How to Set Scorecard Band Boundaries Without Gaps
How to set scorecard band boundaries: write ranges so each score lands in one band, state the inclusive rule, pick a gap policy and recheck after every edit.
To set scorecard band boundaries, write each range so that every possible score falls into exactly one band, state whether the ends are inclusive, and choose in advance what happens when a score matches nothing. Boundaries are arithmetic, not judgement, and an accident here becomes a wrong verdict.
The score is only a number. The band is what the respondent actually receives, repeats to a colleague and remembers. So a boundary mistake is not a rounding quibble: it tells someone the wrong thing about themselves. This post covers the mechanics of drawing the lines. For what to call each band afterwards, see how to name scorecard result bands.
How to set scorecard band boundaries: state the contract
A boundary contract is a plain statement of how matching works. In the engines this method draws on, matching is normally inclusive at both ends. A range from nought to thirty includes both nought and thirty. That sounds harmless until you author the next band as thirty to fifty.
Now a score of exactly thirty belongs to two bands at once. Which one wins is decided by sort order, in other words by an accident. Nobody chose it and nothing warns you about it.
The fix is to author bands so that no two share a bound. Start the next band one hundredth of a point above the end of the previous one. Instead of nought to thirty followed by thirty to fifty, write nought to thirty, then a band that begins a hundredth above thirty and runs to fifty, then another that begins a hundredth above fifty. Every score now lands in exactly one band.
| Authoring style | What happens at the shared number | Verdict |
|---|---|---|
| Bands that touch (the end of one equals the start of the next) | The score matches two bands, and sort order picks one | Wrong |
| Bands that start a hundredth above the previous end | The score matches exactly one band | Right |
Write the contract down in one sentence somewhere you will find it again: "Ends are inclusive, and each band starts a hundredth of a point above the one before." Anyone who edits the ladder later then knows what the rule is.
Why a near miss still does damage
The best teaching story is about a near miss. Picture a per-pillar threshold row that begins a hundredth of a point before the true band does. A respondent whose score sits exactly on the real band's first value is matched by the earlier row instead, because it is found first. Nothing errors. Nothing looks wrong. The respondent is told the wrong tier, and may go on being told it for months.
That is the nature of this class of fault. It does not crash, and it affects only the people who score exactly on or near an edge, so it is easy to miss in testing. Every boundary on the page deserves a deliberate look, including the per-pillar rows, not only the overall ladder.
Know your gap policy
A gap is the opposite fault: a score that matches no band. There is no safe default, and the behaviour is usually chosen per screen, not once for the whole product. Three behaviours exist.
| Policy | What it does | When to use it |
|---|---|---|
| None | Shows nothing for an unmatched score | The honest choice, and right for the headline result a respondent sees first |
| Clamp | Snaps the score to the nearest end of the ladder | Acceptable where an approximate band is better than a blank |
| Last | Hands the person the top band | Avoid. It is the weakest policy and can be dangerous |
The danger of the last policy is easy to see. If one screen shows nothing and another uses the last policy, a ladder with a gap can show no archetype at the top of the page and the best archetype in the action plan, on the same page for the same person. They would be right to doubt everything else on it.
You cannot rely on a default, because the right answer differs by screen. So know which policy each of your screens uses, and then make gaps impossible in the first place.
Validate that the ladder tiles
A well-formed ladder tiles the whole scale from nought to one hundred with no gap and no overlap. Treat that as a check you run, not a hope you hold. The routine is short:
- List every band from lowest to highest, with its start and end.
- Confirm that the first band starts at nought and the last ends at one hundred.
- Confirm that each band starts a hundredth of a point above the end of the one before.
- Repeat the check for any per-pillar ladders, which are separate and just as capable of failing.
- Repeat the whole check after every edit.
Where the editing screen refuses broken ladders, use that help, but do not rely on it alone. The validation usually lives in the editor, not in the database. Anything created another way, such as bands loaded in bulk, bypasses it entirely. That is exactly how overlapping ranges survive in a live instrument: they predate the screen that would have refused them.
Boundaries come from the distribution
Where you place the lines matters as well as how you write them. A first guess made on paper is only a plausible ladder. The way to set cut points that separate people is to look at real scores and place the lines so that every tier is populated. If most respondents fall into one band, the cut points are wrong, not the respondents, and the fix is arithmetic. The loop is set out in how to calibrate a scorecard.
Before launch, a rough ladder is fine. After launch, treat changes carefully, for three reasons.
- Recalibrating usually does not backfill. People who already took the assessment keep the band they were given.
- A per-band count in your editor can therefore legitimately disagree with a chart that recomputes live.
- Scores are written once, so every change to a boundary is a fork in the instrument. Batch changes, date them in your own notes and never read a trend across an edit.
Boundaries and the weights beneath them
Band lines sit on top of the scale, so anything that moves the scale moves the people. A richer option added to a live question, or a changed pillar weight, shifts scores and can push respondents across a line they would not have crossed before. That is another reason to settle the model first. The rules for that are in how to weight scorecard questions, and it is best to finish them before you fix any boundary.
Reading a ladder for faults
When you review a ladder, look for a short list of faults. Each one has a recognisable symptom, so you can often spot it by reading the table of ranges slowly from top to bottom.
| Fault | What you will see | What it does to a respondent |
|---|---|---|
| Overlap | Two ranges that share a number | The band depends on sort order |
| Gap | A stretch of scores no band covers | A blank, or the wrong band, depending on the screen |
| Wrong start | The lowest band does not begin at nought | The lowest scorers match nothing |
| Wrong end | The highest band stops short of one hundred | The best scorers match nothing |
| Duplicate label | Two bands share a name | Reports and tags can no longer tell them apart |
Before you delete or rename a band, also look at what depends on it. Follow-up rules, tags and past assessments may all be attached. Removing the band without checking them leaves people with a rule that waits for something that no longer exists.
Boundaries decide, labels display
The numbers that draw the lines belong to the part of the instrument that decides. They are never translated, and nothing that depends on a translated word should be allowed to influence which band a person lands in. A respondent answering in one language and a respondent answering in another, with identical answers, must land on the same side of every line. Treat the boundary values as fixed facts of the model, and keep names and descriptions on the display side. The next post explains how to handle the names themselves.
A sensible order of work
- Settle the option ladder and the weights, so the scale stops moving.
- Draft a rough set of ranges and write them to the contract.
- Check that the ladder tiles from nought to one hundred.
- Collect real completions and look at where scores fall.
- Move the lines so every tier is populated, and check the tiling again.
Keep the checks in this order. Fixing lines before the scale is stable wastes effort, because the next change to a weight will move everyone again.
A boundary checklist
- The first band begins at nought and the last ends at one hundred.
- No two bands share a bound, and each starts a hundredth above the previous end.
- Per-pillar ladders have been checked as carefully as the overall ladder.
- I know the gap policy of every screen that shows a band.
- The ladder has been rechecked after the latest edit.
- Every band is populated when I look at real scores.
Frequently asked questions
Why not simply use whole numbers for the boundaries?
You can, as long as no two bands claim the same whole number and the ladder still covers every possible score. Scores are not always whole, so a ladder built only from whole numbers can leave tiny gaps between bands.
Which gap policy should I choose?
For what a respondent sees first, showing nothing is the honest choice. Whatever you choose, make gaps impossible by checking that the ladder tiles.
Will changing a boundary change old results?
Usually not. Earlier respondents keep the band they were given, so past counts and live charts may disagree. That is expected, not a fault.
Do per-pillar bands need the same care?
Yes. They are separate ladders, and a faulty row can quietly assign the wrong tier to a pillar while the overall ladder looks perfect.
The full method, with the checklists, is in The Scorecard Playbook, a free 39-page playbook.