Calibration

For a long time, ratings ended in each manager's hands. A manager decided the grades for their own reports, sent them to HR, and that was that. Nobody outside the team ever saw what standard had been applied next door.

The trouble is that people hold different rulers. One manager, wanting to encourage the team, hands the top grade to four people out of six. Another believes the top grade belongs to genuinely exceptional years and gives it to nobody. Two people who did similar work in those two teams end up with a top grade and a middle grade, and the difference came from their managers' habits rather than from their work.

That difference doesn't stop at the grade. Grades usually feed into raises, bonuses, and the shortlist for promotion, so a year's compensation shifts based on having reported to the generous manager. The moment people compare grades over lunch and someone says "apparently everyone on that team got an A," the results stop reading as judgments about performance and start reading as a consequence of team assignment.

Calibration is the extra step bolted on to fix the rulers. Instead of locking in each manager's draft, the managers first sit together, present their own reports, and question each other's reasoning. A grade nobody can support gets adjusted on the spot.

Managers bring their draft grades, present one person at a time, take questions from the room, and leave with adjusted grades that are final.

1. Collect draft grades in advance. Each manager submits a grade and its supporting evidence for every report a few days before the meeting. If people start deciding grades in the room, the two hours go entirely to deciding grades.
2. Put the overall distribution up first. Show how many people landed in each grade, team by team, side by side. If one team's share of top grades is three times everyone else's, that's where the conversation starts.
3. Present one person at a time. The manager states the grade and the evidence behind it. Evidence means things that actually happened during the period, not impressions like "worked hard." What the person took on and how far they carried it goes on the table alongside what colleagues wrote in peer review and 360 feedback.
4. Let the other managers ask. This is the point of the meeting. Someone says "a person at that level was a B on my team — why is it an A on yours?" and if the answer doesn't land, the grade comes down. The reverse also happens: a manager hears the discussion and raises one of their own.
5. Lock in the adjusted grades in the room, and record whose grade changed and why. Without that record, a manager has nothing to say when it's time to deliver the result.
6. Each manager delivers the result to their report. Not a transcript of the meeting, but the reasoning behind the grade in the language the room tested.

The room holds the managers who use the same rating scale, plus whoever facilitates. Five to eight team leads under one org leader is typical, with someone from HR running the session and keeping notes. That HR person doesn't set grades — they put the distribution on screen, watch the clock, and call out statements with no evidence under them.

What decides the boundary of the room is whether the people in it can actually be compared. Engineering and sales do different work and are hard to weigh against each other, so many companies calibrate by job family. Managers of the same job family on different teams, though, always go in the same room. That is precisely where the rulers drift apart.

The timing is the last stage of the review cycle: after performance review has produced each manager's draft, and before anything reaches the people being rated. Run it out of order and the meeting loses its teeth. A manager who has already told someone "you're an A" won't lower the grade, however thin the evidence turns out to be.

A session runs two to three hours and happens once a half or once a quarter. Covering 40 people leaves about three minutes each, so most teams flag the cases that need discussion ahead of time and move quickly through the rest.

"Let's take another look at that grade in calibration."

Not deciding here — waiting to compare against what the other managers bring.

"Please write up the evidence for each of your reports before calibration."

Get the facts you'll present on paper in advance.

"Two A's came down to B in calibration this round."

The evidence was thin against other teams' standards, so the grades were adjusted mid-meeting.

When Pagedeck ran its first half-year review, the six team leads submitted drafts in which the number of top grades ranged from zero to three per team. Grace pulled all six into a room, put the distribution on screen, and had them present one person at a time. Theo's argument for an A on his team came down to "put in a lot of late nights," and Mia asked what that work had produced. No answer came, and the grade became a B.

The same meeting produced an adjustment in the other direction. Sofia had submitted a middle grade for someone, and while she was presenting, another lead added what that person had done on his team's project. The evidence looked different after that. Afterward Grace wrote up what had changed and why on a single page and sent it to all six.

The most common misreading is that calibration exists to force a distribution. The room does put the distribution up and does press the team that sits high, which makes it look that way. But the goal is to level the quality of the evidence, not the percentages. If one team genuinely did the hardest work of the year, three top grades survive the meeting. Start from a target distribution instead and the session turns into picking who to knock down — and from the next cycle on, managers submit inflated grades so they have something to give up.

The other complaint is that the loudest manager wins. That does happen: someone talks at length with nothing underneath and walks out with the grade intact. It's a facilitation failure more than a property of the practice. Cap speaking time per person, accept evidence only as specific facts, and let the facilitator say "that's an impression, not evidence." Left alone, the meeting becomes a debate contest, and from where an employee sits, their lead's rhetorical skill is what set their grade.