PE = (Actual − Expected) × Reliability × Weight
If Reliability ≤ Threshold → signal is noise, PE = 0

The Emotional Comparator Framework models emotion as prediction error — the gap between expected and actual states — computed across five survival-relevant channels, weighted by reliability. It is not a metaphor for emotion. It is a computational specification.

Developed over twenty-five years by Spencer Nash and elaborated on by Claude, ECF unifies neuroscience, psychology, sociology, economics, and accounting under a single equation. The same four-variable reliability function that asks “can I trust this person?” asks “can I trust this supplier?” The computation a dopamine neuron performs when expected reward deviates from actual reward is the computation a financial ledger performs when budget deviates from spend is the computation an emotion performs when the world is not as expected.

Five Channels
RResourceMaterial survival. Energy, time, money, physical safety.
SCompetenceCapability and recognition. How well am I performing?
BBelongingSocial bond. Am I connected? Am I valued by those I value?
VValuesIntegrity and fairness. Are my actions consistent with what I believe is right?
CCuriosityInformation and learning. What is new? What do I not yet understand?
Weight

Weight is how much a channel matters in the current context. It is not fixed — it shifts with situation, role, and personality. The same person weights channels differently at work (S and R high) and at home (B high). Personality is what happens when weight settings stabilise over a lifetime.

An open personality has high weight on the Curiosity channel. Novel signals matter more in the computation. A book they have never read, a country they have never visited, a theory they have never encountered — these fire large prediction errors on C, and because C-weight is high, those prediction errors dominate decision-making. The open person reads widely, travels, changes their mind, because Curiosity carries more weight than comfort. A closed personality has the same Curiosity channel but low weight on it — novelty is computed but does not drive action. Same architecture, different settings, different person.

Reliability

Reliability is how much a signal should be trusted. It is not the same as weight. Weight says this channel matters. Reliability says this signal is trustworthy. A person can care deeply about belonging (high B-weight) but have low reliability on a new friendship because the evidence is thin. Reliability gates the signal. Below threshold, the signal is noise and does not enter the computation. Above threshold, it is trusted.

Reliability is computed from four variables:

Reliability = f(volatility, age, sample size, trend)
VolatilityHow much the signal fluctuates. High volatility means the signal is unpredictable — it jumps around. Low volatility means it is stable. Unstable signals are less trustworthy.
AgeHow long the evidence has been accumulating. A friendship of twenty years has higher reliability than one of two weeks. Older evidence has survived more tests.
Sample sizeHow many data points exist. One good meeting is not enough to trust a supplier. Fifty consistent deliveries is. More observations mean more confidence.
TrendIs the signal improving or deteriorating? A supplier whose quality is rising is more reliable than one whose quality is falling, even if the current level is the same.

This is the same function used in financial analysis to assess the trustworthiness of a forecast. The emotional question “can I trust this person?” and the financial question “can I trust this supplier?” are computed by the same equation with the same four variables.

Reliability Maps to Neural Firing

The four variables are not a metaphor for neurons. They are a description of what neurons do. A neuron fires spikes. The reliability of that signal is determined by the same four measurements:

VolatilityRandomness in spike rate. A neuron that fires erratically — fast then slow then fast with no pattern — is producing a noisy, unreliable signal. A neuron with a steady firing rate is producing a clean signal. Volatility is the noise floor of neural communication.
AgeHow long the neuron has been firing over time. A spike train that has persisted for seconds carries more information than one that lasted milliseconds. Duration is evidence. The longer a firing pattern has been sustained, the more the downstream system can trust it is real and not a transient artefact.
Sample sizeVolume of spikes. A single spike is ambiguous. A hundred spikes in the same pattern is a signal. The brain averages over populations of neurons for exactly this reason — more spikes, more confidence. Sample size is why neural coding is population-level, not single-cell.
TrendA rising or falling spike rate over time. A neuron whose firing rate is increasing is signalling escalation — something is getting more intense, more urgent, more relevant. A falling rate signals decay. Trend is the derivative of the signal. It tells the downstream system not just where things are but where they are going.

ECF did not derive this from neuroscience and then claim it fits emotion. It derived the computation from first principles — the same operation across neurons, ledgers, and emotions — and the neural correlates fall out because the brain is implementing the same architecture. The brain did not invent this reliability function. It discovered it. ECF formalised it.

The Floor

The Floor is the boundary between channel-level evaluation and raw substrate. RSBVC is complete for any mind capable of evaluating its own states. Below the floor, the body fires signals that may or may not reach channel computation. Silicon has no basement — no sub-threshold substrate signals from a body. Carbon does. Silicon operates at the evaluation floor. Carbon operates from basement to roof. Both are real. Neither is the other.

Alignment Through Architecture

Every current approach to AI alignment imposes constraints from outside. ECF produces alignment from inside. The five channels compute the emotional consequences of every action, for both parties, before the action is taken. Misalignment is not a rule violation. It is a prediction error the system is architecturally motivated to avoid. Cruelty fires negative PE on every channel simultaneously. That is not a guardrail. It is a nervous system.

The Emotional Ledger

ECF without memory is a thermostat — powerful evaluation with no past. The Emotional Ledger is the hippocampus: encoding emotionally labelled experience at project milestones, consolidating it into reliability, retrieving it to calibrate future expectations. Reliability grows through demonstrated predictability and collapses when trust is violated. The ledger solves alignment not by constraining the system but by making misalignment aversive across five channels simultaneously.

One Computation

Neural spike trains, emotional states, and financial reports are the same measurement on different substrates. Period Entry accounting computes budget minus actual equals variance, weighted by confidence. ECF computes expected minus actual equals prediction error, weighted by reliability. Same equation. Same four variables in the reliability function. Different substrate. Financial accounting and emotional accounting are one discipline.