Comparison

One is a living page.
One is a fixed record.

They look interchangeable when you are choosing where to put a decision, and they diverge completely eighteen months later when somebody needs to know what you believed at the time. The difference is not features. It is whether the document is allowed to change.

The test

Ask what the team believed in March

Every evaluation should start here, because it is the only question that separates the two. Everything else — search, permissions, editor, templates — both do fine.

The wiki page

Has been edited nine times by five people. The current version describes what the system does now, which is useful and is not what was asked. The reasoning that justified the original choice was tidied away in edit four as out of date.

The decision record

Was written once in March and never touched. It says what was chosen, what was rejected and who objected. If the decision was later reversed, a second record says so and links back — and both are still readable.

Side by side

What each format is actually good at

What you needWiki pageDecision record
Answers "what did we believe in March"
Preserves the option you rejected
Immutable once written
Reviewed in the pull request that caused it
Versioned with the code it describes
Readable by a coding agent in the repo
Long-form onboarding narrative
Comfortable for non-engineers to edit
One page that stays current as things change

Three rows favour the wiki, and they are real. A living page is the correct home for a narrative that should always describe the present — an onboarding guide, a runbook, a system overview. Freezing those would be a mistake.

The honest split

Most teams need both, doing different jobs

The failure is not picking wrong. It is using one document type for two jobs with opposite requirements — one that must always be current, and one that must never change.

Wiki for the present

How the system works today, how to run it, how to join the team. Should be edited freely and should always describe now.

Records for the past

What was decided and why, frozen at the moment it was true. Superseded by new records, never overwritten by them.

Neither for the habit

Both formats assume somebody sits down afterwards and writes. That assumption is what fails, in both cases, for the same reason.

The real problem

Neither format fails. The writing step does.

Worth stating plainly, because choosing between the two is usually a way of avoiding the thing that actually broke.

Almost every abandoned documentation system was abandoned for the same reason, and it is never the template. The decision gets made across fourteen commits, an argument in a review, and an incident channel at two in the morning. By the time anyone could write it down, the work is finished, the context has gone cold, and the next thing is already urgent. Most ADR directories stop at record four. Most wikis go stale in a quarter.

The version that survives is the one nobody has to remember: an agent that reads the work that already happened and writes the record itself, in whichever format you keep. The format debate matters much less than the fact that a record exists at all.

Questions

What teams ask before deciding

01Can I just use a wiki page for decisions?+
You can, and plenty of teams do. It works while the team is small enough that everyone remembers the conversation. It breaks at the point where the page has been edited nine times, nobody can tell which version was the decision and which was a later revision, and the person who wrote it has left.
02What actually makes an ADR different from a wiki page?+
Immutability and scope. One decision per file, written once, never edited — if the decision changes, a new record supersedes it and both stay readable. A wiki page is a living document, which is exactly right for a guide and exactly wrong for a record of what you believed in March.
03Do ADRs have to live in the repository?+
No, but it is the reason most teams that stick with them do. Decisions versioned alongside the code they describe get reviewed in the same pull request, travel with a fork, and are readable by anything that can read the repo — including your coding agent.
04Why do most ADR directories stop at four or five records?+
Because writing the record is a separate act of will after the work is done. The format is not the problem — the discipline requirement is. Every team that has abandoned ADRs abandoned the habit, not the template.
05Should we migrate our existing wiki?+
Almost never worth it. Leave the wiki where it is and start recording new decisions as they happen. A migration project converts stale pages into stale files, and it spends the enthusiasm you needed for the habit.
06What about Notion or Confluence with AI turned on?+
Summarising a page you already wrote is a different problem from producing the record nobody wrote. The stale wiki is not stale because it lacks features — it is stale because the writing step competes with shipping. AI on top of the storage does not remove the step.

Pick a format in a minute. Fix the habit in an afternoon.

Generate your first decision record from a merged pull request, then decide whether writing the next hundred by hand sounds likely.