What calibration did to the ratings

Managers entered ratings; the calibration room changed 34 of them, 31 downward. The published distribution is the after picture. This page keeps the before, so the room’s work is a fact you can audit instead of a rumor you manage.

RuleKeep the pre-calibration rating beside the final one. Calibration is a decision process; a process whose inputs vanish cannot be audited.

Neverlet the final distribution overwrite the entered one. If 31 ratings moved down, that is a fact the cycle owns, not a draft to discard.

◆ Entered versus finalthe same 296 workers, before and after the room.

Rating	As entered	After calibration
5, exceptional	24	18
4, exceeds	103	92
3, meets	138	151
2, developing	26	29
1, below	5	6
Completed reviews	296	296

The final column is the published distribution, to the worker. Managers entered 127 ratings of four and above; the room let 110 stand. The compression happened at the top, which is where it always happens. Sample values are illustrative, never client data.

◆ The movesthirty-four changes, each one a decision.

Move	Workers
5 down to 4	6
4 down to 3	19
3 down to 2	5
2 down to 1	1
3 up to 4	2
2 up to 3	1
Unchanged	262
Completed reviews	296

Thirty-one down, three up. Every row reconciles the two columns above: start from the entered counts, apply the moves, land on the final distribution exactly. A calibration session that cannot produce this table happened, but did not leave evidence.

◆ The owned answerboth ratings kept, per worker, per cycle.

The rating fact keeps the entered value and the final value on the same row, with the calibration session date. Shift analysis, manager patterns, and the audit trail are all queries; the leniency report reads the same rows.

-- the shift matrix for one cycle
SELECT r.entered_rating, r.final_rating, COUNT(*) AS workers
FROM fct_rating r
WHERE r.cycle_id = '2026-ANNUAL'
  AND r.status = 'Completed'
GROUP BY 1, 2
ORDER BY 1 DESC, 2 DESC

Use case

Problem

Calibration overwrites entered ratings in place, so nobody can say what the room changed, whose team got compressed, or whether the process is consistent.

What we build

A rating fact keeping entered and final values per worker per cycle, with the shift matrix and per-manager movement as standing queries.

What you get

An auditable calibration: the 34 moves named, the top-of-scale compression visible, and next cycle comparable to this one.

What did your last calibration actually change?

We build the rating fact that keeps both values and shows every move.

Talk to us

Terms on this page

calibration: The session where entered ratings are aligned across teams.
entered rating: The manager’s rating before the room.
final rating: The rating of record after calibration; what the worker sees.
compression: Moves concentrated at the top of the scale.
shift matrix: Entered versus final, counted; the session’s fingerprint.
rating of record: The value every downstream report uses.