Three separate things happen inside a performance review, and almost every company I've worked in runs all three through a single document, then wonders why the document satisfies nobody.
Running review cycles at Honor, Lyft, Grammarly and Pitch, for teams from a handful of people to dozens later, I've written a few hundred of these and been on the receiving end of about twenty. I've also sat in the calibration rooms where they get decided. That is where you learn what a review is really for, and it is not what the form implies. One I sat with asked, in its first step, whether this was a promotion, a raise, a PIP or an exit. Everything after that was written to justify an answer already given.
The three jobs
A review is asked to do three jobs at once. Help someone improve, rank them against their peers, and decide what they get paid.
Development is the conversation about what someone should get better at. It works when it is specific, frequent, low-stakes and forward-looking.
Ranking compares this person against their peers at the same level. It works when it is evidenced, consistent across managers, and written for people who have never met the person.
Money divides up a fixed pool. It works when it is predictable and explainable.
Those three have almost nothing in common, and running them together is the whole problem. Development wants candor. Ranking punishes it, because anything you write about a weakness becomes an exhibit. Development wants frequency, while ranking and money run on an annual or semi-annual clock because that is what finance can absorb. Development is about a single person. Ranking is about a distribution.
Fuse them and you get the document everybody recognizes: warm, slightly evasive, heavy on adjectives, and carrying one buried sentence that will decide the outcome. A manager writes defensively because they know the audience. A person reads it looking for the number. Nobody develops.
Calibration is the real meeting
Most of a review is written for a room the person will never sit in.
Calibration is the meeting where managers put their people up against the written standard for that level, and argue. It is the most consequential hour in the cycle, and almost nobody being discussed has a clear picture of it. I think that's a straightforward failure of transparency rather than a necessary secrecy.
What actually happens in that room is worth knowing, because it changes how you write.
Managers who tell a story lose to managers who bring evidence. A vivid narrative about how someone grew this year gets picked apart in ninety seconds by anyone with a level rubric in front of them. Three examples, each with a result and a colleague named, survive. Three, not one, and not ten. That happens because the room has eleven other people to get through and no way to audit taste, not because anyone in it is cynical.
Second thing is uglier. Visibility skews all of this, everyone knows it, and nobody has fixed it. Somebody who worked on the surface the executives use every day is easier to advocate for than somebody who spent the year making a migration land without incident. I have watched excellent glue work fail calibration because nobody in the room had a way to see it. What a manager can do is spend the year deliberately creating traces of that work in places other people read, rather than complaining that the traces do not exist.
Managers who tell a story lose to managers who bring evidence. That room is not cynical. It has eleven other people to get through and no way to audit taste.
What the season does to the work
There is a cost people rarely price in. For about six weeks either side of a review cycle, a company gets measurably worse at its actual job.
Managers disappear into writing. Everyone else quietly shifts their attention to what will read well, which is a sensible response to a real incentive and not a character flaw. Cross-team favors dry up, because a favor is invisible in a rubric. Risky work gets deferred to the following quarter, since nobody wants a half-finished experiment as their headline.
Run that twice a year and you have given up something like three months of ordinary behavior to the measurement of behavior. That is a serious price and it should be argued for explicitly rather than absorbed as a fact of nature. I do think it is worth paying once a year. I am not sure it is ever worth paying twice.
The rule worth more than the rest of it
Nothing in a written review should be new information to the person reading it.
If that rule holds, the review becomes a summary and a record, which is all it was ever any good at. If it breaks, you have used a formal process to deliver feedback you were too uncomfortable to give in March. In a document that goes in a file.
I have broken this rule. Not often, and never with anything disastrous. But I have let a thing I noticed in the spring turn up in writing in the autumn, because in the spring I wanted three examples and a well-formed argument before I said anything. I've written about that habit in the context of 1:1s. Here it shows up in its most expensive form. The delay is now in a permanent record, and the person is reading it for the first time with their manager watching their face.
Mechanically: if you are about to write a sentence you have not already said out loud, stop writing, book fifteen minutes, and say it. Then write it.
The part nobody puts on the form
A review is also how the money gets rationed, and pretending otherwise is what makes people cynical about it.
There is a fixed pool. There is usually a soft expectation about distribution as well, whether or not anyone will admit to a curve (nobody ever admits to a curve). Which means that in a given cycle, some number of people who did strong work will receive an ordinary outcome. Their manager then has to explain it in language that implies it was about their performance, when it was largely about the arithmetic.
I do not think you can fix this from a manager's chair. I do think you can decline to lie about it. Tell them the work was strong, the outcome was ordinary, and the constraint was the pool rather than them. It's a hard conversation. It also leaves them able to trust the next thing you say. Otherwise you invent a development area to justify the number. That is the single most corrosive thing a manager can do. The person now has a false picture of their own performance, and will carry it into their next job.
That distortion is common enough that I can usually spot it on a reference call years later. Someone whose sense of their own level is off by a full step, through no fault of their own.
What I would change
Split the three jobs. Development runs continuously and stays out of the file. Ranking happens once a year, in writing, against a rubric the person can read before it is applied to them. Then hold the compensation conversation on a separate day, so the money does not colonize the feedback.
And publish what each level requires. If people cannot read the standard they are measured against, you are just handing down a verdict.
I've run review cycles for design, research and product organizations with dozens of PMs, designers and engineers, at Honor, Lyft, Grammarly and Pitch. One counter-argument. A number of good companies have abolished ratings entirely and report better outcomes. Read honestly, that evidence says the ranking job may be worth less than it costs. I have not run a company without ratings, so I am arguing from inside the system I know.
© 2026 Renato Valdés-Olmos