Peer Review

When a manager was the only person writing evaluations, only what a manager could see got evaluated. That means deliverables, dates, and whatever was said in meetings. Whether someone spent two hours unblocking the person next to them, or absorbed four urgent requests from another team, happens outside the manager's field of view.

So results drifted away from contribution, over and over. At Pagedeck, Theo spent an entire half handling another team's data requests for them. None of that work was on his assigned list, so none of it reached his written evaluation either. In the same half, the person who talked the most in meetings came out looking busier than the work supported. The manager wasn't being lazy — the sample visible from a manager's seat is simply narrow.

From the employee's side the problem is blunter: the people who know your work best have no say in how it is judged. The colleague who trades code reviews with you every day and the designer who sat with you on the same spec for two months contribute nothing to the record.

Peer review exists to fill that hole. It doesn't replace the manager's judgment; it widens the material the manager judges from. That's why it usually isn't published as a separate result at all — it flows into the manager's write-up.

The cycle has five steps.

1. Settle the reviewer list. The employee names three or four people they actually worked with this period, and the manager looks the list over. If someone shows up in the work records but not on the list, the manager adds that name.
2. Write the questions. Fewer than five items works better than more. Prompts that make the writer recall a specific moment — "what helped most when working with this person," "if they changed one thing next quarter, what should it be" — produce usable answers. "Rate their collaboration from 1 to 5" mostly produces fours.
3. Give people time to write. Someone reviewing four colleagues needs a week to ten days. Written the night before the deadline, every answer collapses into "doing great."
4. Collect and sort. Separate the points six people raised from the points one person raised before handing anything to the manager. One strong opinion and six people agreeing carry different weight, and passing the raw pile along means the last thing read lands hardest.
5. Deliver it to the person. Most often the manager reads through it with them during the performance review — the process that turns a period's results and capabilities into a judgment. Peer review enters that judgment as one input among several.

Step 5 restarts the same argument every cycle: show the original wording or not. Pass it through untouched and the author becomes guessable; smooth it out and the point goes blurry. Most teams settle on keeping the facts and editing only the phrasing. "I asked why it was built this way and never got an answer" survives as written. "Difficult to work with" gets sent back with a question about what was difficult.

Three to five reviewers per person is the workable range. Two, and you get two friends. Eight, and the burden on the writers grows until the answers shrink. In a 45-person company where everyone writes four reviews, that's 180 pieces of writing — a week of work spread across the whole organization before anyone reads a word of it.

Letting people pick their own reviewers is the default, but the list doesn't stay as submitted. Picking friends is easy, and collaborations that involved friction quietly go missing. The manager checks the quarter's project records and confirms that the people who were actually in the room are on the list. Adding those names is the largest thing a manager does in this process, and it is the one step that decides whether the round produces anything useful.

The timing sits ahead of the performance review. On a half-yearly cycle, peer review has to close two or three weeks before write-ups begin, so the manager can read it before forming a view. Reverse that order — reviews written first, peer input arriving after — and the peer input becomes a set of quotes used to support a conclusion already reached. After the write-ups comes calibration, where managers meet to line up their standards against each other.

Once every half year is the most common rhythm. Run it quarterly and collaborators rarely change enough in three months to produce different sentences. The gap between cycles is covered by feedback — the running conversation about what went well and what should change. Peer review is that conversation written down twice a year.

"Have you picked your reviewers yet? Please include Theo, you two shipped the payments work together."

A manager heading off a list made up entirely of people who get along.

"Three separate people raised the same thing in your peer reviews — that you don't leave written notes behind."

Not one person's impression, but a point several people arrived at independently.

"This isn't something to save for peer review. It sounds like a conversation you should have today."

Six months and a form are the wrong container for what's being described, and saying it out loud now costs less than writing it later.

Pagedeck runs the cycle every half year. Mia submits the people she worked with, and Grace checks the project records and adds anyone missing. Last half Mia listed three names, Grace added two more to make five, and one of the additions was Noa, who had clashed with her a few times over specs.

What Noa wrote was short and specific: screen requirements shifted a little in every meeting, so the same screen got redrawn three times. Mia started dating the requirements she considered locked. Her rating didn't move because of that paragraph. The way she worked did.

The most common reading is that peer review is a popularity contest. Colleagues write it, so friends should do well. And if you let people pick their own reviewers and feed the output straight into a rating, that's exactly what happens. The two safeguards against it are the manager checking the list, and the results entering as reading material rather than as a number in a formula. Drop both and the company holds a twice-yearly ritual of saying nice things about each other.

The other mistake is expecting something to change the moment the reviews land. Six people naming the same issue adds one item worth working on next half — that's the size of it. The decisions still belong to the manager and the person. Convert the results directly into ratings or promotion calls and nobody writes honestly in the next round.