Performance
Companies change their rating scale every few years and silently break their own history.
A 3 out of 5 becomes a 3 out of 4. Nobody migrates the old ratings, because nobody can. Three years of performance data quietly stops being comparable, and the only people who notice are the ones defending a promotion decision.
The decision that protects the history
Publishing a cycle freezes the scale it was rated on.
A published cycle keeps its own scale. Change your rating scale next year and last year's ratings keep their anchors and their meaning, because the scale travelled with the cycle rather than being looked up at read time.
It is a small design decision and it is the difference between a performance history and a pile of numbers whose meaning depends on when they were entered.
2024 cycle
Five anchors
- 1Below
- 2Developing
- 3Meets
- 4Exceeds
- 5Outstanding
“Meets” on a five-point scale.
2026 cycle
Four anchors, adopted last year
- 1Developing
- 2Meets
- 3Exceeds
- 4Outstanding
Also a 3 — but the third anchor of a different scale.
- The reviewer is snapshotted when the cycle opens —
Whoever they reported to at that moment stays their reviewer, so a re-org in March does not hand a stranger a review of work they never saw. HR can reassign a row deliberately; nothing reassigns itself.
- Weights must total a hundred before it opens —
Not a warning. A cycle with dimensions that do not add up cannot be opened, which is the only moment anybody would have noticed.
- A mid-year cycle can be comments only —
Ratings are refused outright on that kind, rather than being a convention managers are asked to observe. A check-in and an annual review are different exercises and the product treats them as such.
- Excluding somebody needs a reason —
A person left out of a cycle is a decision on the record, not an absence from a list.
- Two scales ship, or design your own —
A preset to start from, or your own anchors — because the words on a scale are the part your managers argue about, and they should.
- Dimensions per cycle —
What is being assessed is set on the cycle, so a mid-year check-in and an annual review are genuinely different exercises rather than the same form twice.
- Calibration is its own stage —
Between manager review and publishing, so any adjustment happens before anybody has been told a number. It is one participant at a time — there is no distribution view or forced ranking, and we would rather say that than imply a grid.
- The phases move themselves, except the last one —
Set the dates and the cycle walks itself from goal-setting to self-assessment to manager review to calibration, and later from published into the discussion window and closed. Calibration to published never advances on its own — the one step that tells somebody a number is always a person pressing a button.
- Publishing freezes the dimensions too, not just the scale —
The weights that produced a rating are frozen with it, so a cycle read three years later shows what was actually being measured and how much each part counted.
- Goal setting is optional —
A cycle can start at self-assessment where goals were set elsewhere, so a mid-year cycle is not forced through a stage that does not apply.
The governance nobody ships
Nobody publishes their own calibration.
The person who ran the cycle cannot be the person who releases the ratings. That separation is enforced rather than advised, and it is on by default rather than a setting somebody has to find.
It matters because publishing is the irreversible act in a review cycle — the moment a number becomes a thing an employee has been told. An approval step that the submitter can approve is not an approval step.
- Build the cycle and open itHR operations
- Write reviews, send assessments backManagers
- CalibrateHR operations
- Publish the cycle they submittedHR operationsrefused — an approval the submitter can approve is not one
- PublishA second approver
What it does and does not infer
It adds up your numbers. It does not invent one.
Where somebody has reflected on each of their goals, the goal-achievement rating is pre-filled from those reflections — their own assessments, averaged and snapped to the nearest point on your scale, with a note saying so and an invitation to change it. It is arithmetic on figures a person entered, not a judgement about them.
What nothing here does is infer a rating from evidence the product cannot see. No model reads their work. No benchmark compares them to anybody. A performance rating carries money, promotion and sometimes exit, and an inferred one is the worst possible use of a confident guess.
What HR asks
Mostly about what happens after the cycle closes.
›What happens to old ratings if we change our scale?
Nothing. A published cycle carries its own scale, so old ratings keep their anchors and remain readable against the scale that produced them.
›Can we run a mid-year cycle without goal setting?
Yes. A cycle can open straight into self-assessment.
›Is calibration a separate step or a conversation?
A separate stage before publishing, which is what makes it possible to adjust before anybody has been told a number.
›Does anything suggest a rating?
One thing, and it is arithmetic rather than inference: where the employee has rated their own goals, the goal-achievement dimension is pre-filled from those, clearly labelled and freely overridable. Nothing anywhere infers a rating from work the product cannot see.
›Can the person who ran the cycle publish it?
Not by default. Separation of duties is enforced on the publish approval, which is the one irreversible step in a cycle.
›Can a manager send a self-assessment back?
Yes, with a reason. The participation reverts, the employee's submission clears, and the manager's own draft is preserved — so asking for a redo does not cost them their own work.
›When does the employee see their rating?
When their manager releases it, during the discussion window, per person — so the conversation happens before the number does. Today that release is a step in the workflow rather than a lock the server enforces, so treat it as a process you run rather than a guarantee you lean on.
›Does calibration show us a distribution?
No. It is one participant at a time — there is no grid, no forced ranking and no cross-manager comparison view. If calibrating against a curve is central to how you run, that is the gap to raise.
›Can we fix a typo in a scale after the cycle opens?
No. A cycle is editable only while it is a draft, which protects the history and does mean a mistake found on day two is a cancel-and-recreate.
›Is there peer or upward feedback?
No — manager and self only. No 360, no peer review, no competency library.
›Can employees see their own reviews?
Yes, in the portal — alongside their goals and one-to-ones, which is where the context for a rating actually lives.
Book a demo
Change your scale and look backwards.
The third step is the one most products fail, quietly.
- Publish a cycle on a five-point scale
- Switch the workspace to a four-point scale
- Open last cycle's ratings and check the anchors
- Run a cycle that starts at self-assessment