CalibrForge — Performance Review, Calibration & Pay Equity Command Center

You cannot compare two people’s ratings until you know who rated them.

A performance rating is a measurement, and every measurement is made by somebody. Managers differ enormously in how they use a five-point scale — some rate almost everybody a four, some have never given one. And when each manager rates only their own team, a harsh manager and a weaker team produce exactly the same numbers. They are perfectly confounded, and no averaging, weighting or forcing separates them.

CalibrForge estimates each rater’s leniency from the people who were assessed by more than one rater, shrinks every estimate toward zero in proportion to how little evidence stands behind it, publishes each one with its confidence interval, and refuses outright when the overlap is not there. It is the only honest way to do this, and most tools do not.

What it finds in the worked organisation

• 1.46 rating points between the most generous rater and the most severe one — wider than a whole box on the scale. The typical difference between two raters is 0.457 of a point.
• 18 of the 27 top-band places are filled by choosing among people the rating does not separate — they all share the same rating at the cut. There is no top twenty per cent on a five-point scale; the achievable bands are 6.6% or 38.2% and nothing in between.
• $106,487 of merit allocated by whatever happens in the room rather than by the rating. Not waste — somebody gets it — but not allocated by the thing you believe is allocating it.
• A pay gap of -2.84% with an interval from -6.03% to 0.46% that contains zero, beside an uncontrolled figure of -24.63%. Both are real. They answer different questions, and the screen explains which.
• 26.5% of ratings have a written case behind them — and the share depends far more on who wrote the review than on who it was about.

Twenty-one screens

Review Command Center · Who Rated Them · The Review Ledger · Every Rater, Measured · A Four Is Not Twice A Two · Whose Outcome Changes · Promotion Decisions · The Merit Pool · Pay Gaps, With Intervals · What Backs Each Rating · The Last Quarter Problem · Goals Against Ratings · Tenure, Level & The New · The Curve You Forced · Who Left Afterwards · Spans Of Control · Cycle Over Cycle · Review Calculators · Letters & Guidance · Cycle Reports · Settings & Backup

What it refuses to do

• Compare two raters who share no ratee, directly or through a chain.
• Publish a rater effect built from one or two shared reviews.
• Report a pay gap for a group of fewer than twelve people.
• Treat a five-point scale as a ruler — and it names every place it comes close.
• Price what it cannot price. The unpriced list is the longest of the four on the reports screen, and it should be.

In the folder

• One HTML file. Double-click it. Nothing installs, no account, no network request of any kind — open it with the wi-fi off and it behaves identically.
• A sixteen-sheet Excel workbook. Every figure a live formula, including a full least-squares pay-equity fit with LINEST. The last sheet names every place the workbook and the application diverge and prints the measured difference beside each one.
• Nine calculators that open on your own figures, six letter templates (including a calibration session note and a letter to somebody who was inside the tie), a seven-step guided tour and a printable cycle report.
• Two guides and a licence that says what you may do with it.

Your data never leaves your browser. There is no account and no server.