Part 5 · Reference
Thresholds
Every number that decides whether something happens, read out of the code.
01What it is
The thresholds behind inference, voice, learning, intel and the visual channels. Each value is imported from the module that defines it.
02Why it exists
A reference written by hand is wrong within a month, and a wrong reference is worse than none: somebody tunes against a number that was true in March and nothing tells them otherwise.
Every value on this page is imported rather than retyped, so the page cannot say 0.35 while the code says 0.5.
03How to use it
- 01Find the constant by name. The name is the one used in the code.
- 02Move inference thresholds down to get more reads that mean less, and up to get fewer that mean more.
- 03Leave the learning floors alone unless you have a reason. They exist to stop conclusions being drawn from eight results.
04What good looks like
- You give it
- Raising MEANINGFUL_EVENTS_TO_SYNTHESISE from two to three.
- You get
- Fewer reads, each carrying more corroboration.
- Why
- Synthesis needs that many meaningful events inside the window before it runs at all.
Derived from the rules in the code
05What weak looks like, and what it costs
- You give it
- Lowering CONFIDENCE_FLOOR to surface more windows.
- You get
- Reads publish that the system does not believe, and the copy that says "confidence is low" stops appearing.
- Why
- The floor is what separates a read from a guess. Below it the product says so rather than telling a story.
Derived from the rules in the code