Measuring Engagement, 2014: model of the Monadnock Building

The model: drag to turn, scroll to zoom. It is a measuring tool and not a portrait, so it has no textures, furniture or people. On the walk page, it fades in as the camera closes on the building.

A model of the room gives the video’s reconstruction something fixed to be measured against. It had to be built first, because the room’s mirrors, lamps, doors and floor are the only things in the video whose size can be measured, and in this room the software could not follow the walker through a door.

What it is built from

The plan and the photographs were measured, not copied. None of them is reproduced on this site, and they remain their owners’.

What measuring it found

• The wall curves. The Monadnock’s brick wall rises straight up to a ledge at about 13.75 ft, then leans back, 5 in by 15.8 ft and 13 in by 22.4 ft, read from one corner’s silhouette to about 3 in. The taper above that comes from the published wall thickness, and the flare at the roof is an estimate.
• The windows are bays. On the west wall each is an 18 in recess from floor to ceiling, with its frame at the back. The counter sits in the window, not in front of it, and the stools go under it.
• The ceiling has two levels. Over most of the room it is 112 in. By the Federal windows it steps up to about 120 in, level with the top of the storefront.
• Two mirrors, not one. The video shows two framed mirrors behind the bar in 2014, as the 2015 photographs do. A single long one appears in a photograph from 2024, with a later bar.
• The entrance’s doors. Two sets close 24.1 and 17.8 ft in from Jackson, each two pairs of leaves either side of a post, read from the plan’s door swings. The plan draws its doors open, so the tips of the leaves are not the doors.
• The floor. A half-offset grid of about 20.5 by 10.25 in tiles, long side east to west.

How well it fits

• Two of the photographs. Where their cameras stood, the bar and the rows of lamps were worked out together from 26 points read by hand. The result misses those points by a typical 2.9 pixels in a 900-pixel image.
• The height of the doors. Seen nearly face on in the video, the bar over the doors on Federal Street stands 85.5 in above the floor, using the plan’s 60 in door width as the scale. The model has it there. The coffee bar’s street door, at No. 55 Jackson, stands about 82 in when seen face on, against the model’s 84.
• Walking through it. At walking height, the walker’s route passes through the street door at No. 55, the inner door, the side door and the two sets of doors at the building’s main entrance, and through no wall.

Placing the walk in it

From 9:13 to 13:36 of the video, the software keeps an unbroken run of 691 camera positions. Seven numbers place the whole run in the room: a size, a rotation and a shift. They were fitted to landmarks read by hand in 8 frames, such as the mirror frames, the doors and the edges of the bar, and to the black floor mat. Because every frame shares the same seven numbers, no single frame can be bent to fit. Then 33 sightings of the hanging lamps were matched, leaving out lamps seen in the mirrors.

• The lamps miss by a median of 4.8 pixels.
• The mat’s fitted rectangle lands on the real one in a frame that was not read by hand.
• The tip jar on the bar, worked out separately from 4 frames over nearly 4 minutes, lands within 2.5 in.
• Against these, the points read by hand fit less well, with a median miss of 6 pixels and the worst, around the fridge at the end of the bar, near 90 pixels in the frames closest to it. So the placement is judged good to about 6 in, and not better.

Times below are minutes and seconds into the video. Places are measured from the building’s west wall, on Federal Street, and its north wall, on Jackson Boulevard.

Which way the head faced

The Glass camera sits on the head, so each camera position also says which way the head faced, from 7:59 to 13:36. Checked against frames worked out separately from landmarks, it agrees to a median of 0.8° in direction and 1.1° in tilt on the four frames where that works well. It differs by up to 11° on three where it works poorly. This is the direction of the head, and not of the eyes.

The software lost its place ten times for over a second, the longest for 13.7 s. Across those gaps, and out to the doors at both ends, points tracked from frame to frame carry the direction along, corrected to the positions on either side. When stretches of 14 s were hidden and predicted this way, the median error was 1.3°, though the worst stretch reached 44°. Looked at against the room, the frames in the gaps fit, apart from a fast turn at 12:51 and 12:56, which was read by eye to about 10°. So inside a gap the direction can be 10° to 20° off for a few seconds.

The Glass sensor readings on the walk page come from a second trip, on 15 May 2014, while the video is from 3 April. At the same moments they differ from the measured head by a median of 81°, as readings from another day should, and their rates of turning do not correlate (r = 0.04).

Estimated, not measured

• Heights. They rest on the 84 in doors and the checks above. The placement of the whole visit puts the walker’s eye at about 62 in, against a measured 69.25 in. Frames worked out alone give 64 to 68 in. That gap is unexplained. Scaling the heights to close it would contradict the doors, and would move the walker a median of 5 in.
• The bar’s distance from the north wall leans on a short wall that TJ marked on the plan.
• By eye from photographs: the frames of the side door and window, the pantry, the back counter, the depth of the bars, the brown cabinet and the retail shelves.
• Copied, not seen: the panes of the entrance’s southern door set, from the northern one; the heights of the Federal windows, from the Jackson window; the east wall’s curve and upper windows, from the west wall’s.
• The room may have changed. The photographs are from 2015, and the visit was in 2014. The frames agree where they can be compared.
• The stair by the coffee bar goes down on TJ’s memory, and because the plan draws the whole flight, unlike a stair further south that goes up.
• The building’s south end was not checked. Only the north end, about 107 ft, is modelled in detail.

The walk and its instrumentation were a team effort by Iota, at SapientNitro. This model was made in 2026, from a leasing plan of the building, street-level imagery from Google, and photographs published by Brian’s Coffee Spot, none of which is reproduced here. Building: the Monadnock, Chicago, by Burnham and Root (1891) and Holabird and Roche (1893).