Everything in the application is one idea and the parts that build it. This page is the idea, drawn — written for counsel, not for engineers.
Before any claim makes sense, this is the object. There is no cable to a computer and no gateway doing the thinking. Everything below happens inside the box.
The dashed amber path is the invention. Everything else is ordinary hardware — the novelty is that the answer produced inside the box is what programs the box's own next wake-up.
A battery-powered sensor computes the diagnosis on the chip bolted to the machine, and uses how fast that diagnosis is changing to decide when to wake up and measure again.
Conventional monitoring draws a line and waits for a reading to cross it. That is the prior art, and it fails twice: slow damage never crosses the line until it is late, and a harmless one-off spike crosses it and cries wolf.
We do not ask how big the reading is. We ask how fast it is moving, compared with how much this particular machine normally wanders. A quiet machine that starts drifting is caught early. A noisy machine that always wanders is not punished for it.
Same machine, same data, two decision rules. The right-hand rule also decides something the left cannot: once the slope lifts, the sensor shortens its own sleep and starts watching more often.
Each of these exists because the loop needs it. They are how the one idea is made to work on a real machine, and they belong in the application as dependent claims, not as rival headline inventions.
A verdict needs independent indicators to agree, and the more severe the verdict, the more agreement it demands. Measured: one extreme reading on its own is wrong 36.6% of the time on healthy machines. Requiring corroboration puts that under 1%.
A physics rule engine and a trained model judge the same measurement separately. When they disagree and neither is confident, the output says so rather than inventing an answer. No commercial system we surveyed publishes uncertainty as a real result.
When the technician closes the job saying "right" or "wrong", that machine's thresholds move — up if we cried wolf, down if we missed it. The model is never retrained. This is the part no prior art was found for.
The box works out how fast the shaft turns from the vibration alone, and also how much that speed wobbled — then discards any measurement taken while the machine was speeding up. This is what makes variable-speed machines work on the same hardware.
The draft currently reads as six independent inventions. That is why it is hard to follow, and it invites an objection that the application covers more than one invention. The fix is not to delete the work — the description should stay, because it is what supports every later claim — but to stop presenting it as eight parallel inventions.
| Now | Proposed | Why |
|---|---|---|
| Six independent claims scheduling · staging · fleet knowledge · on-device pipeline · the system · cadence |
One concept the loop: on-device diagnosis drives its own measurement schedule |
Everything else is an embodiment of that loop. One idea is defensible; six competing headlines are not. |
| 27 claims | Deferred | A provisional's job is to describe the invention and secure the date. Claims are drafted for the complete specification, twelve months later — with a year of market evidence to aim them. To confirm with counsel. |
| Full description, 20 experiments, 29 figures | Keep all of it | Length in the description costs nothing and buys everything: you can only claim later what you disclosed now. This is the part that should stay long. |
The practical reading: file quickly and broadly, argue narrowly later. Every week the application sits undrafted is a week the priority date is not held — and the supporting material is already written.