In a 2 m corridor the far wall counts as "something in front of the wall": ground past it is marked hidden and the phone asks to see past it #160

In a corridor about 2 m wide, the corridor's far wall counts as "something in front of the wall". The ground behind it is marked hidden, and the phone asks the tester to see past it: "Something is in front of the wall here. About 8 ft right of your meter. Look at it from the side or step around it." Nothing stands there, and there's no way round a wall. A side yard with a fence 5–6 ft from the house would hit the same thing.

Clip: The corner, then "something in front" in the corridor (1:20–1:45 in the recording)

Seen in the build 5.1 field test on 2026-09-27 (04:00): TestFlight 0.1.0 (5.1), iPhone 15 Pro Max (iPhone16,2, LiDAR, scene depth and mesh on), iOS 26.6.2. Indoors: a wall phone standing in for the meter, on a short wall with chairs, which turns a corner about 2 ft right of the meter into a corridor about 2 m wide.

What happens (video times; the packet clock is video + 246535.6 s).

• 1:27–1:31: "What's at the right end?" appears at the corner. The tester answers "It turns a corner" and the next wall is marked (the guidance log has walk "Walk round the corner…" met at 1:31). The exported chain matches the LiDAR mesh: wall from −2.4 ft to the corner at 2.0 ft, then wall-right-1 down the corridor to 8.6 ft.
• 1:35–1:44: in the corridor, "Something is in front of the wall here", first "About 5 ft right of your meter" (wall band), then "About 8 ft" and "About 9 ft" (ground band). Purple dashed "Hidden behind something" shapes cover the corridor wall and floor, and the arrow points at empty floor.
• 3:00: a dashed outline sits on the floor at the end of the corridor.
• The tester, a little later: "there's so many random values."

Data (packet 1.1; scene.json).

• The mesh measured the corridor's far wall 5.0–5.4 ft out from wall-right-1 (facing[].depth_ft 4.9999–5.3999 over 2.0–8.6 ft).
• The ground there was seen out to 5.5–6.0 ft (coverage.observed ground out_ft), which is as far as the floor goes.
• The ground depth rows run out to groundDepthReach = 5.18 m (17 ft). Every row past the far wall projects onto a pixel whose depth is the far wall, nearer than the row, so it is recorded as hidden (CoverageMap.recordDepth, CoverageMap.swift#L87). The see-behind task then asks for those rows.
• Later the server asked to "Walk along this stretch about 5 ft out from the wall" (4:59), which in this corridor is the far wall itself, and "Show the ground from 9 ft to 15 ft right" (5:09). Both ended "I can't get there".

Cause. The hidden-row test can't tell "an object stands in front of the wall" from "the open space ends here". A facing surface the mesh has already measured is where the clear ground ends, not something to look past.

Fix direction.

• Treat ground rows beyond the measured facing depth (from MeshProbe / facing on LiDAR phones, or the depth image's far surface) as the end of the space: not hidden, and never a see-behind target.
• Only raise see-behind for hidden rows nearer the wall than that facing surface.
• Word the card for this case differently, for example "There's a wall or fence about 5 ft out. That's the space you have."
• Test (HouseScanKit): a camera in a 2 m corridor, with depth frames seeing the far wall, adds no hidden ground rows past it, and GuidancePlanner raises no seeBehind there.

Not explained yet: the wall-band "About 5 ft right" hint (1:35.9–1:37.5, band wall). Near the corridor entrance the mesh has a gap at the wall line, possibly a door recess, so what stood in front there isn't clear.

Also seen in this run.

• "Take a step back" was up about 80 s in total, 24 s straight at 1:44–2:08 in the corridor. The card has no button. The trajectory has the camera 0.3–1.1 m from the wall there, so the prompt followed the rule (under 1.2 m, cleared at 1.4 m). But in a 2 m corridor, 1.4 m means standing against the far wall. The tester: "How can I go more back? It took 200 photos already." The same space problem as above; see The walk card changes every ~3 s, and target distances jump backwards #84 for the card churn.
• The manifest's guidance span_m values for tasks on the turned piece are meter-frame x, not chain s. "About 8 ft right of your meter" is logged as [0.569, 0.606], and the server's "Show the wall from 1 ft left to 9 ft right" as [-0.3048, 0.5859]. Anything reading the packet sees these tasks at the corner.
• The server's wall task: progress went 5% → 16% → 5% → 0% (4:21–5:00), with no target dot on screen ("Where's the blue dot?"). See The target ring and arrows show no progress, aren't explained, and the arrow can contradict the card #81.
• 3:48: "Wall ends here", tapped while standing back by the meter's wall. The strip had warned "leaves out 13 ft you walked". The exported right end was set again at 4:13, after the server's past-end task, 2.0 m down the corridor (8.5 ft along the chain). See "Can't get there" during the walk can put the wall's end at the meter, and the rest of the walk is dropped #24, No way to finish a scan early: stopping took 7 rapid "Can't get there" taps, each silently setting ends #82.
• 1:29, 2:21 and 3:06: three "Your phone lost its place for a moment" prompts, with tracking limited.