Most US firms use evaluation output in compensation decisions, whether or not the evaluation was designed for it. The form was written to help an associate improve. The committee reads it to decide what somebody is paid, and those are different tests.
What a committee actually needs
A compensation committee has to be able to defend a decision to the person affected and to the partners who disagree with it. That takes three things, and only one of them is a rating.
It needs to know who assessed the associate and how many of them. It needs to know that the scale meant the same thing in one reviewer's hands as in another's. And it needs at least one specific instance behind any rating at the extremes, because an unexplained outlier is where the conversation will go first.
An evaluation process that produces scores without those three produces a number the committee cannot use, and in practice it falls back on the impressions of whichever partners are in the room.
The attribution question comes before the rating
Before anyone argues about whether an associate deserved a 3, somebody asks who filled the form in. This is a reasonable question and a process built on emailed reviewer lists answers it poorly: the instrument went to the people the group suggested, which invites the obvious follow-up about who was not suggested.
Derived rater pools answer it in advance. Nine of the eleven associates who logged more than fifty hours on these matters responded, against a threshold fixed before the cycle opened. That sentence ends the procedural argument and moves the discussion onto what the feedback says, which is where a committee's time is worth spending.
Calibration is the prerequisite, not a refinement
Reviewers use scales differently and always will. A partner who gives nothing above a 4 and a partner who gives nothing below one are both describing real observations, and their numbers are not comparable until somebody puts them side by side.
Calibration before compensation is what makes the ratings commensurable. Without it, an associate staffed mainly by a severe reviewer is paid for the reviewer's habits, and the firm has no way to see that happening. Run it after individual reports exist and before any number reaches a committee paper.
Lockstep changes what the data is for
At a firm on strict lockstep, evaluations do not set the number. They set whether somebody advances a step, whether a bonus is discretionary, and whether a conversation about fit is due. That is a narrower use and it lowers the stakes on the rating itself, which tends to produce more honest reviewing.
At a firm with meaningful merit variation, the rating carries more weight and has to bear more scrutiny. The same instrument serves both, but the amount of calibration and documentation a firm needs scales with how much money turns on the output.
The instrument that should never reach a pay decision
Upward review data should not feed partner compensation.
The argument for using it is obvious: supervision quality is a real contribution and firms say they want to reward it. The problem is what happens next. Associates work out quickly whether their answers affect a partner's pay, and once they conclude that they do, the instrument stops measuring supervision and starts measuring how much an associate wants to affect somebody's income. Scores compress toward the top, written comments thin out, and the firm loses the only source it had on how its partners actually supervise.
The workable arrangement is to say plainly which instruments feed compensation and which do not, and then hold to it. Downward evaluations and self-assessments feed the record. Upward reviews are developmental and reported above a threshold, and a partner's results reach that partner and the people responsible for their development. If a firm wants supervision quality in its compensation model, it should assess it some other way rather than by repurposing the one instrument that depends on candour.
Timing
The review calendar and the compensation calendar are usually set independently, which produces one of two problems. Either the evaluation closes so long before the decision that the data is stale, or it closes so close to it that calibration gets cut to make the deadline.
Working backwards from the compensation meeting is the fix. Reports need to exist at least three weeks before, which puts fielding roughly eight weeks before, which usually means starting instrument design in the quarter prior. A cycle designed forward from whenever the form was ready last year will keep colliding with the same deadline.
What to write down
Whatever the firm decides, the part worth recording is which instruments feed what, who sees which output, and what an associate can ask to see about themselves. Participants behave differently depending on the answers, so the answers are worth being explicit about rather than leaving to inference.
SRA administers downward evaluations and upward reviews on exactly that separation, with calibration packs built for the committee conversation. The methodology sets out what the firm receives, and the wider picture is in the associate evaluation guide.