salience engine
From molecule to meaning
Salience does not live inside an object. A dopamine molecule is not the good chemical, a glucose cue is not food, a text message is not hope. These things become signs only inside systems organised around survival, repair, reward, attachment, and self-continuity. The playground walks one cue up an eight-rung ladder: from a neutral physical difference, through constraint relevance and proto-salience, into sign function, incentive value, attention, narrative binding, and finally over-salience.
A normal sign points at the world. A runaway sign is no longer in the world; the world is now read through it.
Wanting is not liking
Salience is not the same as liking. Incentive-sensitization research separates wanting from liking: a cue can seize the attention budget and drive pursuit while delivering little pleasure. Select the possible rival, an object with low intrinsic reward but high cognitive and affective charge, and watch it climb the field anyway. An object you do not want can still run the day. It is the same reason a phone notification can feel more charged than the conversation it announces.
Uncertainty is the fuel
Limerence is not the strongest love. It is the runaway regime of an ordinary salience system. The dangerous term is uncertainty: ambiguous reciprocation keeps the prediction loop computing, where clear affection or clear rejection would let it settle. Switch the signal regime from stable to ambiguous to volatile and watch over-salience climb without any change in how much the object is actually liked.
What the model is and is not
The eight object dimensions and eight field weights are a deliberate compression. The scoring formula has no biography and no time axis; a single signal regime stands in for the loop dynamics. It is a comparative instrument, not a predictive one. The calibration table runs each of the six objects under the limerence preset and compares the model's salience to a reader-assigned expected value. Close agreement means the model's shape matches a careful reading of how that kind of cue typically behaves. It does not mean the model has met any particular person.
Lowering the loop
The path back from over-salience is not an argument with the interpretation. It is structural. Raise reality correction and habituation, raise attention temperature so the allocation stops being winner-takes-most, cut the intermittency feeding the uncertainty term, and introduce competing salient objects. The reality-correction preset shows the field calming: salience pulled back toward evidential weight, attention spread again across the world.
Model changelog
- replaced the original demo time-wave with a deterministic signal regime, so the configuration alone is the model.
- kept the eight-dimensional object profile but renamed it to matter, constraint, proto, sign, reward, affect, cognition, narrative.
- made the eight-rung ladder a first-class structure: matter, constraint, proto-salience, sign, value, attention, narrative, over-salience.
- added a sixth object, a shared song, to isolate narrative binding from reward.
- replaced the single salience number with a six-metric panel: salience, over-salience, attention share, concentration, meaning, stability.
- four status tiers, ordinary to runaway, classify the focal object by its salience band.
- softmax attention over all six objects feeds the attention rung and the concentration metric.
- sweep, sensitivity, snapshot comparison, and a comparative ranking added throughout.