The Simulated Other

What It Is

The simulated other is the mind's auto-completed rendering of another person — their answer, their verdict, their reaction — delivered internally with the felt certainty of a memory. Coherence is not evidence supplies the parent law: coherence is a property of the story, evidence is a property of the world, and a confident claim with no sample behind it is a rendering, not a read. This article is that law applied to people. When the missing evidence is another mind, the generator does not leave a blank where their answer should be. It renders them — fluently, in their voice, with plausible objections — and hands you the result formatted exactly like knowledge.

Will caught the mechanism live, in his own mouth:

"I have no idea what somebody else is going to say, and I do not know if they will pay — the only way to find out is to ask them, but I only know what they are GOING to say."

Both halves are true at once, which is the whole pathology. He has no idea what the other person will say, because he has never asked. And he knows exactly what they are going to say, because the simulation has already run. The certainty is real; its referent is the rendering.

Rendering vs Read

The simulation feels like knowledge because the brain is a generative model (predictive coding), and pointing it at another person produces a complete, detailed answer at zero cost. Nothing about the answer's felt quality marks it as generated — the same missing marker that coherence-is-not-evidence documents for factual claims. But a rendering and a read are different operations with different information content.

The load-bearing row is surprise. A rendering is sampled from your own priors, so it can only return things you already contain; it is structurally incapable of telling you something new about the other person. The one thing an ask can do that a simulation cannot is come back with an answer you did not have. That single bit — could this have surprised me? — is a provenance question: generated or retrieved?

The Tripwire Sentence

The mechanism has a verbal signature: "I already know what they're going to say." The sentence is worth installing as a tripwire because it is literally true about the rendering and unverifiable about the person, and the speaker cannot feel the difference. Whenever it fires — before a customer call, before an ask, before a hard conversation — it is announcing that the simulation has already run and its output has been cached as the other person's answer. The response is not to argue with the cached answer, because arguing with it is more simulation. The response is to notice its provenance: this was generated, not retrieved, and the certainty attached to it is the generator's fluency, not the world's testimony.

When the Rendering Absorbs Years

The most expensive documented run of the mechanism lasted nearly two years. A founder builds a product almost entirely inside his own head and the AI chat, and the central doubt — "isn't this just a wrapper?" — is raised and answered inside that loop over and over, always against a simulated customer whose objections he can render and whose satisfaction he can render too. Then the first real customer conversation of the whole period shatters on a single question: how is this different from an off-the-shelf chatbot with custom prompts? There is no answer, because the question has only ever been answered against renderings. The post-mortem's verdict: the whole edifice had been optimized against an imagined user.

The specimen shows the scaling behavior. The simulated other is not just a small daily error about queues and phone calls. Left unaudited it can absorb years, because every internal "answer" raises coherence, and coherence never terminates the loop — only a sample does. One read returned more information in an hour than a hundred renderings had in two years.

The Imagined Judge

The rendering also runs as an audience. The hostile judge who prices every ask as already-refused is usually not a memory of any real person. Will caught this one while watching his own EV sensor recalibrate after real exposure:

"I feel like my EV sensors are readjusting to feeling like, hey, if I do this, I'll be able to convince somebody that it's valuable — rather than them challenging me all the time. Because them challenging me was actually me myself challenging myself in my mind."

The hostile audience was internal. The sensor had been training for years on rehearsed rejections that never happened — renderings of the other, run as training data. This is the worst configuration of the mechanism, because the estimator being trained and the generator producing the labels are the same model. A system training on its own output converges on its priors with rising confidence, which is why the imagined judge only gets harsher with time and only softens after real contact. Reality's actual response rates are unaffected by any of it — as the sensor article puts it, reality doesn't care about your internal EV.

Contact Collapses the Rendering in Minutes

The rendering is not only unreliable; it is expensive. It re-runs every time the topic comes up, at whatever resolution the anxiety demands, and it has no stopping condition, because coherence can always be elaborated further. A read terminates in minutes and its output is reusable.

The small specimens make the asymmetry visible. A clinic visit mentally prepared for as a four-hour wait took five minutes ("It's a good lesson to follow reality contact"). The imagined boredom of a no-phone hour was worse than the hour itself. A multi-hop transit trip stayed intimidating as an imagined graph for months, and then one boarding ended it:

"The first traversal simultaneously completed the trip and learned the topology; the route is now a known reusable affordance rather than an abstract burden."

That last line generalizes past errands. The first ask, the first call, the first send does two jobs at once: it completes the task and it learns the topology — what asking this kind of person actually costs, how they actually respond, what the real distribution looks like. Every later traversal inherits that map. The rendering, run a hundred times, produces no map at all.

This is the same collapse reality contact documents as the red pill: the customer conversation that turns "they want an AI agent platform" into "we just need to automate these three repetitive tasks" in one meeting. The simulation absorbs months; the sample takes an hour.

The Ask Is the Instrument

For the questions that matter most — will they pay, will they say yes, will they show up — there is exactly one instrument that measures another mind, and it is the ask. Willingness-to-pay in particular cannot be simulated even in principle, because it is a fact about their constraints, their budget, and their alternatives, none of which your model contains at the resolution the answer needs. The ask also does double duty: every real answer is a labeled example for the EV sensor, an action→outcome pair the estimator can train on. A rendering is worse than unlabeled data — it is self-labeled.

The de-abstraction move belongs here too, because "the market" is the simulated other at its worst: an entire crowd rendered from priors, faceless, high-dimensional, and unaskable. A real person restores geometry — constraints, vocabulary, a budget band, a believable next conversation. It is never "someone like the first customer"; it is the first customer, by name. "Someone like them" is a rendering wearing a real person's name. The named person can be asked.

Simulation Debt

Anxiety owns the meter: anxiety is proportional to simulation time over reality-contact frequency. The simulated other gives that numerator a concrete population — accumulated un-sampled renderings. Every imagined conversation kept alive without an ask, every imagined queue, every imagined judge is an open loop the system keeps servicing, and the aggregate load is felt as the anxiety bar. The forager self-model names it directly:

"There's also a meter for hallucinations, because basically if you overuse your cognition without reality contact, there's like an anxiety bar… how you get rid of that is you close the gap between your thinking and reality."

The debt frame explains why the discharge is an action rather than an insight: no amount of further simulation reduces the numerator, because only reads close renderings. It also explains the anxiety article's self-fulfilling loop from the inside. The neutral face mid-conversation gets read as disapproval because perception in a sparse channel is rendering-heavy — even while looking at the actual person, the parts they are not saying get auto-completed by the same generator, from the same priors, with the same felt certainty.

The parent pathology — substituting a lower-anxiety proxy like whiteboarding, architecture, or AI dialogue for putting the thing in front of a stranger — runs on this mechanism. The proxy feels sufficient precisely because the rendering says you already know how the stranger would respond.

Failure Modes

Integration with the Mechanistic Framework

Connection to Coherence Is Not Evidence

The parent law. "Generated, not retrieved" is the general tripwire; this article is the case where the generated object is a person. The promotion gate applies unchanged: the rendering is a fine hypothesis about what they will say, and forbidden as a foundation until sampled.

Connection to Reality Contact

The practice. The red-pill collapse — simulation narratives dissolving on territory contact — is this collapse; the clinic wait, the dreaded errand, and the first traversal are further specimens of the same event.

Connection to Anxiety

The meter. Simulation debt is the anxiety formula read as an inventory: the numerator is the stock of renderings held open without a read, and the discharge is closing them.

Connection to EV Sensor Calibration

The corruption channel. The imagined judge trains the sensor on rejections that never happened; the ask is simultaneously the read and the labeled example that retrains it.

Connection to Provenance

The audit that catches it. "I already know what they're going to say" is a provenance flag on a social belief: the answer on file was generated by you, not retrieved from them.

See Also

• Coherence Is Not Evidence — the parent law: renderings promoted to beliefs without samples
• Reality Contact — the red pill: contact collapses the simulation
• Reality Contact Metabolism — the physiology of a mind running rendering-heavy
• Anxiety — the meter: simulation time over contact frequency
• EV Sensor Calibration — the imagined judge as sensor corruption; the ask as training data
• Provenance — generated or retrieved, applied to any value
• Predictive Coding — the generative substrate that renders by default

Core Principle: The mind auto-completes the other person, and the felt certainty of the rendering substitutes for the ask. A rendering is drawn from your own priors, so it cannot surprise you — the one thing a read can do that a simulation cannot is return an answer you did not already contain. The tripwire is the sentence "I already know what they're going to say": literally true of the rendering, unverifiable about the person. Contact collapses in minutes what simulation sustains for months, the first traversal learns the topology as a reusable map, and the only instrument that measures another mind — willingness to pay above all — is the ask itself.

The other person is one ask away. The simulation of them is infinitely elaborate, always available, and made entirely of you.