Reading this map
lowattenuationhigh
unsureopacity = supportsure
Solidity is support, relative to this frame: the best-supported cell in the picture is drawn at full strength and anything carrying at least half of it is drawn solidly enough to see. A sparse mesh supports nothing strongly in absolute terms — this one reaches an unmeasured amount — and drawing it in proportion made a real person invisible. The ranking is unchanged; the scale is not a probability.
Dashed ring around a person = position with no calibrated error bound. Support (opacity) is not precision.
In 3D the dot sits at the height the field peaked at, and its shell is that radius across and the whole storey tall. The nodes sit in a near-flat plane, so the reconstruction tells floors apart and does not resolve metres of height within one. The shell says so.
A person is on exactly one storey. A ring on the floor under the dot says which. Whoever is on the storey you are looking at is drawn in gold and called a person; the same person seen from another storey’s tab is greyed and labelled with the floor that owns them, never with the word “person”. Nobody should read three people upstairs when everyone is downstairs.
When the shell is wider than the room, the room is the answer. A disc asserts a centre, and on a sparse mesh in a big house the calibrated radius is sometimes half the floor across — at which point there is no centre to assert. The map then fills the storey instead of drawing a disc, keeps a small hollow ring where the evidence points so you know which end to look at, and says someone on this floor · not located within it, in those words, on the plan and in 3D. That is a weaker claim than a dot and it is the one the radio supports. The fix it implies is more nodes: the same table draws a tight disc on a mesh with enough of them.
Two shells that touch are one mark. At this error, two people standing closer together than the sum of their radii cannot be told apart, so they are counted once. The undercount is the honest direction: two dots there would claim a resolution the radio does not have.
A soft blob is an estimate, not a body. Its width is the measured uncertainty in where the disturbance was, so a wide blob means the radio narrowed it down to a wide area — not that a wide person is standing there. There is no silhouette, no skeleton and no anatomy in this picture, because none of that is measured.
A filled blob would mean a detector that has been checked against a person whose position was known. Nothing here has passed that check, so every blob is drawn as an outline and labelled as movement rather than as a person.
Every number on a mark carries the time it was last measured. When a mark stops being measured it fades to a hold, then to unknown, then off the map. A mark that is being held is a memory of a measurement.
If the hub stops receiving from the mesh, every mark goes to unknown at once and the map says it is not measuring. A page that quietly kept drawing a calm, breathing person off a dead radio is the failure this was built to avoid.
Breathing is only ever shown when a still session with reference readings has been scored and passed. When it is shown, the gentle expansion follows the measured rate and is an illustration of that rate — it is not a measurement of how far anybody's chest moved. There is no heartbeat here, at any rate and at any timing.
Posture and whether a mark is a person or an animal are both shown as unknown, and they mean it: nothing in this house has ever been labelled to tell them apart.
A small hollow ring around a node is an experimental score, not a person and not a place. It says one node's own links changed more than everybody else's, which beats chance and is far from usable.
Breathing and heart rate. There is no heart-rate reading here and there will not be: the nodes report amplitude and no usable phase, the links cross rooms, and a heartbeat moves a chest by micrometres. Breathing is used only to keep a person who has stopped moving on the map — a presence claim, never a rate. The reasoning is in docs/vital-signs.md.
Raw shows what is being filtered. The switch above the map is off by default and the default is the honest picture: solidity is support, and a reconstruction the mesh cannot back is drawn faintly. Turn it on and solidity becomes magnitude instead — every cell the reconstruction put something in is visible in proportion to it, whether or not the geometry supports it — and every blob the tracker considered is marked with a small dashed ring and its support number, including the ones the gates then removed. Most of those are the ringing an underdetermined inverse leaves beside a bright blob, which is why they are gated; being able to count them is why the switch exists.
Unpainted areas are not empty rooms — they are places the hub has no reading for, either outside the building you drew or not reconstructed. Colour is magnitude; how solid it looks is how much the radio supports it.
The greyed area is not covered. Every link runs from one node to another through the inside of the house, so a person outside a wall — or in a corner no link crosses — is off every link path, and the reconstruction has no voxel to put them in. The veil is drawn from the mesh’s own weight matrix, not from a guess: clear where the geometry supports a position, light where it barely does, heavy where it cannot. Coverage for this level is unknown. This hub does not report how coverage varies with height. This hub does not report how far past the walls it reconstructs.
Approach is not detected, and height is not measured. Approach detection is unproven. Movement outside the walls should lift the variance on the nearest node's links, and on this building it does not: with someone standing outside the front door the nearest node separates from the empty house with AUC 0.563, and beside the kitchen 0.504, where 0.500 is a coin. Treat the halo layer as silent, not as an all-clear — the house cannot currently tell you that somebody is at the door. Height is not measured. The reconstruction does compute a vertical centre, but on labelled windows in this house a three-metre difference in real height moved it by 0.18 m (AUC 0.588), and feeding random link activity through the same geometry lands in the same 2.5–3 m band — so the number describes where the mesh's attention sits, not where a body is. Which floor someone is on is not something this install can tell you yet.