Counterfactual Proximity Weighting

Prime #

757

Origin domain

Cognition Judgment Decision Making

Subdomain

reward and signal learning → Cognition Judgment Decision Making

Core Idea

The internal signal an agent assigns to an outcome is graded not by the outcome alone but by its distance to a nearby counterfactual of different value: a near miss carries a near-reward signal, a near-catastrophe a near-loss signal. The system updates as if the counterfactual partially happened, modelled as s(o) = α·r(o) + (1−α)·w(d)·c(o).

How would you explain it like I'm…

So Close It Counts

Imagine you almost catch a ball but it slips right through your fingers. You did not catch it, yet your heart jumps as if you nearly did, just because you were so close. A near miss feels almost like the real thing. How close you came changes how big the feeling is.

The Near-Miss Feeling

Counterfactual proximity weighting means the signal a system feels about an outcome depends not just on what happened, but on how close a different outcome was. A near miss carries a little of the reward you almost got; a near-disaster carries a little of the loss that almost happened, even though nothing bad actually occurred. The closer the other outcome was, the more strongly it colours the feeling. This is about how the signal is made, not how it is read later: your behaviour or worry updates as if part of the close-by outcome really happened. And it is not silly, because in a noisy world a near miss really does tell you that you were operating close to a failure line.

Distance to the Almost

Counterfactual proximity weighting is the pattern where the internal signal an agent assigns to an outcome is not a function of the outcome alone but of the distance from it to a nearby counterfactual outcome of different value. A near miss carries a near-reward signal though no reward was delivered; a near-catastrophe carries a near-loss signal though no harm occurred. It concerns signal generation, not interpretation: the internal update, to behaviour, attention, or a risk register, is graded by proximity to a counterfactual rather than by the realised value alone. When the actual outcome was a clean miss but a catastrophe lay close by in the relevant state space, the system updates as if the catastrophe partly happened. This is not irrational when outcomes are noisy, because closeness genuinely carries information about how near a failure boundary you were; it becomes maladaptive only when the proximity weight is miscalibrated relative to the true counterfactual probability.

 

Counterfactual proximity weighting is the structural pattern in which the internal signal an agent or system assigns to an outcome is not a function of the outcome alone but of the distance from that outcome to a nearby counterfactual outcome of different value. A near miss carries a near-reward signal even though no reward was delivered; a near-catastrophe carries a near-loss signal even though no harm occurred. The pattern concerns signal generation, not interpretation: the internal update, to behaviour, attention, a risk register, an investment posture, is graded by proximity to a counterfactual rather than by the realised value alone. The commitments are an agent generating an internal signal in response to an outcome, a nearby counterfactual of meaningfully different value, a proximity measure in the relevant state space (causal distance, perceptual similarity, temporal nearness), a decreasing function of that distance weighting the counterfactual's value into the signal, and a downstream effect calibrated to the signal rather than the realised outcome. A compact model fits it: with r(o) the realised value and c(o) the value of the nearest counterfactual at distance d, the signal is s(o) = alpha*r(o) + (1 - alpha)*w(d)*c(o), where w decreases in distance. The classical assumption alpha = 1 (signal depends only on the realised outcome) is the limiting case; the prime is everything to its left. The signal is not irrational under noise, since a near miss genuinely carries information about how close to a failure boundary the agent was operating; it becomes maladaptive only when the proximity weight is miscalibrated relative to the true counterfactual probability.

Broad Use

• Operant learning and gambling: the near-miss effect — "two cherries and a lemon" drives approach behaviour resembling a partial win though nothing was paid.
• Decision theory and emotion: regret and relief scale with the proximity of the alternative, not its mere existence.
• Safety engineering: leading-indicator weighting enters near-catastrophes into the risk register at a fraction of the catastrophe's cost, by causal proximity.
• Sports: almost-victory drives re-engagement more than clear defeat, the near-win carrying a partial-reward signal.
• Negotiation: an offer "just barely" rejected anchors the next round more than one wildly off-base.
• Driving: a near-collision recalibrates risk perception disproportionate to the objective hazard density.

Clarity

It separates the generation of an internal signal from its interpretation, reframing apparently irrational responses as correct responses to a counterfactual-proximity signal the outcome record does not capture, and makes the proximity weight itself the auditable object.

Manages Complexity

It reduces a confused mix of "irrational" behaviours to one mechanism with three direction-keyed interventions: suppress the counterfactual neighbour to dampen, surface it to amplify, or re-ground its weight in a causal model to correct it.

Abstract Reasoning

It relocates the pathology from the response to a single parameter — whether the proximity weight w(d) tracks true counterfactual probability or mere perceptual vividness — since a near miss carries genuine information about operating margin when outcomes are noisy.

Knowledge Transfer

• Across reward systems: a gambling regulator predicts that reducing near-miss displays dampens play across loot boxes, retention loops, and prediction-error engagement.
• Safety across industries: a safety officer surfaces close-call data, weights it by causal proximity not vividness, across aviation, medicine, and nuclear.
• Negotiation: a narrowly-rejected offer functions as a partial-acceptance signal that anchors the next round.

Example

A slot machine is engineered to display near-misses far more often than chance, deliberately decoupling the proximity weight from the true near-zero counterfactual probability — so the player's re-engagement is not irrationality but a signal miscalibrated by design, dampened by removing near-miss displays.

Relationships to Other Primes

Parents (1) — more general patterns this builds on

• Counterfactual Proximity Weighting presupposes Counterfactual Reasoning — The SIGNAL-GENERATION mechanism that weights a nearby counterfactual into a reward/loss signal by distance; it presupposes counterfactual_reasoning to CONSTRUCT the alternative, then folds it in by proximity. The file: 'reasoning constructs the alternative; proximity weighting weights it into a signal'.

Path to root: Counterfactual Proximity Weighting → Counterfactual Reasoning

Not to Be Confused With

• Counterfactual Proximity Weighting is not Counterfactual Reasoning because reasoning constructs the alternative, whereas this weights an already-constructed alternative into a reward/loss signal by distance.
• Counterfactual Proximity Weighting is not Anchoring because anchoring is insufficient adjustment from an initial number, whereas this grades a signal by how near an alternative outcome was in a state space.
• Counterfactual Proximity Weighting is not Attention because attention is the selective allocation of processing, whereas this computes how an outcome's internal signal is generated, which may drive attention downstream.