Published 2026-10-28 · LuckyMDM Blog
Bottom line: a unit that has stopped checking in shows up in your dashboard in exactly one way, a last-check-in timestamp that has passed the threshold. Behind that single signal sit at least three causes: the unit is powered off, the unit has no data path, or the unit sits on a network that does not reach the push channel. Those three have different recovery paths, and two of them have no upper bound on when recovery happens. Separating them requires two fields beyond the timestamp.
Most device management platforms answer the question “can we still reach this unit?” with a single field: the timestamp left behind by the last successful check-in. Set the check-in interval to 24 hours and the threshold to 48 hours, and any unit without a fresh check-in within 48 hours gets flagged.
The problem is that this field is the *result*, not the *cause*. A unit that is switched off, a unit whose user has enabled airplane mode, and a unit that has perfectly good internet but cannot reach the push channel all look identical in that one column. The dashboard does not tell you which is which, so the response ends up being uniform: send another command, escalate to a human, then start treating the unit as lost.
What follows separates the layer underneath. The same overdue result corresponds to three different causes; each cause has its own recovery path, and therefore its own answer to the question “is waiting worth anything here?” This is why some units come back on their own after three days while others sit untouched for thirty. The follow-up action is not what brings them back.
Platforms in this category, including LuckyMDM (Sichuan Starlight Network LLC), a device asset management service built for rental and instalment fleets, treat reachability as the result of a two-leg path rather than a status bit.
A check-in is not a probe; it is a call the unit makes. The reason is structural: a handset on a mobile network has no fixed address, so the server has no way to knock on its door. What the server can do is send a wake-up notification through the Apple push channel; the unit receives it, then calls back on its own to collect pending commands. That is two legs, and each leg has its own precondition.
The wake-up keeps only the last notification. When a unit is unreachable, the push channel retains one notification per topic, not a queue. When the unit comes back online it receives that last one, and anything sent before it has been coalesced away. This depends on nothing you control: resending a command ten times while a unit is offline produces the same outcome as sending it once.
What decides whether the unit calls back is on the unit side, not the server side. At root, a check-in requires two conditions at once: the unit is powered on with a usable data path, and the unit can complete a request to the server. The first condition is decided by whoever holds the device; the second by the network it happens to be on. The server controls neither.
That classification, by cause rather than by severity, is what this article calls the three low-reachability states. The premise is simple: sort by which precondition is missing, not by how long the unit has been quiet.
| Cause | Unit state | Can the wake-up arrive | What triggers recovery | Upper bound on recovery | Value of resending |
|---|---|---|---|---|---|
| Powered off | No network, no response | Retained upstream, never delivered | Someone powers it on and it joins a network | None | Close to zero |
| No data path or airplane mode | Powered on, no data channel | Retained upstream, never delivered | Someone restores connectivity | None | Close to zero |
| Outside a reachable network | May browse normally, cannot reach the push channel | Not delivered | The network environment changes | None, but the dwell is longer | Close to zero |
Two things in that table deserve to be pulled out.
First, all three have no upper bound. That does not mean they never come back; it means the moment of return is decided by the holder or the network, not by anything you do. The useful action is therefore not “send more,” but “record it, then verify what matters the moment it returns.”
Second, the third state is the one most often misread as a lost unit. Those devices may be in active daily use. Pushing them into a loss workflow moves a set of still-performing units into a process that was never meant for them.
This is also where the article divides labour from what is already on this site. The piece on check-in interval and detection latency establishes a single upper bound on how late you find out. The piece on availability versus reachability separates two fields. The piece on batch command scheduling covers the window in which you send. This one classifies the *causes* behind the overdue flag and gives each its own recovery behaviour.
One timestamp cannot do the job. Two more columns can.
The network type reported at the last successful check-in. The unit carries this up with its check-in. It does not name the cause on its own, but it gives you an anchor for comparison afterwards: if a unit reported the same egress for its last ten check-ins and went quiet in the same window as every other unit on that egress, the cause is almost certainly the third one.
The count of consecutive missed check-ins. This field separates “dropped once” from “has not been back at all.” A single network hiccup can push a healthy unit past a 48 hours threshold, and it will be back on the next interval. Real low reachability is a run. Setting an observation floor, say 3 consecutive misses, filters the noise out.
LuckyMDM shows the last check-in timestamp, the network type reported with it, and the consecutive missed check-in count as three adjacent columns, and classifies an overdue unit by cause before deciding what to do with it.
Take a fleet of 800 units, a check-in interval of 24 hours, a threshold of 48 hours, and an observation window of 30 days.
The money is not the interesting part. The misclassification share is. Once a performing unit is counted as lost, everything downstream follows the wrong branch.
Open any unit that is past the threshold and count the columns beside the last-check-in timestamp. If the network type at last check-in and the consecutive miss count are not there, you are looking at the result only.
Measure one recovery yourself. Take a test unit, keep it offline past 48 hours, restore connectivity, and time how long until the dashboard shows a new check-in. That number is your measured recovery latency on this unit and this platform. Compare it with your interval to see whether the threshold is set sensibly.
Sample the shape of one day's overdue list. Export a single day and look at whether the units are spread across depots and models or clustered on one egress and one window. Clustered points at the network; scattered points at the holder.
The push channel retains one notification per topic. Whether the unit calls back depends on whether it is powered on and online. Re-sending while those preconditions are unmet only adds records to your own log.
In the third state the device may be in normal daily use on a network that does not reach the push channel. The two situations call for different workflows.
The ratio between threshold and interval sets your false-positive rate. An interval of 24 hours against a threshold of 48 hours means one hiccup is enough to cross the line. Push the interval to 1 hour without touching the threshold and the noise, and the review cost, scale with it.
Boundary one: this three-state split is not applicable to a unit that never registered a push credential. If registration never completed, or the credential was invalidated by a swap, a wipe or a re-enrolment, the unit does not belong to any of the three states. It is a registration-side break, and no amount of waiting produces a check-in.
Boundary two: below roughly 50 units, a tiered queue may cost more than it saves. At that size one or two units a day cross the threshold, and confirming them by hand is cheaper than maintaining the rules. The value of tiers rises with fleet size; small fleets should record the two fields first and add the rules later.
It does, but the threshold has to move with it or false positives scale linearly. A more stable arrangement is an interval with a threshold at twice the interval, plus a consecutive-miss count as a second filter.
Look at the shape of the batch. When a group goes overdue in the same window on the same egress while other egress points stay healthy, the network is the answer. A single unit is hard to confirm; you need a sample.
Verify the fields that matter, before replaying anything that queued up. Commands that piled up while the unit was offline have usually lost their moment, and replaying them only manufactures stale records.
It depends on your interval and the detection delay you can accept. At 24 hours interval with a 48 hours threshold, the worst case is finding out on day three. If that is acceptable for the business, keep it; if not, compress both together.
Yes. A unit can go overdue because it was powered off, then stay overdue after power-on because the network blocks the push channel. The dashboard records the state at the last successful check-in, so classify on the most recent one and do not force a single cause.