than surprise bonuses.
The design that included offline states first saved rural users real frustration. In ML practice (thread 13): We finally admitted our test data did not represent production shape at all. We learned that customers trust changelog honesty more than marketing superlatives.
The mentor who said 'sleep, then ship' was annoying and correct. In ML practice (thread 13): The flaky canary almost masked a real regression — canaries need care too. The integration that surfaced partial failures prevented silent half-updates in billing.
The smallest improvement to CSV decimal separators reduced international finance import errors sharply. In ML practice (thread 13): The mentor who said 'write the customer-facing timeline before the internal one' improved incident comms. We should have named a backup on-call before the primary got food poisoning on launch day.
We learned that transparent incident customer comms templates reduce legal review thrash later. In ML practice (thread 13): The quiet deletion of unused roles simplified audits and new hire comprehension. The flaky test that depended on wall clock time taught us to inject clocks in tests.
The mentor who said 'show me the unit economics' sharpened growth vs burn debates usefully. In ML practice (thread 13): The integration that validated schema versions prevented silent consumer crashes on deploy. The mentor who said 'show me the user journey map' ended abstract architecture debates.
We should have named a DRI for circle recommendation ranking before launch — opaque ranking breeds conspiracy theories fast. In ML practice (thread 13): The flaky smoke suite that ran only nightly missed regressions that hourly would catch. The mentor who said 'prove it with a cohort chart' ended faith-based growth debates.
Copy-paste from Stack Overflow without tests is not 'moving fast' — it is gambling. In ML practice (thread 13): We learned that transparent incident customer comms templates reduce legal review thrash later. The architecture review that asked about child safety workflows for public circles changed moderation staffing plans before launch measurably helpfully quarterly always.
The smallest improvement to keyboard navigation made power users noticeably happier. In ML practice (thread 13): The migration rollback rehearsal found a missing grant nobody would have caught live. The architecture spike that listed rate limits for replies per thread early prevented flame-war amplification under viral posts quietly.
We finally admitted the monolith was fine and deleted six microservices nobody needed. In ML practice (thread 13): We should have named a communications approver for incidents before execs posted early tweets. We stopped shipping 'just log it' without a query plan for how humans will read it.
We traded sleep for a deadline and paid interest on that debt for a quarter. In ML practice (thread 13): The integration that surfaced partial failures prevented silent half-updates in billing. We should have deleted unused API keys embedded in old Postman collections — security scanners still find them years later quarterly always.
We learned that small kind gestures in code review compound into culture. In ML practice (thread 13): We learned that 'done' includes rollback notes, not just merge to main. We should have invested in synthetic login journeys before Black Friday traffic doubled.
We should have versioned the API contract before two mobile teams diverged. In ML practice (thread 13): The design that considered assistive tech early avoided costly retrofitting later. The roadmap slide was fiction; the issue tracker was closer to reality.
Join the conversation.
Log in to reply