Tilt-down prompts never ask for the ground in front of the meter

Part of the cause of #24. Split out so the guidance can be fixed separately from the "Can't get there" tap.

In field test run 1 (PR #23), "Tilt down to show the ground" asked for the strip 3 ft 3 in, 2 ft 9 in, 4 ft 6 in, 7 ft and 8 ft 9 in right of the meter (0:27–0:46 of the recording). It never asked for the ground in front of the meter, which is the first thing GuidancePlanner.reach needs. Replaying the scan in PR #23 shows the ground 2 ft and 4 ft out by the meter was seen twice only in photos 41–43 of 44.

Why. laggingBand only looks within 1 m of the phone's position along the wall, and it only asks for the ground where the wall is already covered (GuidancePlanner.swift#L141-L155). At the start, beside the meter, the wall wasn't covered yet, so there was no ground prompt. By the time it was covered, the tester had walked on. This is also why the requested distance follows the phone and can go backwards (3 ft 3 in, then 2 ft 9 in).

Possible fixes (your call):

ask for the ground in front of the meter right after the close-up, before "Walk slowly to your left";
or have laggingBand or firstHole prefer the gap nearest the meter, since reach counts outward from it.

Clip, 0:27–0:46 of the recording:

02-tilt-prompts-skip-ground-at-meter.mp4

Addressed in 3c7b83a on t3/ios-mvf (merged in 75dea41).

What changed

GuidancePlanner.preferredTask now asks for the ground in front of the meter (about 1 ft either side of it) before anything except "Take a step back" and the look-past step. It keeps asking until 80% of that ground is covered or skipped. "Can't get there" skips it and the walk moves on.
An aim request now keeps its distance while the planner's window drifts by less than 0.3 m. The card shouldn't go from 3 ft 3 in to 2 ft 9 in for the same ground.

What's left
Nothing for this issue that I can see. One behaviour to know: this request outranks walking. If you walk off before it's met, the card keeps saying "Tilt down to show the ground, at your meter" until you go back or tap "Can't get there".

To confirm on the first TestFlight build with 3c7b83a
After the close-up (and "Take a step back" if you're close), the first walk card should be "Tilt down to show the ground / The strip along the wall, at your meter", before "Walk slowly to your left". Tilt down at the meter from two spots a step apart. The card should clear and move on. Later tilt prompts on the same stretch shouldn't change their distance by a few inches.

Same in field test run 2 (build 3.1). The tilt-down prompts asked for the strip at 5 ft 3 in, 4 ft 9 in, 4 ft 6 in, 4 ft and 4 ft 3 in right of the meter, and never in front of it or to its left. That's why reach was still 0 on the left at the 0:53 "Can't get there" (#24). The distance also went backwards again, 5 ft 3 in → 4 ft, as in run 1.

Checked run 2's symptoms against t3/ios-mvf (3c7b83a). Build 3.1 came from e9e405d, which doesn't have 3c7b83a.

Tilt prompts only 4 to 5 ft right of the meter, never in front of it: fixed by 3c7b83a. The ground in front of the meter is now the walk's first request, and it outranks the lagging stretches.
Distance going backwards, 5 ft 3 in to 4 ft: only partly fixed by 3c7b83a, and fixed in Tests for #24, #28 and #29: synthetic walk-guidance cases modeled on field runs 1 and 2 #56. 3c7b83a held a request only while the planner's target stayed within 0.3 m of it. Run 2's steps were each under 0.3 m, but 5 ft 3 in to 4 ft is 0.38 m, so the card would still have jumped. Tests for #24, #28 and #29: synthetic walk-guidance cases modeled on field runs 1 and 2 #56 keeps the request while its ground is within 1 m of the phone.
New in Tests for #24, #28 and #29: synthetic walk-guidance cases modeled on field runs 1 and 2 #56: a ground cell only counts once it's seen from two places at least 0.25 m apart. Run 2's amber cell had been seen from one place only, and a homeowner who tilts down without moving can't clear the request at the meter. The card now adds "Take a small step sideways as you look."

Next field test (a build with #56): the first walk card after the close-up should be "Tilt down to show the ground ... at your meter". Tilt down, take a small step sideways, and check the card moves on. Later tilt prompts should keep their distance while you stay near that spot.

Fixed on t3/ios-mvf in 3c7b83a, merged at 75dea41. Build d02cc11 or later for runs 2 and 3; it is green (CI run 36279804539).

The walk's first request is now "Tilt down to show the ground" at the meter, before "Walk slowly to your left". It's met once 80% of the ground within 0.3 m of the meter is covered from two positions. "Can't get there" moves on from it.
The tilt prompt no longer jumps backwards (3 ft 3 in, then 2 ft 9 in). An unfinished tilt request keeps its stretch while the new one is within 0.3 m of it.
Since d0ff1be the walk asks for the wall only up to 4.5 ft, the most a view from the 2 m stand-off shows together with the ground. On the synthetic replay the wall tilt-up prompts along the walk went from 3 to 0. Greater heights, when a check needs them, come as one server-requested view after the upload. That height rests on an assumed field of view of 30° to 31° half angle. If your phone still gets frequent "Tilt up to show more wall" prompts during the walk, please note where in the recording.

Build 4.1 field test (TestFlight 4.1, beta 45017ba, iPhone 14 without LiDAR, three runs on 2026-09-26): inverted. The ground in front of the meter is now the first thing asked, and until it's met it is the only thing asked, so it blocked the start in all three runs.

Run 1: "Tilt down to show the ground / The strip along the wall, at your meter" took about 151 s of the 225 s walk (0:33–4:13). It was met once at 0:54, then re-asked 3 s later. Cause: the ground came from a table plane (Ground height taken from a table plane, so every height is about 3 ft too high #62).
Run 2: 70 s (0:28–1:38), with the correct ground. It finished only once the tester stepped back, because each ground row needs sightings from two spots 25 cm apart, and nothing on screen says so.
Run 3 (a real house): 73 s (0:21–1:34), never met. The tester ended the test. Likely cause: the wall was placed behind the real wall (Meter tap on an estimated plane put the wall 1.1 m behind the real wall and 46° off, and nothing corrects it #69).

There's no timeout or skip. See #77. Analyses: PR #60 (run 1, run 2, run 3).

WS-07 object understanding (#28)

Navigation Menu

Repository navigation

Tilt-down prompts never ask for the ground in front of the meter #28

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Tilt-down prompts never ask for the ground in front of the meter #28

Description

Activity

AidenJohnston commented on Sep 26, 2026

AidenJohnston commented on Sep 26, 2026

AidenJohnston commented on Sep 26, 2026

SamGu-NRX commented on Sep 27, 2026

AidenJohnston commented on Sep 27, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions