Focus room pilot, week 1

Rooms with a display averaged — occupied against — without, but the gap comes from one room. 1145-D was occupied — of the workday, while the other three ranged from — to —. Treat display vs no display as unresolved until more weeks are in. Short visits are undercounted: neither 5-minute visit in the Oct 5 log was detected, so these figures are a floor for brief use.

Weekdays 7am to 7pm, Fri Oct 2 noon through Thu Oct 8 noon. Occupancy is estimated from Govee temperature and humidity; no sensor counts people directly.

Config A, no display (1114, 1144) — of business hours occupied, range — to —
Config B, display (1115-D, 1145-D) — of business hours occupied, range — to —
Typical session length, A vs B — vs — average minutes; — and — confirmed sessions
Oct 5 logged visits detected — of — Both were 5-minute visits. The sensors show no change at those minutes, so visits this short sit below the detection floor.

1145-D is the busiest room by a wide margin

Share of weekday business hours occupied. Bar is the base estimate; the whisker runs from strict to lenient detection.

Use clusters late morning and early afternoon

Share of each hour occupied, averaged over the weekdays observed. Darker is busier.

Every detected session

Each bar is one stay in a room, 7am to 7pm. Dashed outlines are temperature-only rises without humidity confirmation; they count only in the high estimate. No sessions were detected over the empty Oct 3–4 weekend.

The two logged 5-minute visits fall below the detection floor

Oct 5, 12:30pm to 4pm, taking the log times as exact. Grey bands are the logged visits; colored bands are the next session the rules found in the same room. At the logged minutes, temperature and humidity keep the trend already under way, with no step that separates them from an empty room.

Room Logged visit Detected What the sensors show Next detected session Min after log end

Tuning the rules to catch those visits fails on an empty building

The loosest rule that fires inside both logged windows was run on every room. It flags almost as many visits on the empty Oct 3–4 weekend as on weekdays, so it cannot tell occupied from empty. The current rules flag nothing on that weekend.

Room Current rules, weekend sessions Tuned rule, empty weekend flags/day Tuned rule, weekday flags/day

Detection rules

The same rules run on every room and every upload. Room thresholds are set from each room's own empty weekend, never tuned by hand.

Step Rule Setting

Daily breakdown

Fri Oct 2 covers noon to 7pm and Thu Oct 8 covers 7am to noon, so their shares use fewer observed minutes.

Day Room Config Occupied min Observed min Occupied Sessions

Session list

Rises are measured from session start to peak; humidity is net of the building-wide trend.

Day Room Start End Min Confidence Temp rise °F Humidity rise g/m³ How it ended Note

How occupancy was estimated, and what to watch

Signal

  • Temperature turned out to be the stronger signal: a person plus a docked laptop and monitors warms a 60 SF room by 1 to 6°F per stay. Humidity confirms that a person, not equipment or HVAC, caused the rise.
  • A session starts when temperature climbs well above each room's weekend noise and humidity rises faster than the other rooms. It ends at the peak before a sustained drop. A sharp drop in both usually marks the door opening.
  • Building-wide swings are removed by comparing each room to the rooms that are empty at the same minute.

Caveats

  • One work week and two rooms per configuration. The A vs B gap depends entirely on 1145-D, so it says more about that room than about displays.
  • 1115-D runs cooler during the day and its rises are smaller, which suggests stronger supply air. It may be undercounted relative to the others; check its diffuser and sensor position.
  • Session edges are accurate to about ±15 minutes. Both 5-minute logged visits were missed, so brief stays are undercounted and utilization is a floor.
  • 1145-D has a 90-minute logging gap on Fri Oct 2 from 4:21pm; those minutes are excluded, not counted as empty.

Next

  • Log six to eight more visits per configuration at 10, 20 and 45 minutes, seated at the desk with the laptop docked, to measure the shortest stay the rules catch and the error on start and end times.
  • Thresholds come from each room's empty weekend, so they recalibrate automatically as more weekends arrive.
  • Keep syncing weekly; rerun this analysis on the cumulative data.

Snapshot published Oct 8, 2026 from Govee exports synced that morning. Figures are estimates from temperature and humidity.