Ratings that keep, cycle after cycle

Ratings get overwritten by the next cycle, the delivered matrix report cannot even compute a median, and a reorg makes last year’s distribution unanswerable. This module keeps every rating as a per-cycle row on the scale it was given under.

RuleStore one rating row per worker per cycle, on the versioned scale it used, under the versioned org it happened in. History survives everything.

Neversummarize ratings with averages alone. A 3.2 average hides the shape; distributions and medians are the calibration audit.

◆ The pagesone model, ten pages, counts that partition.

Page	What it answers
The talent star	The model: per-cycle rating rows, review process events, versioned scales and orgs.
Rating distribution	The shape by department, with the median the delivered matrix report cannot compute.
Review completion	The cycle tracked with statuses that sum to the eligible population, and the chase list.
Nine-box placement	Two ratings joined per worker per cycle, the grid summing to the rated population, movement between cycles.
Ratings vs outcomes	The cohort followed forward: whether ratings predicted turnover and promotion, with regretted attrition surfaced.
Calibration shifts	Entered versus final for the whole cycle: 34 moves, 31 down, the top of the scale compressed, all reconciled.
Manager leniency	Distributions by manager group on one scale; Sales rates 42.7 percent at four-plus and the outcomes join gives the verdict.
Succession coverage	24 key roles: 13 covered now, 6 later, 5 with nobody named, and plans that go stale visibly.
Promotions vs ratings	The promise tested: 30 of 42 promotions on the top bands, the passed-over bench counted beside them.
The scale change	Why a new scale breaks every trend, and the percent-of-scale view that keeps cycles comparable.

◆ The three delivered wallsreported consistently in the field, all structural.

Wall	Consequence
Matrix reports cannot compute a median	Practitioner-verified: the delivered summarizations stop at counts and averages, and averages hide the shape calibration exists to find.
Two ratings, one data source each	Performance and potential live apart; the nine-box join is manual gymnastics, per the single-source wall on delivered vs owned.
Reorgs rewrite the rollup	Last cycle’s distribution by department becomes unanswerable after a restructuring, the same mechanism as turnover.

◆ Where the numbers tiethe arithmetic the battery checks.

The mid-year 2026 cycle runs on 312 eligible of the headcount page’s 351; the distribution and the nine-box both sum to the same 296 completed reviews; completion statuses partition the 312 exactly. Every table crossfoots, and the eligible and rated totals are forced equal across pages by the battery. Sample values are illustrative, never client data.

Use case

Problem

Ratings vanish into the next cycle, medians are uncomputable in delivered summaries, and reorgs erase comparability.

What we build

Per-cycle rating rows on versioned scales and orgs, with review process events beside them.

What you get

Distributions, medians, completion, and nine-box grids that survive scale changes and restructurings, cycle after cycle.

Can you show last cycle’s distribution under last cycle’s org?

We ship the per-cycle rating model that answers it, in 10 days.

Talk to us

Terms on this page

battery: The automated set of checks, including the crossfoot arithmetic, run on every build of this catalog.
cycle: One review period: its window, its scale, its population.
rating scale: The configured levels ratings are given on. Versioned, because it changes.
distribution: How many workers landed at each level. The shape, not the average.
median: The middle rating. Uncomputable in delivered matrix summaries, practitioner-reported.
calibration: The review of the shape: are managers rating alike?
nine-box: Performance against potential, three bands each, nine cells.
partition: Every eligible worker in exactly one status, statuses summing to the total.