Instrumented People, Places and Things: an office day, 17 August 2012

Boxes

A Bluetooth logger, as the team's viewer drew it, with north up and east to the right. Its edge and color follow the count of the scan it is on: one for each thing it heard. A scan that heard nothing still counts one, because the log writes a line saying so. A small pale box, breathing, means no log at that time.

Disks

Added in 2026. For the chosen device, how often each logger heard it over the last 5 minutes. A disk says nothing about where the device is.

Small boxes

A Kinect's tracked center of mass, one for each reading in the last 10 seconds, colored by tracker id. An id is the tracker's label for as long as it holds on to something, so it need not be one person. Orange lines: its field of view.

Colored squares

A Grid-EYE's 8 × 8 heat readings, on the floor under it. The viewer's colors, red hotter than yellow, than green, than blue. The sensors read far apart, so a color can't be compared between them. Green lines: its field of view.

Bars

A spider's readings. Four recorded sound in seven bands; two recorded light, humidity, air pressure and temperature, and one of those motion too. Red boxes, breathing: no reading at that time.

The track

Added in 2026, under the replay. For the chosen device, a row for each logger, in 5-minute steps, deeper when more of that logger's scans heard it. Blank: that logger was not running. Neighbors are beside each other: 2, 16 and 14 along the brick wall, then 13, 1 and 5 through the middle, 11 to the east and 12 to the north.

Timeline

The team's rows: G for a Grid-EYE, S a spider, K a Kinect, B a Bluetooth logger. Gray where a sensor has no data. The red line is the replay's time; click to set it. The labels are added. Sound's seven rows show the highest of each band over each pixel's stretch.

Devices

Added in 2026. A row for each device, a column for each minute, deeper when more loggers heard it. Click a row to choose its device.

The team's viewer places each sensor in a 3D model of the office, the same model as the open house, file for file. It was exported two days after this day, and the sensors were moved into place by hand, 5 inches at a press. So the places are the team's, not a survey, and they could be wrong. They were tested three ways:

Kinects

Their people land on open floor: under 2% of readings fall in a wall or past the room's edge, against 75 to 97% at the middle of each Kinect's other headings. Someone may have placed them by eye with the data showing, so this says the model fits more than that the placing was blind. It can't see a turn of 10° or a shift of a few inches. The team's model has a table at the east end, and its shadow is baked into the floor. The viewer does not draw it, and 15% of its readings there fall inside its outline, so whether a table stood there that day isn't known.

Loggers

Loggers that stand closer hear more in common (r = −0.51, as or better in 631 of the 40,320 ways of assigning eight loggers to eight places). The four computers' hearing fits the places better than all but 45 of those ways. Neither can tell the plan from its mirror image.

Grid-EYEs

Where the Kinects put people, two of the Grid-EYEs see warmth only with their picture a quarter turn from the way the viewer drew it, the same turn the open house's sensors needed. The page draws all four turned. The other two have no Kinect to test against.

At once

While all eight ran, 10:51 to 12:39, the four computers heard most were heard by three or more loggers in most 10-second spans, and by seven or eight in about half of their minutes.

With one logger

A computer's runs at one logger are short: a median of two or three scans for the three heard most. It comes and goes for that logger with no move at all, so one logger's silence means little until they all fall silent.

Limits

Only a device set to answer can be heard, so many never show, and a device heard is not a person. A computer may be left on a desk all day. Whether the team set its own computers to answer that day isn't recorded.

TJ: “A sensor has a range and resolution.” (Computational Design, p. 4.)

No list of the study's devices has been found, and the loggers heard anything set to answer, including people passing. So the page sorts devices by how they were heard, and not by who held them. Stable labels only: device 1 was heard most. No address or name is published, and a device's kind is read in 2026 from what its name says.

Stayed, 6

Heard in 30 minutes or more: five computers, one phone. They hold 99% of the sightings.

Returned, 6

Fewer minutes, on two visits or more, over 10 minutes apart.

Passed, 63

One visit, nearly always within seconds. They are heard most by loggers 2 and 16, the two nearest the brick wall. No file says what lay beyond it.

TJ: “Placing sensors in one room does not, by itself, make their measurements compatible.” (Computational Design, p. 9.) Each device stamped its own time. Phones heard for a moment measure the difference: as stamped, two loggers hear the same phone about a minute apart, again and again. Of the 20 passing devices that loggers 2 and 16 both heard, logger 2 hears every one first as stamped, and 12 of the 20 on one clock, so an order of loggers read from the stamps would be the clocks and not a walk. The Clocks view shows it. Loggers 14 and 1 read an hour and two hours fast, as their own files' dates show too.

The team's viewer played each logger by its own stamps, and so does the Timeline view when One clock is off. Everywhere else the loggers play on logger 2's clock, which is not itself known to be right. Played as stamped, device 6 seems to move from logger 13 to logger 14, an hour apart. On one clock the two hear it together.

The other sensors' computers are left as they stamped. Where it can be tested, two Grid-EYEs and two Kinects agree to within a few seconds. The sound spiders' clocks are not settled, and nothing here ties the loggers to the rest.

Kept from the team's viewer: the white ground, the model and the places, the colors, each sensor's drawing, the timeline's rows and the red line. Like the viewer, the page does not draw the table's model, and plays at 10× unless told otherwise.

Changed, each marked here: the loggers play on one clock; the Grid-EYEs are turned a quarter; a Kinect's height is the right way up, where the viewer drew it upside down about the camera, which changes nothing in plan; readings are thinned to one a second, the highest for sound and the median for heat; light and air to 5 seconds, and air pressure is only a difference.

Added: disks, tracks, the Devices and Clocks views, labels on sensors, and two sets of readings the viewer did not load.

As a way to see a day the approach held up: which computers were in the office, when one left or moved, and that phones go by. It needs nothing carried for the study, only devices set to answer, and those included people with no part in it, which is also its concern. These logs hold other people's device names, kept out of this page.

It cannot say where a device was or who carried it, and sound, heat and Kinect points name no one. TJ: “An observation is not an activity model.” (Modeling Activity, p. 3.) The patterns here are observations, from which a model of a working day would still have to be made and tested. TJ, of this day and the office beacons of 2014: “They use different tech and different scopes.”