Pick the strongest person on your team whose last review you have read. Somewhere in that document is a sentence about how they come across. It does more work than the rest of the page, and it is the one line you cannot audit.
Most review templates ask for a rating on collaboration, communication or executive presence, and attach no definition to any of them. The ratings arrive anyway. They go in the file, feed the calibration meeting, and follow the person into a promotion discussion three years later.
What a manner line actually records
A manner line records how closely an employee's output matched the reviewer's expectation of good work. It reads as an observation about the employee. What it measures is the fit between two people, and only one of them is named in the document.
An evaluative sentence describes something the person did, checkable against a deliverable or a date. Or it describes what they were like while doing it.
The second kind is not a character flaw in the reviewer. It is how the equipment works. Willis and Todorov measured trait judgments of competence and trustworthiness from a hundred-millisecond exposure to a face against judgments made with unlimited time, and they barely moved (Willis and Todorov, First Impressions: Making Up Your Mind After a 100-Ms Exposure to a Face, 2006). Posture, eye contact, vocal tone, the way someone occupies space. It feels like perception. It is a pattern match against a template, and the template was built by one kind of nervous system.
So, "needs to be more engaged with company vision" gets written in good faith about someone who would have memorized the strategy in thirty seconds had it arrived as a document. The observation is real. The causation is wrong.
An outcome line outlives the reviewer. A manner line leaves with them.
I want to be clear where that reading came from. Autistic adults have been making it for decades, long before it reached a management journal: language about how a person presents describes the observer's experience rather than the person's capability. The argument is theirs. The vantage point, a review template and a P&L, is mine.
The two-column count
The proportion is the number. Take last cycle's completed reviews, mark every evaluative sentence as describing an outcome or describing a manner, and divide. A second number comes free. Count how many of the template's rating fields carry a written definition.
Both are assumptions and I will label them as assumptions. Sentence-level marking is a judgment call, two analysts will disagree on the borderline cases, and the fix is marking the same set twice. Substitute your own boundary if you have a better one.
What an explicit criterion buys you
An explicit criterion is cheaper to run than an implicit one, because it can be argued with before it reaches a file. A defined rating can be contested, calibrated and corrected. An undefined rating can only be overruled by someone with more authority than the person who wrote it.
Undisclosed expectations are a process defect an employer owns. In three focus groups, sixteen self-selected autistic US STEM undergraduates named opaque processes and unclear interview expectations as what they ran into while building technical skill on their own initiative (Newbutt et al., Barriers and enablers shaping autistic college students' pathways into advanced manufacturing careers, Australian Journal of Psychology, 2026). Sixteen students in one sector carries no rate and does not reach an org chart on its own. An undefined review criterion is the same defect one stage later.
Here is my own accounting, and it is a controlled case rather than a credential. Culture fit never made sense to me. I could not read the signals that tell most people this one is like us, so I hired people who could do the work and graded them on what they delivered. The implicit channel was unavailable to me, so I built the explicit one. Written promotion criteria. Structured feedback. Transparent decision-making. Retention ran near ninety-five percent across the companies I have run. One operator, and I am not going to tell you the sample generalizes.
You have the channel I did not have. It is running right now, producing evaluations that read to you as observations.
Run it on last cycle
- Pull the completed reviews from the most recent cycle for one function.
- Mark each evaluative sentence O for outcome or M for manner. A sentence naming a deliverable, a number, a decision or a date is O. One naming a style, a presence, an attitude or a degree of engagement is M.
- Report the M share by reviewer, not by employee.
- Count the template's rating fields with no written definition attached.
- Look at where the M share clusters.
An afternoon for one analyst. It is a cut that names managers, so decide beforehand who sees the output.
What you could notice this week
Read one completed review and count the sentences that would mean the same thing if the reviewer had never met the person. Four minutes, no data pull.
Where a review holds none of those, the decision that produced no document is a silence worth measuring.
Where this stops
The count does not tell you which manner lines were wrong. Some are load-bearing. A person who cannot be put in front of a customer is a real constraint, and writing it down is the right call. The ratio tells you how much of the record is pattern match. It does not say which of those sentences was accurate. Clustering by reviewer is the nearest proxy.
Two numbers instead of a rating distribution, and a sentence-level test any manager can apply on Monday. Few organizations have run this once.
What share of last cycle's review language would survive the reviewer leaving the company?
If someone on your leadership team owns the performance review template, this is their number to run.
Read closely: the single phrase that most often ends this sequence has its own post, alongside the labels that cost the most, the same criterion priced at the hiring debrief and the five explicit systems. For a manager inheriting a team mid-cycle, the first decision is the one no training covers.
Measurements you can run on data you already hold: the SHIELD tools.
Peter Allen Mann spent thirty-seven years in leadership, four of them as a Navy surface warfare officer, then at Dell and Tech Data, then founding and running two companies. He is the author of The Perception Revolution (Fast Company Press, 2026) and The Projection Fallacy (April 2027). He writes about late recognition and what it costs at peterallenmann.com.
Sources
- Janine Willis and Alexander Todorov, "First Impressions: Making Up Your Mind After a 100-Ms Exposure to a Face," Psychological Science 17, no. 7 (2006): 592–598. Sage, Psychological Science
- Nick Newbutt et al., "Barriers and enablers shaping autistic college students' pathways into advanced manufacturing careers," Australian Journal of Psychology (2026). Three 90-minute focus groups, 16 autistic US STEM undergraduates, thematic analysis; qualitative, no rate, no effect size. PMID 42891363. PubMed, Barriers and enablers shaping autistic college students' pathways
- Nalini Ambady and Robert Rosenthal, "Thin Slices of Expressive Behavior as Predictors of Interpersonal Consequences: A Meta-Analysis," Psychological Bulletin 111, no. 2 (1992): 256–274. Behind the snap-judgment mechanism; reference taken from the endnote in The Projection Fallacy and not opened this run.
- Lauren A. Rivera, "Hiring as Cultural Matching: The Case of Elite Professional Service Firms," American Sociological Review 77, no. 6 (2012): 999–1022. Carried here for the linked sibling post. No figure from it is used in this piece.
- Peter Allen Mann, The Projection Fallacy (Fast Company Press, April 2027, manuscript), Ch. 4 on presence and power dynamics, Ch. 5 on observation and causation, Ch. 8 on designing feedback. Manuscript in copyedit; check against final text before publishing.