Canonical URL: https://density.io/resources/occupancy-sensor-pilot-scorecard

# How to score an occupancy sensor pilot: a worked example

An occupancy pilot needs separate scores for the questions it is meant to answer. Distinguishing an empty room from an occupied one is a different test from identifying group size. Density's [product-specific accuracy methodology](https://density.io/resources/the-truth-behind-99-accurate-occupancy-sensors) explains why those tasks need different measures.

The worked example below shows how to keep those calculations separate. **Every reading is invented for instruction. This is not a test of Density or another vendor, a performance benchmark, or a proposed acceptance threshold.**

## Fix the output before scoring it

[Density Waffle](https://density.io/waffle) reports zero, one, two or three-plus people. For this example, map the reference headcount into those same categories. A reference count of four becomes three-plus; it does not become an exact sensor reading of four.

Suppose a buyer has two questions: can we distinguish empty rooms from occupied rooms, and can we distinguish individual use from group use? Keep the original reference count alongside the mapped category so either calculation can be checked.

Each row below represents one equal-duration observation interval, with the reference and reported output aligned to the same room and time. In a real pilot, agree how the reference is collected, how timestamps align and how transitions are treated before starting.

## Work through ten hypothetical intervals

| Interval | Reference people | Reference category | Hypothetical reported category |
| --- | --- | --- | --- |
| 1 | 0 | 0 | 0 |
| 2 | 0 | 0 | 0 |
| 3 | 0 | 0 | 1 |
| 4 | 1 | 1 | 1 |
| 5 | 1 | 1 | 0 |
| 6 | 2 | 2 | 2 |
| 7 | 2 | 2 | 3+ |
| 8 | 4 | 3+ | 3+ |
| 9 | 6 | 3+ | 3+ |
| 10 | 1 | 1 | Unavailable |

Nine intervals have both a reference and a reported category. One has a reference but no usable reported output. Keep that interval in the coverage record; do not convert it into an empty room.

## Calculate coverage first

**Hypothetical paired coverage: 9 ÷ 10 = 90%.** Here, coverage means the share of scheduled intervals with both a usable reference and a usable reported output. It is a definition for this scorecard, not a vendor uptime measure.

Report why an interval was unavailable. A missing reference and a missing sensor output need different follow-up work. If intervals have different durations, use their durations for time-based calculations rather than treating each row as an equal amount of time.

## Score empty versus occupied

Treat categories one, two and three-plus as occupied. Among the nine paired intervals:

- **Correct empty:** intervals 1 and 2.
- **False occupied:** interval 3, where the reference is empty but the reported category is one.
- **Correct occupied:** intervals 4, 6, 7, 8 and 9.
- **False empty:** interval 5, where the reference is one but the reported category is zero.

**Hypothetical empty-versus-occupied agreement is 7 ÷ 9, or about 77.8%.** Two errors remain visible: one false occupied interval among three reference-empty intervals, and one false empty interval among six reference-occupied intervals. The hypothetical conditional error rates are **1 ÷ 3** and **1 ÷ 6**, respectively.

Keep the denominators beside the results. Reporting only the overall agreement would conceal which kind of mistake occurred. In this example, interval 7 is correct for occupied-versus-empty even though its group-size category is wrong.

## Score the reported categories separately

The reported category matches the reference category in intervals 1, 2, 4, 6, 8 and 9. **Hypothetical category agreement is 6 ÷ 9, or about 66.7%.**

That differs from the empty-versus-occupied result because interval 7 reports three-plus when the reference category is two. The room is correctly reported occupied, but the distinction between two and three-plus is incorrect.

Intervals 8 and 9 match the three-plus category. They do not establish exact headcounts of four and six. Choose a measurement output that supports the question; a category score cannot validate a capability the output does not provide.

## Take the scorecard into a Density pilot

The example is deliberately small enough to audit by hand. Ten invented intervals cannot establish expected field performance. A real evaluation needs representative rooms, empty and occupied periods, quiet individual use, groups and transitions, with enough observations to assess each agreed condition.

Bring Density the room list, the decision, the output definition and a proposed scorecard. Agree the observation schedule and acceptance rules before collecting results. Review coverage, empty-versus-occupied errors and category agreement together, then inspect the individual disagreements before deciding whether to expand.

Use the [occupancy accuracy guide](https://density.io/guides/how-accurate-are-occupancy-sensors) for the broader test design and the [RFP acceptance schedule](https://density.io/resources/improve-your-occupancy-sensor-rfp-with-our-best-practice-guide) for responsibilities and handover. This scorecard supplies the calculation between those steps.

[Book a Density demo](https://demo.density.io), or email [sales@density.io](mailto:sales@density.io) with the room decision you want the pilot to resolve.