Performance reviews without corridor evidence: what to write down

A review built on memory rewards the people who talk most, not the people who produce most. Ravi’s rating said his impact was unclear, and the rating was wrong.

The review that said unclear

Ravi had been in the support team for three years and took the tickets nobody else wanted. His written review said his impact was hard to see.

The rating was not malicious. His manager had six reports across two countries and remembered the meetings rather than the queue of unresolved cases.

What the manager remembered was who spoke in the weekly call. That is a memory test, and it measures confidence rather than work.

Corridor evidence was never evidence

In an office, a manager collects impressions between meetings. Those impressions are unreliable, but at least they are based on something observed.

Remote work removes the corridor but leaves the impression behind, intact. What remains is whatever the loudest channel happened to contain that month.

The result is a review that measures presence in conversation rather than output. Quiet competence scores low, and the low score is defended as judgement.

Start from the artefacts

We rebuilt Ravi’s review from four sources: closed tickets, incident notes, code comments, and requests for help. Three of those were written by other people.

None of those required him to perform. All of them were timestamped, and three of them named him inside someone else’s thread.

Artefacts are not a complete picture of a person. They are a better picture than a manager’s recollection of who was memorable in March.

Write a sentence a stranger can check

Handling the billing queue migration without a customer-visible outage is checkable. Bringing a strong technical presence is not checkable by anyone at all.

The test is simple. Could a new manager verify the claim in ten minutes, using systems they already have access to? If not, delete the claim.

We rewrote all six reviews this way over two afternoons. The documents became shorter, duller, and far more useful in the calibration meeting.

Volume is not the same as output

Ravi closed fewer tickets than the team average, because he took the ones that took three days. Volume alone would have rated him below a colleague he trained.

We now record difficulty alongside count. A ticket requiring four systems counts differently from a password reset at nine in the morning.

This is imperfect and everyone in the team knows it. It is still better than a metric that quietly punishes people for taking the hard work.

Account for the weeks nobody saw

Two of Ravi’s months went into a migration with no visible output until the end. A monthly review would have scored both months as empty.

We now ask people to flag long-running work at the start, with an expected date. The flag costs one sentence and prevents a wrong rating.

Long work also needs intermediate markers. A design note in week three is evidence of progress even when nothing ships until week nine.

Let people nominate their own evidence

Each cycle opens with a form where the person lists three things they would point to. Ravi’s list included an escalation guide nobody had asked him to write.

That guide cut the average handling time for two colleagues. Nothing in the manager’s notes mentioned it, because it never appeared in a meeting.

Self-nomination is not vanity. It corrects a structural bias, since the person doing the work is the only one who knows what it cost.

Peer input needs a specific question

We stopped asking colleagues for general impressions. The question became: what did this person do in the last six months that you could not have done?

Answers arrived with examples, because the phrasing demanded one. General prompts produce general praise, which is close to useless in a calibration room.

Three questions are enough for a peer note. More than that and people start summarising the relationship instead of describing the work itself.

Work that leaves no trace

Some of the most valuable work leaves no trace at all. It includes unblocking others, answering the same question for the fifth time, and keeping a fragile process alive.

None of that generates a ticket. We now ask for a monthly line in a shared note describing what somebody unblocked and for whom.

It is a low-tech answer to a hard measurement problem. The alternative is to reward only what the tracking system happens to record.

Ratings should follow the record

When the written evidence was complete, Ravi’s rating moved up one band. The change came from the record rather than from a negotiation.

That order matters. A rating set first and justified afterwards invites the manager to hunt for support, which is how the bias started.

We now write the evidence before anyone proposes a number. The calibration meeting argues about facts instead of about adjectives people prefer.

The conversation still has to happen

A document is not a review. Ravi read his, disagreed with one line, and we changed it because he had the ticket history to prove his point.

The meeting lasted fifty minutes and most of it covered the next six months. That is what a review is for once the past is settled.

He said the written record made the conversation feel less personal. He meant it as praise, and I have taken it that way since.

What the next cycle changes

We now collect evidence quarterly rather than annually. Waiting twelve months means reconstructing events that nobody remembers with any real accuracy at all.

The template has four fields: what shipped, what was unblocked, what was hard, and what the person wants next. All four take ten minutes to fill.

The surprising cost was manager time rather than hire time. Reading six honest records takes longer than skimming six paragraphs of confident adjectives.

Evidence that needs the author present is not a record of anything. If a reviewer cannot verify a claim without asking, whose memory is the review measuring?