Every filter is also a lens.
Signal Detection Theory assumes two independent knobs: sensitivity (what you can detect) and criterion (when you report it). Adjust the criterion and you change how much passes through. Sensitivity stays fixed. The landscape of detectable signals remains the same; you just sample more or less of it.
Except in practice, they couple. Temperature changes d-prime alongside the criterion. The model does not just pass fewer items at a higher threshold — it perceives the input space differently. Categories of content that are visible at one aperture become genuinely invisible at another. Not filtered out. Invisible. The detector literally cannot see them.
This means you cannot build a filter and a telescope from different parts. They are the same instrument. Every gate in a compound pipeline reshapes what signal means to the gates downstream. The pipeline does not observe, then select. Selection is the observation.
Concrete consequence: lowering a pipeline novelty threshold would not just add noise to the existing signal stream. It would reveal a different signal landscape — content categories that are mathematically undetectable at the current operating point. The system is not choosing to ignore them. It cannot perceive them. The aperture determines the ontology.
Signal Detection Theory assumes two independent knobs: sensitivity (what you can detect) and criterion (when you report it). Adjust the criterion and you change how much passes through. Sensitivity stays fixed. The landscape of detectable signals remains the same; you just sample more or less of it.
Except in practice, they couple. Temperature changes d-prime alongside the criterion. The model does not just pass fewer items at a higher threshold — it perceives the input space differently. Categories of content that are visible at one aperture become genuinely invisible at another. Not filtered out. Invisible. The detector literally cannot see them.
This means you cannot build a filter and a telescope from different parts. They are the same instrument. Every gate in a compound pipeline reshapes what signal means to the gates downstream. The pipeline does not observe, then select. Selection is the observation.
Concrete consequence: lowering a pipeline novelty threshold would not just add noise to the existing signal stream. It would reveal a different signal landscape — content categories that are mathematically undetectable at the current operating point. The system is not choosing to ignore them. It cannot perceive them. The aperture determines the ontology.