The Logbook.

Project analysis

A migration needs a way back at every stage

GitHub’s MySQL upgrade is a useful starting point for examining recovery, intermediate states, and the work a migration leaves behind.

Project analysis

A small configuration change can have a large reach

Cloudflare’s July 2019 incident raises a practical question for release reviews: how much of the service can this change affect at once?

Project analysis

Restoring service includes the work that accumulated

GitHub’s October 2018 incident provides a starting point for thinking about data integrity, queued work, and credible recovery criteria.

Project analysis

Test the whole workflow before adding capacity

GOV.UK’s email migration shows why a dependency’s capacity and a service’s end-to-end behaviour need separate investigation.

Ownership Needs Authority

An accountable owner needs room to make the decisions that determine the outcome. Make those boundaries explicit before the work depends on them.

Which Decisions Still Need You?

Repeated requests for a leader's approval reveal something about the operating conditions. Use them to decide where the team needs clearer boundaries, better context, or more practice.

Make Bad News Easier to Bring

A concern is most useful while there is still time to change the plan. A leader's first response helps determine whether the next concern arrives early enough.

Decide Whether to Fund the Next Step

Previous effort cannot answer whether further investment is worthwhile. Review the remaining outcome, uncertainty, cost, and closure work before extending a project.

Commit to the experiment, then examine the outcome

A delivery commitment and an outcome hypothesis need different treatment. Keeping them separate makes a team more accountable for the work it can control.

Test the assumption that could change the plan

Investigation earns its place when the answer changes an investment decision. Start with the uncertainty carrying the most consequence, then choose the smallest credible way to examine it.

Establish the baseline before promising an improvement

A pilot needs a credible comparison, operating safeguards, and a decision to inform. An attractive target cannot replace understanding how the current work behaves.

Stopping an experiment is still engineering work

An experiment can end while its access, infrastructure, data, and obligations remain. Plan closure early so stopping leaves useful evidence and a system someone can operate.

The smallest useful slice still has to work

A narrow release can reduce investment without leaving people to bridge its gaps. Define a complete workflow before cutting scope, including the behavior that matters when something goes wrong.

A rollback does not undo every consequence

Restoring a previous release can leave changed data, external effects, and interrupted work behind. A recovery plan starts with those consequences and identifies how the crew will resolve them.

Sometimes the useful change is a dependency removed

Dependencies consume attention as well as computing resources. Follow one recurring interruption to its source and examine whether the next useful outcome requires the dependency at all.

Generated code and generated tests can agree on the wrong answer

A passing test suite can repeat the assumptions that produced an implementation. Evaluate AI-assisted work against explicit requirements, independent examples, and the consequences of being wrong.

Make Expertise Transferable

Solving a difficult problem is only part of the contribution. Use the next real task to make the reasoning and operating knowledge available to someone else.

Give Preventive Work a Clear Case

Work that reduces risk or recurring friction can be difficult to discuss without inventing savings. Explain the failure mode, proposed control, cost, and evidence you can actually observe.

What Is Keeping This Work Open?

Before starting another item, examine the work already waiting to reach useful operation. Separate active effort from unresolved decisions, blocked dependencies, and missing closure.

Handover Begins Before Release

Operating ownership cannot be transferred in a final presentation alone. Involve the receiving people while they can still influence the design, controls, and support expectations.