Known-body place test during fleet captures

In plain words

When one person we know about walks past a radio link, does that link react more than the links far away?

Why it matters: One walk on 28 August said yes, but bystander phones in the library could not confirm it, because a phone is not a person and its position is 2 m off. Only a walker whose path we know can settle it.
How it is done: Repeat the 28 August ground-truth walk during normal fleet captures, fit the phone's path to the surveyed plan, and compare links near the walker with links far away.
Where it stands: Proposed. No capture request exists yet; the only evidence is the one walk of 28 August, recorded under EXP-C1.

Asks, formallyDuring a scheduled fleet capture, one known walker within about 5 m of a link's excess path raises that link's injector-scoped CSI spread relative to links at 5 m or more, and the effect repeats across at least three separate ground-truth walks with an interval that excludes zero. Falsified if the near-minus-far contrast is not detectable on repeated walks, which would make the single 2026-08-28 walk an outlier rather than a place signal.

StandingDesigned, not yet run

RoomReal hardware: the ten-node Pi 5 + AX210 fleet in the FIIT library, ch48 HT20, monad01 star (walk-abba-48-night / illum-walk-48-night) or a later mesh; ground-truth walks with the phone's own path fitted to the surveyed plan, ideally with the library empty.

Why this card exists

The fleet's place test on bystanders found nothing: a device within 1 m excess path of a link changes that link's z by +0.04 [−0.07, +0.17] (diary 2026-09-27). On the one ground-truth walk, a link's spread did rise when the walker was within about 5 m of excess path (amp_sd +1.0 to +1.6 z against links ≥ 5 m away). The two cannot be reconciled from existing data: located phones undercount people and carry about 2.2 m BLE error, so the null may come from invisible bodies or from exposure error. A known body is the only test that separates them (diary 2026-09-27).

Setup

Proposed, not booked. A ground-truth walk runs during a scheduled fleet capture (walk-abba-48-night receivers, illum-walk-48-night on monad01), ideally with the library empty. The phone records its own path and is fitted to the surveyed plan on surveyed anchor points, with the fit residual reported per walk. The loaders now scope to the real injector MAC (02:6d:6f:6e:01:NN, not the receiver fill ef:be:ad:de:ad:de), so the pipeline is ready for new walks (diary 2026-09-27).

Evidence base, cited not claimed: the 2026-08-28 walk (CR-003 raster and CR-015 survey sessions, owned by EXP-C1).

Procedure

Repeat the 2026-08-28 statistics under each configuration: the band model, link choice by excess path, and moving against still. Primary contrast: per 2-s bin, a link's injector-scoped spread when the walker is within 5 m of its excess path, minus links at 5 m or more, with two-way fixed effects (bin, link × segment). Report the interval over walks, not over bins. Declare the analysis before the first new walk is read.

Results

None. No session is claimed here.

Limits

One building, one channel, one transmitter geometry unless a mesh (EXP-015) passes first. A star leaves most of the floor inside no link's first metres of excess path, so a null on the star does not rule out a place signal on a mesh. A walker is one body: the test says nothing about crowds, sitting people or counting.

Next steps

Write a capture request for repeated walks inside a fleet capture, with an attended slot and the walk quest named. Move the card to planned when a merged plan holds it.