Skip to content

symfonic.capabilities.memory.candidate_signals

candidate_signals

What a candidate is scored on, and how far it travelled to be seen.

Two small decisions the coordinator makes about every candidate, kept together because they are the same decision seen twice: one fills in the signals a store did not measure, the other measures the distance a store did not report. Both answer 'what do we know about this candidate that the store did not tell us', and both must stay conservative -- inventing either would rank a memory on a number nobody produced.

fill_signals 

fill_signals(signals: RelevanceSignals, *, salience: float) -> RelevanceSignals

Supply the one signal the record already carries, if the source did not.

Salience is on the record, so a source that restates it is duplicating state and a source that omits it is not admitting ignorance. Filling it here means every candidate is scored on at least one signal.

Source code in src/symfonic/capabilities/memory/candidate_signals.py 
22
23
24
25
26
27
28
29
30
31
def fill_signals(signals: RelevanceSignals, *, salience: float) -> RelevanceSignals:
    """Supply the one signal the record already carries, if the source did not.

    Salience is on the record, so a source that restates it is duplicating state
    and a source that omits it is not admitting ignorance. Filling it here means
    every candidate is scored on at least one signal.
    """
    if signals.salience is not None:
        return signals
    return replace(signals, salience=salience)