How to interpret 360 feedback without flattening it

Here's how most 360 feedback gets read. You open the report, your eye goes to the number at the top — overall 3.8 out of 5 — then to the two auto-generated lines beneath it: strength: communication; development area: delegation. You nod, feel vaguely fine, and close the tab. Ten minutes later you couldn't tell anyone a single specific thing you learned, because there wasn't one. The report did its job of summarizing and, in doing so, quietly destroyed the only parts that were worth reading. A 360 is valuable precisely for the things a summary throws away — the disagreement between people, the gap between how you see yourself and how they do, the one verbatim comment that stings because it's true. Flatten those into an average and a tidy phrase, and you've paid for a mirror and then frosted the glass.

Why the average is the enemy

An average is a compression algorithm, and like all compression it discards whatever doesn't fit the summary. The trouble is that in a 360, the discarded part is the point.

Take that 3.8. It could mean everyone thinks you're solidly good — a genuinely reassuring, low-information result. Or it could mean half the respondents rated you a 5 and half rated you a 2, which is not "pretty good" at all: it's a person who lands completely differently on different people, which is one of the most important things a leader can learn about themselves. The mean reports those two opposite realities as the same number. So the single most useful signal in the whole dataset — are people seeing the same person? — is the first thing the average deletes.

This is why a 360 read only at the summary level tends to feel simultaneously fine and useless. It's engineered to feel fine. Useful is one layer down, in the stuff the summary smoothed over.

What actually carries the signal

If the averages are the least useful layer, here's what the useful layers are — the things worth protecting from any tool or habit that wants to flatten them.

• The spread, not the mean. High disagreement between raters isn't noise to be averaged out; it's the flashing light. It usually means your behavior is inconsistent across people or situations — you're one person with your peers and another under pressure — and that inconsistency is exactly the thing worth understanding.
• The self-versus-others gap. The comparison between how you rated yourself and how others rated you is the entire engine of a 360. Tasha Eurich's research on self-awareness is blunt about how rare it is to see yourself accurately — so the places where your self-score and their scores diverge are your blind spots, handed to you directly. A summary that blends your answers in with everyone else's erases the one comparison that matters most.
• The verbatim comments. The open-text answers are where the specifics live — the actual moment, the actual behavior. "Cuts people off in meetings when he's under time pressure" is worth more than any rating, because it's observable and you can act on it. Scores tell you that something's off; comments tell you what.
• Who said it. A peer, a direct report, and your own boss are watching different slices of you. Averaging their answers together produces a number that describes nobody. The interesting questions are between the groups — why do your reports rate your clarity far lower than your peers do?

The flattened read vs the real one

Why an AI summary makes it worse, not better

There's a strong temptation right now to solve the "reports are hard to read" problem by having an AI write you a tidy paragraph: "Your feedback suggests you're a strong communicator who could delegate more effectively." It feels like help. It's the opposite. An AI narrative summary is the flattening problem with a confident voice bolted on — it averages the averages, launders the specifics into smooth generic phrasing, and resolves the disagreements that were the entire signal into a single pleasant sentence. The variance was the data; a summary's whole job is to remove variance. You end up with something that reads well and tells you nothing, and worse, tells you nothing authoritatively, so you trust it and stop digging.

This is why Mirorly does statistical synthesis — distributions, gaps, per-group breakdowns — and deliberately no AI narrative synthesis. The math can surface where the disagreement is without dissolving it; a generated paragraph can only dissolve it. Reading your own verbatims and gaps is slightly more work than skimming a summary, and that work is the point — it's the same reason self-assessment has to come first: the friction is where the learning happens.

How to read a 360 without flattening it

1. 1

   Read the verbatim comments first, before any score

   Start where the specifics are. The open-text answers anchor everything else in real behavior and stop you from resolving the whole report into a number. If you only have time for one layer, make it this one.
2. 2

   Look at the spread before the mean

   For each question, check how much the raters disagreed. Wide disagreement is your most important finding — flag those items first, because that's where you're landing differently on different people.
3. 3

   Separate the answers by who gave them

   Never read one blended average. Look at reports, peers, and your manager as distinct views, and pay attention to where they diverge — the between-group gaps are usually more revealing than any single group's score.
4. 4

   Hunt the self-versus-others gap

   Put your own answers next to theirs, question by question. The biggest divergences are your blind spots, and they're the highest-return place to spend your attention. Related patterns show up in your management blind spots.
5. 5

   Resist compressing it into one action

   Don't reduce a rich report to "be a better delegator." Pick one specific, observable behavior out of the specifics, change that, and re-measure next round. The detail you'd have flattened is exactly the detail you act on.

Where Mirorly fits

Mirorly is built to keep the signal intact rather than summarize it away. You answer behavioral questions about yourself, send the same set to the people you work with, and the results show you the distribution for each question, your self-view next to theirs, and a breakdown by relationship — managers, peers, and reports in their own columns — so you can see exactly where the views diverge instead of reading one blended number. There's no AI paragraph telling you what to think, on purpose: the core leadership behaviours template gives you the structured questions, and the interpretation stays yours, where it belongs. If you're choosing what to measure, start with questions that surface useful answers, and run it on a cadence so the gaps become a trend you can track.

Common questions

The one-line summary

A 360's value lives in the parts a summary throws away — the disagreement between raters, the gap between your self-view and theirs, and the specific verbatim comments — so the average score is the least useful number in the report and an AI-written summary is worse still, because flattening variance is the one thing you must not do; read the specifics first, treat disagreement as signal, separate answers by who gave them, and act on one concrete behavior rather than a smoothed-over phrase.