The distribution, with the median

An average rating of 3.2 tells you almost nothing: a tight middle and a split of extremes both land there. Calibration needs the shape and the median, and the delivered matrix report can compute neither, practitioner-verified.

RuleAudit ratings by shape: the full distribution, the median beside it, by department, per cycle, on the scale that cycle used.

Nevercompare distributions across a scale change as if the numbers align. A rebased scale is a new ruler; the model marks the seam.

◆ The distribution, mid-year 2026 cycle296 completed reviews, median 3.

Rating	Workers	Share
5, exceptional	18	6.08
4, exceeds	92	31.08
3, meets	151	51.01
2, developing	29	9.80
1, below	6	2.03
Completed reviews · median 3	296

The median is 3, over half the population sits exactly there, and 37 percent sits above: that shape is a conversation about central tendency and rater generosity that no average starts. The 296 is the same rated population the nine-box places, forced equal by the battery. Sample values are illustrative, never client data.

◆ The delivered gapcounts and averages, and nothing in between.

Practitioners report the delivered matrix summarizations stop at counts and averages: no median, no quartiles, no shape. For a bounded five-point scale that is the wrong toolkit entirely, because the mean saturates and the tails vanish. The owned model computes order statistics trivially, per department, per manager, per cycle, which turns calibration from an argument about feelings into a reading of shapes.

◆ The queryshape and median, one grouping.

-- distribution and median by department, one cycle
SELECT o.department_name,
       r.rating_value,
       COUNT(*) AS workers,
       MEDIAN(r.rating_value) OVER (PARTITION BY o.department_name) AS dept_median
FROM fct_rating r
JOIN dim_org o ON r.org_key = o.org_key
WHERE r.cycle_code = '2026-MY' AND r.rating_type = 'performance'
GROUP BY 1, 2, r.rating_value
ORDER BY 1, 2 DESC

Use case

Problem

Rating summaries stop at averages, so central tendency, generous raters, and split departments all hide inside 3.2.

What we build

The full distribution with median and quartiles, by department and manager, from per-cycle rating rows.

What you get

Calibration run on shapes: who rates high, who rates flat, and where the middle actually is.

What is your median rating, by department?

We ship the distribution the delivered summaries cannot compute.

Talk to us

Terms on this page

median: The middle rating when all are ordered. Robust where averages saturate.
shape: The whole distribution: tails, middle, skew.
central tendency: Raters clustering everyone at the middle level.
order statistics: Medians, quartiles, percentiles: position-based summaries.
matrix report: The delivered summarization report type. Counts and averages only.
rebased scale: A redesigned scale. A new ruler, marked as a seam.
rater generosity: Systematic differences in how managers use the scale.