The electricity meter, with a receipt.
Think of the electricity meter at home. It measures consumption non-stop, and the utility keeps only what it needs for the monthly bill. Nobody keeps every second of every home: it would cost too much.
Now imagine the bill is disputed. The utility must show that what it kept is faithful to what the meter read, and that nobody changed the numbers afterwards. That is the problem of all telemetry: keep less without losing the ability to prove.
Telemetry is exactly that: readings machines send on their own, from a vibration sensor in a mine to a satellite 500 km up. "Certified" means everything we do with that data comes with a proof anyone can check. See the glossary for the words of this section.
One question per tool.
Each tool solves one concrete question. They combine: data stored by TUBE is queried by CLAMP, timestamped by PLUMB and recorded by GIRDER, and each proof adds to the previous one.
The tools together, inside the client's infrastructure.
Operating structures built on TUBE.
With the guaranteed-tolerance compressor in place, it started to carry other everyday pieces for people who run sensors. All of them were measured on real data (robot motion, car CAN bus, energy meters, depth video, self-driving laser scans), with the measurements next to the code, including the ones that did not work.
- SLUICEthe queue that, when bandwidth runs short, delivers everything at a wider, declared tolerance instead of dropping readings.
- COFFERthe black box that blurs the past instead of forgetting it, and states how much it blurred each stretch.
- STRATAage-based retention that, for any question, returns a range guaranteed to contain the answer.
- SPLICEjoining sensors with different clocks: every pair comes out right, ambiguous or impossible, instead of silently mismatched.
- CAULKgaps in the data filled by the sensor's physical limit, with the likely range or a refusal stated with a number.
- TRESTLEthe fleet that answers questions without shipping raw data to the cloud, only small indexes.
- TUBE2Dimages and depth video with a per-pixel guarantee.
- PILEthe laser-scanner point cloud that proves "there is no obstacle in this area".
- WEIRrefuted layered delivery per subscriber. It did not work: the simple alternative won on everything, and the result stayed on record.
The cases.
Every case starts from a real problem, with public data, and shows what the tool did and what was measured.
50 demos on public data.
Earthquakes, blackouts, glucose, a nuclear treaty, gravitational waves, stock trades, flights, batteries, ships and robots. Every demo runs in the browser, on real data, with the tolerance checked.
How much each reading may vary after it is stored. Whoever uses the data chooses it. in the glossary →
The guarantee comes with what you need to redo the check. in the glossary →
When a reading escapes the prediction, the compressor notices. It becomes a free detector. in the glossary →
Where this could be wrong.
The tolerance is the user's choice
The guarantee holds for the agreed tolerance. If it is too wide for the job, the file is faithful to the rule and useless for the question.
Not everything wins
There are cases where a competing compressor stores less, and one case where the idea was refuted. The pages show both sides.
Proof is not interpretation
The proof guarantees the data is what was measured, within tolerance. What it means is still the analyst's job.
Mostly public data
Most measurements come from open data. In production, each sensor needs its own tolerance and its own test.
Does this apply to your case?
Tell us in two lines what you need to decide or measure. The first conversation is to see whether measurement solves your case, and if it does not, we say so.
- The tool family and the structures were measured on real data; each page lists its own sources and numbers.
- The technical section describes the systems in detail.