What the ratings predicted

Every cycle produces a distribution, and there it stops. The delivered reports keep ratings per cycle, with no path forward to what happened next. So the question a leadership team actually has, did these ratings predict anything, who left, who advanced, has no report, because the answer is a join between this cycle’s ratings and the following year’s outcomes.

The distribution shows the shape. This page carries that same cohort forward, into the turnover and promotion the turnover and movement models already track.

The ratings did predict. Attrition climbs as the rating falls, from 6 percent at the top to half the ones, and promotion concentrates at the top, 44 percent of the fives.

But the join shows what the distribution hides: seven of the thirty-one who left were rated four or five, the regretted attrition a shape never surfaces and a manager most needs to see. The cohort ties to the same 296 the distribution counts. Sample values are illustrative, never client data.

Join the rating cohort to the movement and turnover events in the twelve months after the cycle, per worker. Outcome rates by band fall out; regretted attrition is the top-rated leavers; promotion validity is whether the fives actually advanced. The longitudinal question the delivered per-cycle report cannot reach becomes one grouping in a model that keeps every cycle.

-- outcomes by rating band, the cohort followed 12 months forward
SELECT r.rating_value,
       COUNT(*)                                          AS cohort,
       SUM(CASE WHEN o.event = 'termination' THEN 1 ELSE 0 END) AS left_12mo,
       SUM(CASE WHEN o.event = 'promotion'   THEN 1 ELSE 0 END) AS promoted_12mo
FROM fct_rating r
LEFT JOIN fct_worker_event o
  ON o.worker_key = r.worker_key
  AND o.event_date BETWEEN r.cycle_end AND r.cycle_end + INTERVAL '12' MONTH
WHERE r.cycle_code = '2026-MY'
GROUP BY 1 ORDER BY 1 DESC

Run it each cycle and the reviews stop being filed and forgotten. They become a prediction with a track record, and calibration gains the one thing it never had: evidence that the ratings meant something.