← All decks

Engineering Practice Patterns

Ship small changes often and sleep at night.

The daily habits of a software team that releases often without dreading it, for developers, leads and anyone running a codebase. The cards cover designing before building, changing code in small reversible steps, testing, reviewing, releasing, watching production, handling on-call and incidents, and keeping the code healthy and secure for years. Nothing depends on a particular language or vendor.

56 cards · 8 suits · 5 recipes · 2 missions

What's inside

• Design7Have you compared designs before writing code?
• Change7Is each change small and easy to undo?
• Test7Will you know when it stops working?
• Review7Do changes get reviewed within a day?
• Release7Can you ship without holding your breath?
• Observe7Would you notice trouble before users report it?
• Respond7Is there a calm plan for the next incident?
• Sustain7Are debt and security getting regular attention?

A few cards

Design

Design Doc First

Argue with a page of prose before anyone argues with a thousand lines of code.

Test

Test Pyramid

Lots of fast, small tests at the base; a few slow, broad ones at the top.

Release

Continuous Integration

Merge to the shared line often, let a machine prove it still works, and treat a red build as everyone's emergency.

One card in full

Respond

Sustainable On-Call

Design the rotation so people can rest between shifts, hand over in writing and see how heavy the load really is.

On-call that relies on a few willing people every other week works until those people leave. A sustainable rotation spreads the load, protects sleep and makes the cost visible, so the team fixes the noisy systems instead of absorbing them.

How to use it

1. Put enough people on the rotation that each is on call no more than one week in four, or split the day across time zones if you have them.
2. Hand over in writing at every shift change: open incidents, anything fragile, recent changes worth watching.
3. Log every page with its time, whether it woke someone and how long it took. Review the totals monthly.
4. Agree a limit, such as two overnight pages per shift, beyond which the following day is free and the noisiest alert gets fixed first.
5. Give the on-call person no feature work. Their job is interruptions and making the next shift quieter.

Example

A hypothetical hospital IT team rotates six people weekly with a written Monday handover. Their page log shows one integration causing most night calls; fixing it halves them.

Watch out

• Treating heavy on-call as a personal resilience issue. Load is a system property; fix the system.

It depends on Alert On Symptoms and Runbook Per Alert; for the human side, see Burnout Signals.

Recipes

• First Hour Of An Outage: Set the severity, put one person in charge, stop the harm before chasing the cause, keep everyone informed at a fixed rhythm, then study what happened without blame and find every condition that let it happen.
• Pay Debt While Shipping: List known debt with the cost it adds, sort it by how it was taken on, let the reliability budget decide when to slow down, keep changes small and clean up a little on every visit.
• Risky Change, Calm Release: Agree the design on paper, split data changes so they stay reversible, hide the new path behind a switch, expose it to a sliver of traffic, keep undo one step away and remove the switch once it has served its purpose.
• Tame Old Code: Record what the old code really does, reshape it until the change is simple, swap internals behind an interface, move whole capabilities out piece by piece and leave each file tidier on the way.
• Unblock The Review Queue: Catch your own leftovers before anyone else sees them, tell the reviewer why the change exists, hand style to tools, label every comment by how much it matters and answer requests within a working day.

Missions

• Draft a design doc: from "we need to build this" to a reviewable design doc with goals, non-goals, two designs compared, a decision record, a rollout plan and a three-point estimate.
• Write up an incident: from a raw timeline, chat log and alerts to a blameless post-mortem with impact, contributing factors, what went well and owned follow-ups.

Related decks

UX Principles

Shape interfaces around how people think.

54 cards

Productivity Patterns

Do your most important work without burning out.

56 cards

AI at Work Patterns

Use AI assistants without losing your judgement.

56 cards

← All decks