Performance Evaluations
Review season arrives, and Meera sits down to write Arjun's review. She clearly remembers his excellent June launch and very little from January to April. She is about to write something that will shape his pay and his next year, based on about six weeks of memory.
The rule
No surprises: the review only writes down what you've already said.
A fair evaluation rests on four things.
- Clear expectations, set in advance. People can only meet a bar they can see. Most companies publish expectations for each level, often called a career ladder. At Haat, a mid-level engineer is expected to own features end to end and to give reviews that others learn from.
- Evidence from the whole period. Keep short notes after one-on-ones and after notable events. Without them, memory plays tricks:
- Recency bias: the last few weeks crowd out everything before them.
- Halo effect: one strength colours every other judgement.
- Similarity bias: people who work or think like you seem better.
- Impact, not activity. "Led the refund flow; refund-related support tickets fell by a third" is impact. "Worked late most nights" and "wrote a lot of code" are activity.
- No surprises. Any concern in the written review should already have come up in one-on-ones, months earlier, with time to improve. The review sums up conversations that have already happened; it is not the place to start one.
A written review usually covers what went well, with evidence, what to grow, a rating if the company uses them, and goals for the next period.
The shortcut
For every sentence you write, point to the evidence and the date you last said it. If you can't point to evidence, the sentence is an opinion. If you have never said it to them, it is a surprise.
Pull the expectations
Mid-level: owns features end to end, and gives reviews that teach. These are the headings Meera judges against.
The trap is writing the review from memory in one evening. Memory hands you recency, halo and similarity bias at no extra charge.
Drills
Answer fast — trust the rule.
Meera remembers Arjun's June launch vividly but little from January to April. Which bias is at work?
Kabir first hears that his code reviews are too slow when he reads his annual review. What went wrong?
Priya is superb at system design, so Meera rates her highly on communication too, without checking any evidence. Which bias is this?
Which is the strongest evidence of impact?
Remember
- No surprises: the review only writes down what you've already said.
- For every sentence, point to the evidence and the date you last said it.
- Keep notes across the whole period to beat recency, halo and similarity bias.