Reading a QPU Specification Page
Read a QPU spec page as a record of evidence. Every number needs a source, date, method, scope, aggregation, and uncertainty. All the fields must describe the same version of the device. "Not publicly disclosed" is a clear finding with a date. It is a different state from a value that is just missing or out of date.
How should you read a QPU spec page?
A QPU spec page looks like a product sheet. But the useful way to read it is as an evidence record. It is a set of claims. Each claim either comes with the facts that back it up, or it does not. A number with no context is not information yet.
Think of a nutrition label that says "200". Two hundred what? Per what serving? Measured how? A spec number works the same way. Every number needs six fields:
- Source: who published it. The vendor, or an outside team?
- Date: when it was measured, not when it was advertised. Calibration (the chip's tuning) drifts, so numbers get old.
- Method: the test behind it. One example is randomized benchmarking. It runs random gate sequences and fits how fast success drops, to get an average error per gate.
- Scope: which qubits or pairs it describes. Is it one golden pair, or the whole chip?
- Aggregation: how many measurements were boiled down into one number. Was it the best, the median, or the average?
- Uncertainty: the error bars. A fidelity (chance a gate works) quoted to four decimal places with no uncertainty is a red flag.
The two worked examples below show how each field changes what a number is worth.
- Worked example: why don't these two 99.x% claims compare?
- Worked example: how do you turn T1 into a depth budget?
- Run it: a sanity-check circuit for any spec sheetINTERACTIVE
- Change one thingINTERACTIVE
- What is the difference between absent, not disclosed, and stale?
- On real hardware
You’ve read the opening of chapter 11. Pro unlocks the other 6 sections — plus every chapter of every course, with circuits you can run right on the page. That’s $11.99 a month, about the price of a coffee, or $99.99 a year (save 30%). The first chapter of every course, and the whole math course, stay free.