Every camera's most recent heartbeat, refreshed automatically.
You're seeing all 52 centers assigned to you. Camera health covers your whole assignment, so the center picker up top is off here.
Every number here comes from one row per camera — the assessor's latest verdict on it.
Three of these are the assessor's verdict and are stored with the reading. The fourth, out of date, is decided in the browser from how old that reading is, and is never written anywhere — which is why it sits over a status rather than replacing one.
assessed_at is more than 1 hour old. The assessor runs every 30 minutes, so
this means two consecutive runs produced nothing for this camera.
Not a status. A block keeps the colour and the shape of whatever it was last
assessed as, loses its saturation, and gains a dashed ring: this was offline, and we have
not heard since is two facts, and a fourth flat colour would have destroyed the first to
show the second. Front-end only — nothing persists it, and it disappears the moment a fresh
row arrives. It marks cameras and nothing else: a center square answers how much of
the site is working, which a reading's age does not change the answer to.Order matters: offline is tested first, so a camera meeting both sets of conditions is offline, not degraded. Nothing in this section confirms a picture — only that the pipeline behind it is producing files on time.
One cell is not one center. It is one unit at whatever zoom the band is drawn at, and the Cell control says which. The zoom starts on whatever is finest and still legible — a cell per center at 73 or 600, a cell per state at 4,000 — because four thousand squares at a readable size is eleven rows of something nobody can read.
0%, 1–24%,
25–49%, 50–99%, all of them. Inside a center cell those
things are cameras, so the fill is its defect rate. Inside a state or brand cell they are
centers, so the fill is the share of them that needs attention.
This replaced a worst-camera rollup, which painted a center with one dead
camera out of nineteen exactly the same red as a center with nineteen out of nineteen. At 73
centers that is already a wall of red; at 4,000 it is a solid field, and a field carries no
information at all.
The unit has to change with the zoom or the arithmetic eats the signal:
average a camera defect rate across a whole state and every state in the country comes out
between 2% and 11% — fifty-six squares, one colour, nothing said. The share of centers in
trouble spreads across the ramp, and blackouts get their own channel below rather than being
averaged into it.offline = total, for every center in the cell.
This is the event someone is on the phone about — parents at a whole site
seeing nothing — and it is almost always one cause rather than a list of faults. It has to be
findable in a glance at every zoom, so it never shares a treatment with a magnitude.541 all reporting — which
expands on click. Cells are sorted worst-first, so if more remain than three rows will hold,
what is hidden is always the least urgent, and a chip says how many.Color is never the only cue: impaired steps carry a diagonal hatch and a total blackout carries an ✕, so the ramp survives a colorblind reader, a projector and a grayscale print. Below 10px the hatch turns to mud, which is why 10px is the floor — under it the band changes zoom rather than shrinking further.
Staleness never reaches this rule. A cell's colour is about how much of the fleet under it is working, and how old a reading is asks a different question of a different object — so the out-of-date treatment lives on the camera blocks inside each center, on the count beside them, and in the Status filter.
(offline + degraded) ÷ total cameras, per center, as a percentage.
Highest first by default.
Rate, not count: a center with 3 of 8 cameras down is in worse shape than one
with 4 of 40, and only the rate says so.offline + degraded — the absolute number of cameras not fully healthy.
Highest first by default.
Count, not rate: this is the one that maps to how many visits or work orders
the day actually holds.max(now − last healthy heartbeat) across
its offline cameras. Longest first by default.
Degraded cameras are excluded — they are still reporting, so they have no
outage clock running. Centers with nothing offline sort last.Ties fall back to center name A–Z, in both directions. The card shows whichever number you sorted by, so the order is always checkable against something on screen.
Filters don't have to be applied in order. Each list is built from the centers that pass your other choices, and the number beside an option is how many centers you'd be left with — so choosing a state narrows the city list on its own, and an option that would empty the board is greyed out instead of offered. Within one category you can pick as many values as you like and they're ORed together (Texas or Oklahoma); separate categories are ANDed (Texas, and parent-pay). The counts also respect whatever is in the search box.