The recording engine captures a window of full-rate data into a permanent archive — manually, or when a condition trigger fires. Those recordings are what engineers compare, what models are trained on and what a quality discussion points back to. A recording with a gap that looks complete is worse than no recording, because nothing warns the person who relies on it.
The rule
A recording is either complete, or it says that it is not — and why. Everything in the engine is designed around that sentence.
What it means in practice
- Captured data is never thrown away to make room. Data that is already in a recording stays in it.
- Incomplete is visible. If a window could not be captured in full, the recording is marked as truncated, with the reason, right in the archive. It never silently looks complete.
- Capture comes first. While a recording is running, background housekeeping steps aside.
- Complete means complete. A window is only closed once all of its data has arrived.
- Retry, honestly. An interrupted recording can be finalised again from the data already stored. A retry that finds nothing fails rather than producing an empty recording that looks valid.
Why it matters
Trigger windows on a grinding cycle are only comparable if each one holds the whole event. A model is only as good as the windows it learned from. And when a number is challenged, the recording behind it has to stand up. A gap-free archive — or an honest label where one could not be guaranteed — is what makes all three possible.