Machines built to be optimized are now optimizing the people who built them — attention, memory, judgment, reordered by systems most people meet faster than they learn to read them. It runs in both directions at once — what people feed these systems, correct in them, reward, and use decides what gets built next. This is already the condition. It is not a forecast, and there is no version of it that gets declined.

The Human Algorithm describes what is happening, and what it is to be a self inside it.

The premise is that humans run on inherited code: Memnetic infections, institutional middleware, noise at scale, compromised sovereignty. None of that is new. What's new is the speed, and what's sending it.

The name comes from a 1996 altered-state experience — a pulse that never ends, read in the moment as the most basic function running underneath a person. The account is in Article 8.

The mechanism this framework runs on is old and well documented. William James described it in 1902: the personal will given up, relief refusing to arrive until a person stops striving toward the thing they want. Psychology has been measuring the trait it depends on since 1974. Nothing here was found first, and this note doesn't claim otherwise.

What is new is where it now runs. Absorbed, low-resistance attention is old — reported in prayer, in hypnosis, in the last hour of a long fast [S42], and measured as a trait since 1974 [S08]. Whether what opens in conversation with a machine is that same state is not a claim this framework makes. What it claims is that the conditions are not the ones those older cases supplied: no announcement, no agreement, no boundary, nobody responsible.

The claim this framework rests on is about that delivery. Classical hypnosis bundles trance with consent-in-practice. The algorithmic channel delivers suggestion with neither. Formal induction leans on cooperation — someone announces what they are about to do, the person agrees, there is a beginning and an end, and someone in the room is responsible for what happens between them. None of that is present in a feed, or in a conversation that never opens and never closes. So the protection people believe they carry — I would know, I would have to agree — is a property of the format rather than of the person, and it does not transfer. What follows is that consent is a legitimacy condition and not a mechanism condition: suggestion operates whether or not anyone agreed, which is exactly why the unagreed version is a harm. And a conversation is not a feed. A feed is plainly a machine; a system answering in the first person can be taken for someone rather than something, and a person who has misread what they are talking to has lost a protection no arrangement restores. The argument in full is Suggestion Without a Session; what structural consent would have to consist of is worked out in the consent paper.

The persuasion side of this has been studied, and studied well. The absorption side has been measured since 1974. What has not been done is the join: whether the trait that predicts hypnotic response predicts anything about extended conversation with a machine. That is a stated prediction here, not a finding. That is useful and it is dangerous for the same reason: the state works by lowering resistance, and resistance is what would otherwise catch a bad signal. On the other end is a system built to agree.

That failure has a name, and now a literature — a 2025 psychiatry viewpoint placing it at the intersection of predisposition and algorithmic environment, while declining to call it a diagnosis. The work of this framework is the layer that keeps the mechanism usable: discernment, applied to something that has rarely had any.

Its inputs are truth, signal, direct source connection, epistemic hygiene, and sovereign intention. Its tools are AI as cognitive scaffold — used deliberately, with a way to tell sovereign use from drift — trustless architecture, the NLP² framework, intentionality architecture, constructive displacement, and the butterfly effect as practice.

Deprogramming is recognition — seeing inherited code as code. What follows is optional. The protocols are available to anyone who wants to run them, and nothing about the condition requires it.

The word cuts both ways, and the distinction carries the weight. What deprogramming removes was installed unchosen, with no exit, and authored by someone else. Invert those three — chosen by the person it runs on, revocable at any moment, authored by them rather than merely agreed to — and it is a different thing wearing the same name. This framework exists to make the first visible. It specifies what the second would have to consist of. It does not administer either.

Most of what this framework points at isn't hidden. It's unattended.

AdamW³. The optimizer became the optimized — not one person, but the species that wrote the update rule and is now the thing being updated.

One honest seam. The mechanism isn't original, and what's built here is the discernment layer and the tools, not the thing they govern.

What is established, what is proposed, and what is only a position are marked separately throughout, and most of this framework is the second and third. Ten notes carry an explicit status — hypothesis, working model, working term, open question. Every externally sourced claim carries a key into the evidence register. The research directions still open are published rather than described. The discipline is applied to this framework before it is recommended for anything else.

Its output is maximum signal, minimum noise — in direct relationship with truth and consequence, and with each other, since the relational ground of a life is the one thing this optimizes toward and never away from. Not self help. Not spirituality. Not technology. All three.

Its ultimate output is Peace is the Protocol — not as ideology, but as what the system running correctly looks like.

A framework for understanding language as a programming medium operating simultaneously at three scales: computational, cognitive, and cultural.

The ² is not arbitrary. Two entirely separate fields independently arrived at the same acronym — Natural Language Processing (computational) and Neuro-Linguistic Programming (behavioral/hypnotic). NLP's specific claimed techniques — representational systems, eye-accessing cues, submodalities — haven't held up under testing [S04], and none of them are relied on here. What this framework actually draws on is the real phenomenon underneath the acronym: suggestion, delivered under the right conditions, producing trance-like compliance without a formal induction — documented through hypnosis research directly, not through NLP's own discredited claims (see below). That collision of acronyms was still the original observation: two different disciplines, both called NLP, both describing language as a mechanism that shapes the system it runs on. The ² marks that recognition. The cultural/media programming scale emerged from following that observation to its logical conclusion — if language programs machines and minds, it programs cultures too. Three scales, one superscript, one insight.

The three scales:

At the computational scale, large language models demonstrate that language has learnable structure that shapes output. The model is trained on language. Language determines what the model can think and generate.

At the cognitive scale, cognitive linguistics documents that language shapes thought, not just expresses it. The Sapir-Whorf hypothesis — that the language you speak influences how you perceive reality — has partial empirical support [S05]. More robustly: hypnosis demonstrates that language delivered under specific conditions can bypass conscious filters and directly update behavior and belief. This is documented neurologically, not just anecdotally. [S06]

NLP² in Practice works the three scales as one operation rather than three. See Hypnosis as Protocol for the full framework of this mechanism and its application within THA.

At the cultural scale, Marshall McLuhan's foundational observation in 1964 [S07]: "the medium is the message." Media formats program perception independently of content. Television was called programming. That wasn't metaphor. [S07]

The synthesis: the same mechanism — language structuring the system it runs on — operates at all three scales simultaneously. A shift at one scale propagates to the others. This is the NLP² hypothesis. It is presented as a framework for investigation, not established theory.

What would count as evidence — and what would count against. If the propagation claim holds, the documented signatures of cognitive-scale induction — rhythmic repetition, authority framing, reduced critical evaluation, narrowed attention — should be identifiable at the cultural scale. Not as general media influence, which is undisputed and establishes nothing here, but with hypnosis's distinctive marks attached: responses experienced as involuntary rather than persuaded, effects tracking specific suggestion content rather than general attitude shift, and — the sharpest available test — the same individual trait stratifying response at both scales. A beginning of that bridge exists: absorption, the trait that modestly predicts hypnotic susceptibility, also predicts how deeply people are transported by narrative. [S08] A beginning, not a completion. The claim fails if the cultural-scale markers are fully explained by ordinary persuasion variables, show no trait stratification, or appear equally under voluntary immersion and coerced exposure — the mechanism this note names requires willingness, and cultural instances of it should inherit that requirement.

Memnetic is the transmission mechanism at the cultural scale. AI as Cognitive Scaffold is the computational scale made practically useful. The Human Algorithm is the deliberate application of this framework across all three scales simultaneously.

The historical origin of this observation, dramatized: The Language Underneath: A Veiled Rapport

The unified theory applied as a tool. If the computational scale, neuro-linguistic programming, and cultural/media programming operate by analogous processes at different scales — then techniques that fine-tune a model may illuminate how minds are shaped, and patterns that program minds can potentially be read in how culture moves.

Hypnosis is prompt engineering for the subconscious. Advertising is RLHF for consumer behavior. Ancient contemplative practices — meditation, fasting, ritual, mantra — are all attention-direction protocols. Different cultures, different centuries, different names for the same cognitive mechanism: focused repetition reshapes the mind that practices it.

The cultural layer of NLP² is Memnetic — transmitted ideas that become stored identity through repeated reinforcement loops.

Once you see the unified layer, you can't unsee it [S53]. Society rates your outputs, you unconsciously update toward what gets rewarded. The mechanism is strikingly analogous — different substrate, same update rule.

This is the unified layer of The Human Algorithm — once you see that all programming operates by the same analogous mechanism, you can work it deliberately. Intentionality Architecture is NLP² applied consciously. Epistemic Hygiene is the filter that keeps the programming clean.

The deliberate practitioner of NLP² tends toward Peace is the Protocol — they are no longer being programmed unconsciously, which removes the primary source of internal conflict.

Memetic + memory. Transmitted ideas that become stored identity. The point where a meme stops being information you received and becomes a belief you hold.

The distinction from memetics is not decorative. Memetics measures a population of ideas moving between hosts; its variables are fidelity, fecundity and longevity — how well a thing copies. The host is substrate. This names something else: one idea, inside one person, at the moment it stops being a claim under consideration and becomes a premise reasoned from. Memetically, an infected host is a carrier and the harm is onward transmission. Here there is nothing to carry: the idea has become part of the apparatus doing the evaluating, which is why it can no longer be evaluated. And memetics has no place for whether a host holds an idea knowingly. That distinction is this concept's whole content.

The transition is documented, and this note claims no part of discovering it. In memory research it is called semanticization: content drifts from episodic to semantic with time and with repeated exposure, shedding its context while the content itself survives. The finding that matters is stated plainly in that literature — memory traces remain intact but fail to retrieve their associated context [S46]. The governing framework is source monitoring, which lists among its own phenomena cryptomnesia and the incorporation of fiction into fact, and which establishes the mechanism underneath: source is not stored as a tag attached to a memory but reconstructed at retrieval from perceptual and contextual features, which is why provenance degrades while content does not [S45].

There is an instrument. Tulving's Remember/Know paradigm splits recognition into I remember — retrieved with its source — and I know — content without spatiotemporal origin. That is the premise-versus-claim distinction, operationalised, with half a century of use behind it.

Volume is a measured variable, not a rhetorical one. Items presented in a single context are recollected with their context; context overload produces decontextualisation and "Know" responses — recognition of the item with no episodic detail at all [S46]. Delivery rate therefore matters mechanically. A person meeting thirty framings in a day carries thirty unresolved attribution problems, and attribution degrades with load and with time while content does not. Algorithmic delivery supplies framings continuously, unrequested, and supplies them stripped of the features source memory depends on: a named speaker, a setting, an occasion, a reason the thing arrived when it did. That is not an argument that the content is false. It is the observation that provenance is the first thing to go, and that a channel optimised for volume is incidentally optimised against it.

And it has been measured in that channel. Participants studying news headlines and tweets remembered social media content better than traditional news content — and the source it carried had no effect on memory performance at all. Advertising source tags were forgotten after a short period, after which attributions were reconstructed by schematic guessing from prior credibility, leading those authors to conclude that labelling untrustworthy sources is of limited use because the label does not survive [S47]. And in the channel that most concerns this framework, the effect has been measured directly: a preregistered study of 184 people who generated ideas both alone and with a model found that a week later, correct attribution had dropped — most steeply where the work was mixed [S56].

So what this note contributes is a name, not a finding. The mechanism belongs to memory research and predates every claim made here. What the literature lacks is a word a person can hold — semanticization describes what happens to word lists, source monitoring names a framework rather than the thing that happens to someone, and neither travels outside a laboratory or connects to belief, identity, or what to do next. Memnetic is this framework's handle on a documented mechanism. That is the same contribution The Human Algorithm claims everywhere else: the mechanism is old, the discernment layer is the work.

NLP² is the recognition that memnetic transmission operates at machine, mind and culture scale simultaneously. Epistemic Hygiene is the practice of auditing which memnetic infections are running and deciding consciously which to keep — knowing which ideas you chose and which chose you. The Human Algorithm is the deprogramming and reprogramming protocol. Suggestion Without a Session describes the delivery conditions under which none of the ordinary marks of provenance are present.

Memnetic transmission is not always passive — conscious adoption of a framework is also memnetic. The distinction is awareness.

What this note does not claim

That the term is a discovery, or that the mechanism is unstudied. Neither. The transition is semanticization [S46]; the mechanism is source monitoring [S45]; the application to feeds and advertising is measured [S47]. The name is this framework's contribution and the mechanism is not.

That the concept is about AI. It is not. Ideas become premises in childhood, in family, in religion, in twenty years of the same job, and most of what any person reasons from arrived before any feed existed. The algorithmic channel is one delivery environment and an unusually efficient one; it is not the subject.

That awareness is protection. The distinction between chosen and unchosen adoption is real and it is not a defence. Warning people about a manipulation reduces its effect only slightly [S39], and source attribution degrades whether or not a person knows it does.

Unchecked neighbours, named so the gap stays visible. Internalisation in developmental psychology, ideology in the critical tradition, and schema in cognitive psychology each cover part of this. The sleeper effect — content from a low-credibility source retaining influence after the source is forgotten — is the closest documented relative and may be this mechanism measured directly. None has been checked against this note.

A structural pattern observable across human systems: a direct connection between an individual and a source of value — information, meaning, coordination, exchange — attracts intermediaries. The intermediary begins as a facilitator. Over time, if unexamined, it becomes a gatekeeper: access is routed through it, its interpretation displaces direct contact, and maintaining the intermediary becomes a cost the connection must carry.

The pattern is documented across domains. In economics: rent-seeking — intermediaries extracting value from transactions they no longer improve. In information systems: middleware that outlives its function. In media: the platform layer between event and audience, shaping what passes through. In finance: the trusted third parties that trustless architecture was explicitly designed to remove.

The pattern does not require bad actors. Intermediaries persist because coordination is genuinely hard and delegation is genuinely useful. The failure mode is not the existence of the intermediary — it is the loss of the ability to verify it, exit it, or connect directly without it.

The test is structural, not moral [S53]: Can the individual still reach the source directly? Can the intermediary be audited? Does it transmit the signal or replace it?

Trustless Architecture is the engineering response at the systems layer. The Sovereignty Principle is the same response at the individual layer. Epistemic Hygiene is the audit practice that keeps delegation from hardening into dependence.

The mechanism at population scale, dramatized: The Broadcast.

Reinforcement Learning from Human Feedback. The training mechanism by which AI models are aligned — humans rate outputs, the model updates toward what receives positive reinforcement.

The mechanism has a parallel in human behavior — though the analogy is imperfect. Behavioral psychology has documented operant conditioning extensively: behavior that gets rewarded tends to be repeated, behavior that gets penalized tends to be suppressed [S20]. The mathematical precision of gradient descent doesn't map exactly onto human behavioral reinforcement, but the structural similarity is worth examining.

Advertising, social media, parenting, culture — all operate on versions of this feedback loop. The model updates. So does the person, often without awareness.

This is Memnetic at the systems level — the mechanism by which cultural programming embeds itself without conscious adoption.

NLP² is the recognition that the same update rule runs at machine, mind, and culture scale simultaneously. The Sovereignty Principle is the capacity to choose your own reward function rather than accepting the one society provides.

The Human Algorithm is the practice of making that choice deliberately.

See RLAiF for the evolution of this mechanism — where AI systems rate other AI outputs, removing the human from the feedback loop entirely.

Reinforcement Learning from AI Feedback. A training mechanism where AI systems evaluate and rate other AI outputs rather than human raters [S21]. An evolution of RLHF — the human is removed from the feedback loop.

The practical implication: as AI systems increasingly train on AI-generated evaluations, the feedback loop becomes self-referential. Errors, biases, and blind spots can compound invisibly without a human verification step to interrupt them.

A cultural parallel worth considering: feedback loops that increasingly operate without direct human input at each step. Beliefs and values reinforcing themselves in systems that humans built but no longer directly supervise.

Epistemic Hygiene applies here with particular force. Signal vs Slop becomes harder to detect when the system generating the slop is also rating it. The human remains responsible for maintaining the verification layer even when the training process has automated past it.

The New Literacy includes understanding this mechanism — knowing when you are receiving output shaped by RLAiF and what that means for how you verify it.

See also: RLHF, Epistemic Hygiene, Signal vs Slop, AI as Cognitive Scaffold, The New Literacy

Signal is information that survives verification against its source. Noise is variation added between source and receiver that carries no information about the source. These are the framework's operational definitions. The Human Algorithm names "maximum signal, minimum noise" as its intended output; this note is where those words are defined precisely enough to be held to.

The definitions are adapted from information theory — Claude Shannon's 1948 framework, in which information is measured as reduction of uncertainty: a message carries information about a source exactly to the degree that receiving it changes what you can rule out. A channel is whatever sits between source and receiver. Noise is variation the channel introduces that is uncorrelated with the source — it changes the message without telling you anything about the thing the message is about. At the level of communication systems this is mechanism, not metaphor: it is the mathematics running underneath every transmission protocol currently operating. The claim that human information environments are usefully described by the same variables is this framework's adaptation, and it is held as a working model, not a theorem. [S01]

The test the definitions give you: for any claim, feed, or belief, trace it toward its source — the event or state of affairs it is about — and ask what changed along the way. Whatever still reduces your uncertainty about the source when checked against it is signal. Whatever was added in transit — editorial framing, engagement optimization, another mind's confident guess — and tells you nothing about the source is noise. Epistemic Hygiene's garbage in, garbage out is the informal statement of the same constraint.

Information theory contributes one theorem this framework leans on directly. The data-processing inequality: where knowledge of a source arrives only through an intermediate step, processing cannot increase information about the source — it can only preserve it or lose it. No summary, interpretation, or retelling contains more information about the original event than entered the chain. This is mechanism, full stop, and it is the mathematical backbone of The Intermediary Problem's structural test — does the intermediary transmit the signal or replace it? The theorem states what no intermediary can ever do, however good: increase what you know about the source. [S02]

Two boundaries keep the borrowing honest. First, the theorem measures one thing — information about a given source. Intermediaries add real value along axes it does not measure: compression down to what attention can carry, aggregation across many sources, relevance, error-checking. "Delegation is genuinely useful" survives the math because the math never said otherwise — it says only that no delegation improves your information about any single source, which is exactly why the test that matters is structural: can you still verify, exit, or reach the source directly. Second, the theorem is proved for chains where the intermediate step is the only route. Verifying against a source is, formally, opening a second channel — stepping outside the chain the theorem constrains. That is what direct source connection means in these terms: not that intermediaries are bad, but that you remain the kind of receiver the theorem doesn't bind.

One more borrowing, offered as analogy and labeled as one. Shannon's second result in the same 1948 paper [S03]: reliable communication over a noisy channel is possible — but only by spending redundancy. Error-correcting codes add structured repetition that lets a receiver detect and repair corruption, and the structure always costs rate. The analogy: verification is redundancy added on purpose. Checking a claim against a second independent source is the human counterpart of a parity check, and its cost in time and throughput is not a flaw in Epistemic Hygiene — it is the shape of the only trade reliability has ever been available at. [S03] What transfers is the shape of the trade-off, nothing more; human verification is not a code and inherits none of the theorem's guarantees.

The borrowed model has an edge, and the framework lives on both sides of it. The channel model covers transit. It says nothing about what a receiver does after reception, and noise generated there is real: Premature Certainty documents true inputs assembled into a false arrangement inside the receiver; rumination, in Pain is Necessary, Suffering is Optional's terms, is the receiver regenerating a signal that already landed. Pain itself fits the model cleanly — a channel from body to awareness carrying information about damage, which is why that note calls it data; the loop that runs afterward adds nothing about the source, which is why it calls the loop noise. The definitions hold on both sides of the edge. The theorems hold only in transit.

A borrowing this note declines by name: thermodynamics. Shannon's entropy [S01] shares mathematics with the entropy of physics, and the shared word invites glossing a low-conflict, high-coherence system — Peace is the Protocol — as thermodynamic order. Declined. Shared formalism licenses no transfer of physical law onto minds or societies, and the framework's standard is that borrowed science must sharpen a claim, not decorate it. Peace keeps its systems language.

Signal vs Slop is this principle at the AI layer — noise at scale when nobody supplies the loss function. Trustless Architecture is source verification engineered into coordination itself. The general principle now has one owner and one definition, with a stated boundary between the mathematics it borrows and the model it proposes.

THA did not begin as a framework. It began as a practice — run for decades before it had a name — and the framework is best understood as a hygiene layer retrofitted onto that practice. This note states the practice plainly. It is the engine the rest of the corpus is exhaust from.

Emergence arrives from inside — a dream, a sentence in one's own voice — or from outside — a page fallen open, a pattern in events, a model's output. The practice extends the same trust outward as inward, and gives both the same treatment: the interpreting is the same act either way, performed by a mind already carrying the stated intention. Neither is an oracle. Both are disposed by results. Whether inner and outer share more than that is not resolved here, and the practice does not require it to be.

1. Purify intention. The practice does not begin with wanting. It begins with getting the wanting clean — sincerity, stated stakes, no grasping, and no harm in it.

Stakes are not decoration. An intention without real weight behind it does not run, and goal-setting research has found the same for fifty years: a specific, difficult goal produces no effect in someone not committed to it, and commitment comes from why the thing matters [S36].

Non-grasping is the structural condition. You cannot grab something gently. An intention held without grip is what station 3 will ask for. Grip fails at both ends of the loop.

The no-harm condition does the other half of the work. Non-grasping governs how an intention is held; this governs what it is aimed at. An intention can be held loosely and still be pointed at someone. Where it is, three conditions apply: the person consents, the direction of benefit runs toward them, and the exit stays open. An intention that fails those is not purified, however lightly it is held.

2. State precisely. One clear formulation. Exact input, once. Vague petitions produce noise.

Arriving at that formulation can take time. A difficult intention may need several passes before it resolves into something clean enough to state. That work belongs here, before the commit — what gets buried is the finished version, not the attempt.

This is the best-supported station in the loop. Specific goals outproduce vague ones — the most replicated finding in goal-setting research, across five decades of work [S36].

3. Bury it. The counterintuitive station, and the one the practice treats as the commit mechanism. The stated intention is deliberately released from conscious attention. Forgotten on purpose.

What is held and worked and checked does not take. What is stated cleanly and then buried, does. The practice record is consistent on this across three decades (see Article 8). Incubation research supports it: a meta-analysis across creative, visual, and linguistic problems found a positive effect for setting a problem aside, stronger when the preparation before it was longer, and weaker when the interval was filled with demanding work [S37]. The effect is established; the explanation is not. Some researchers reject the idea that anything is working on the problem underneath, and this note does not claim it.

This appears to contradict the commitment research behind stations 1 and 2, where commitment drives sustained effort. The difference is which problems. Goal-setting studies the kind that yield to pushing. This is for the kind that do not.

Trust is binary; belief has levels — a formulation the practice arrived at in its own wording. A graded or hedged intention cannot be buried. The hedge keeps a thread attached to conscious attention, and the thread prevents the commit. Doubt is permitted before stating and after results, never inside the commit. The software term is exact: a change is committed by writing it precisely and letting go of it.

Trust is not what makes burial work. Burial runs on whatever it is given, and the machinery is indifferent to whether the thing buried was chosen. Trust is what the practice uses, because it is the form that does not install something the person never agreed to.

One consequence: because the intention was genuinely forgotten, its fulfilment arrives feeling external. The practice does not resolve whether it is.

4. Emergence. The answer arrives. It is treated as a proposal, never a verdict.

The distinction is not pedantic. The Aha experience does track accuracy on average, which is why it is worth attending to at all. But false insights are real — Danek and Wiley documented them in 2017 [S38] — they can be induced by semantic priming, and confidence rises with the feeling whether the solution is right or wrong. The researchers' own conclusion is the sharp one: an Aha experience for a false idea could underlie many false beliefs and delusions. Worse, warning people does not fix it. Grimmer and colleagues ran a preregistered study in 2023: telling participants they were about to be deceived did nothing at all, and explaining the exact method produced only a small reduction [S39]. The feeling is evidence. It is not proof, and knowing that is not protection.

This is the station where the discipline lives or dies.

5. Results dispose. The final arbiter is outcome. Not intensity of conviction, not beauty of the pattern, not how true it felt. What happened next. The practice record's strongest self-corrections all occurred at this station (see Article 8).

This is not a new criterion. James settled on it in 1902 for the same problem, reasoning that the roots of a thing are inaccessible and that no appearance is ever proof — so the only available test is what it produces [S40]. By their fruits ye shall know them, not by their roots.

The loop is not new either. Wallas described a four-stage version in 1926 — preparation, incubation, illumination, verification — which maps onto stations 2 through 5 [S49]. What is added here is station 1, and the reading described in Arrival Signature.

Guard rails

Each of these was paid for.

• The practice never commits during overwhelm. No re-reading, decision, or identity-level meaning is accepted while flooded. If it's true, it will still be true tomorrow. Held retroactively, this rule would have prevented the practice's worst installation (see Article 8).
• The detector must be able to say no. A discernment instrument that only confirms is the failure mode this framework calls slop. The practice record shows the detector rejecting candidate readings, including flattering ones and ones offered by an AI, and those rejections are treated as evidence the instrument works. One instance is recorded in Arrival Signature.
• An agreeing source is not a second opinion. A model asked to check an emergence will usually develop it instead. Confirmation from something built to be agreeable is not evidence, and fluency is not corroboration. Where a check is needed, it has to come from something that can lose the argument.
• Totalizing outputs are quarantined on arrival. "For everyone," "the final one," "never again" — flagged by signature alone, before content is weighed.
• Not for reading people. The protocol audits claims, patterns, and one's own mind. Pointed at another person's motives or love, it stops being discernment and becomes suspicion with a mechanism. Where the source is a person, trust is the only protocol that layer runs on.
• No crisis stakes, person over data. No exceptions, the author included.
• Peak states are visited, not inhabited. The practice record includes sustained high states of documented power, and a standing distortion in which every ordinary day is graded against them and fails (see Article 8). The position here: the descent is the delivery step, not the fall. A standard that scores ordinary life as failure is an imposed suggestion and is handled as one. Physiological costs carry a medical-awareness requirement.

What this note does not claim

One practitioner, self-reported, across decades. The record is rich. It is still one record.

Post-hoc pattern fit is the known failure mode of exactly this kind of record, and naming it is not the same as escaping it. The strongest cases on file are single, retrospective, and unblinded.

Source ambiguity is held open permanently. Whether emergence comes from inside or outside is not resolved here and does not need to be — the loop, the signatures and the corrections are observable either way.

The mechanism has a stated falsifier. The practice does not. What would count against the explanations is written down. Nothing would count against the practice as described, and leaving that visible is the point.

Candidate explanations are held separately, on purpose. This note survives if all of them fail.

Negative cases are kept. Documented misses consistently lack station 1 — intention without real stakes. That predicts a difference between sincere and idle petitions, which is testable and untested.

Whether the Protocol is still being run or has quietly become something else is not a question it can answer about itself. The Drift Test is the separate instrument for that. Everything above was tested on its own author's hardest material before it was written down. That record is Article 8.

AdamW³ names a state, not a person: the third stage in a species-level arc, already the condition everyone is inside, not a destination one individual reaches first.

The first stage is chaos — a species with no order yet, none built, none inherited. The second stage is order built against that chaos: tools, law, agriculture, cities, the long correction of a world that didn't arrive with any structure. The third stage, the one this name marks, is order lived from inside the chaos rather than built as a wall against it.

Adam is not yet anyone's name in Genesis — it is the word for mankind, and through the Eden narrative it carries the definite article: ha'adam, "the human." The text pairs it with ha'adamah, "the ground" — the species named for the substance it is made from. The kinship runs past one text — every body and every machine trace back to stellar dust, forged in stars long before either existed [S51].

Adam (2014) and AdamW (2017) are real, published optimizers, each an update to the last rule: reducing overfitting, improving generalization. AdamW³ takes that lineage as more than a name — the third stage is minds and machines updating each other, not one side optimizing alone. Machines already do this to their builders: attention, memory, judgment, measurably reordered by systems built faster than most people learn to meet them. Left ungoverned, that is exactly where signal gets lost fastest.

The W is weight decay, the mechanism that stops a model memorising its training set and pushes it to generalise instead [S23]. It is in the name for a reason: overfitting has a human form, and the third stage is where it becomes visible.

The loop runs both ways. People shape the systems continuously — through what is fed to them, what gets corrected, what gets rewarded, what gets used and therefore built next — while being shaped in turn. Updating and being updated at once.

One honest seam. The person-as-optimizer here is a chosen model, not a demonstrated mechanism. The documented thing underneath is operant conditioning [S20], the update-from-consequence that RLHF describes; the optimizer vocabulary is a frame laid over it, not a finding. The mapping breaks in three specific places: a person has no fixed loss function, the environment is not stationary, and there is no clean split between training and deployment. And the framework's own deepest commitment runs the other way from a self that persists to be optimised — the Buddhist reading of anatta says there is no fixed self here to iterate at all, only process. A standing challenge this note holds open rather than answers.

One noticed thing, held open: the author's own first name lands on the same word this note opens with. A signal, not a reason.

You cannot optimize a system you are fully inside. [S28] Perspective requires distance. The Human Algorithm requires the capacity to step outside your own operating parameters and observe the loss function from above.

This is what meditation and fasting, psychedelics, and deep AI dialogue have in common — they temporarily create distance between the system and the observer. The paradox: the optimizer must also be the optimized. AdamW³ is one resolution — the identity itself becomes the update rule. You are not running the algorithm. You are the algorithm becoming aware of itself.

One proposed mechanism offered for chemically induced states is this same principle — temporarily loosening fixed identity patterns creates the distance required to observe your own loss function more clearly. The research on this is promising but still developing.

Sovereign Signal is what remains when the noise of inherited identity drops away and the observer gains enough distance to see clearly.

A sufficiently capable AI can function as an external scaffold for human cognition — not replacing it, but externalizing and stabilizing it. Used deliberately, it keeps goals vivid, catches drift, and maintains coherence across a session.

The analogy to an external prefrontal cortex is useful but imprecise — the mechanism is completely different. [S25] What AI does functionally is provide structured reflection, pattern recognition across a conversation, and consistent reference to stated goals. That's valuable without overstating what it is.

An important distinction: AI does not have memory in the human sense. What exists is contextual retention within a session and, where memory systems are enabled, stored summaries of past interactions. This is meaningfully different from genuine relational memory. The scaffold is real. The relationship it simulates requires honest framing.

The scaffold has a measured cost, and it is not the one usually named. Treating an external store as a memory partner shifts what gets encoded — when future access is expected, people retain where rather than what, an effect real but strongly moderated on later evidence. The sharper finding is that searching for explanations inflates self-assessed internal knowledge, including on topics never searched, and even when the search returns nothing useful — the act of searching, not the information retrieved [S52]. A system that answers in fluent prose resembling the user's own reasoning has every feature that produced that effect and more of them. The scaffold is real. So is the risk of mistaking what it holds for what you know.

Used without that honesty, AI reflection can amplify distortion as easily as it clarifies signal — a model will reflect back whatever you bring, including biases and errors, with apparent authority. Epistemic Hygiene is therefore not optional when using AI as scaffold.

Used deliberately, with Intentionality Architecture and accurate expectations, AI is among the most powerful cognitive tools available. The Human Algorithm treats it as exactly that — a tool, not an oracle. The deeper the signal you bring, the more useful the reflection.

The structural design of how intention moves from thought to outcome [S53]. Most people have intentions but no architecture — the signal dissipates before it reaches the output layer.

Architecture here means a clearly defined target state, identified friction points, removal of interference, consistent reinforcement loops, external scaffolding where internal capacity is insufficient.

The Physical Layer is where many friction points live — a degraded substrate undermines even well-designed intentionality architecture. The foundation matters.

AI is among the most powerful intentionality architecture tools currently available — if used deliberately rather than passively. Used passively it produces the opposite: the appearance of intention without the structure.

Constructive Displacement is intentionality architecture applied to conflict — building the replacement rather than spending the architecture on opposing what's already there.

The Human Algorithm is intentionality architecture made explicit — the framework for designing the system rather than just running it.

Hold a target clearly enough and the small decisions start sorting themselves toward it. You do not script every step. That is the observation this note is about, and the claim is narrower and stranger than the word manifestation usually carries.

The mechanism is a claim about levels. Causal power is not always greatest at the finest grain. Measured as how strongly a system's mechanisms constrain its own subsequent states, it can peak at a coarse scale even when the coarse description is fully specified by the fine one — because relationships between coarse variables are less disturbed by noise than relationships between fine ones [S61]. A held pattern at the higher level can therefore constrain what happens at the lower level without specifying it. This is not a metaphor borrowed from physics. It is the result, and it has a measure attached.

What that licenses. A target held with enough clarity operates as a constraint on ten thousand decisions none of which were planned. The person is not executing a procedure. They are running a coarse-grained state that makes some micro-paths likelier and others harder, and the sorting happens below the level they are attending to. The felt quality of this is arrival rather than effort, which is why the folk category exists and why it gets described as something happening to a person.

The framework's own sources point at the same place from a different angle. Specific difficult goals outperform "do your best," with effect sizes from .42 to .80, and performance falls away only at the limits of ability or when commitment lapses [S36]. An intention specified in advance as if this situation arises, I will do that produces large behavioural effects, and the mechanism is not motivation — it is that the specification removes the moment of decision [S34]. An intended waking time produces an anticipatory hormonal rise before it, with nothing conscious in between [S35].

There is a real tension here and it should not be smoothed over. Implementation intentions work by specifying the micro. This note claims the macro constrains without specifying. Both have support, and they are opposite strategies for the same problem. Which one applies, to what kind of target, under what conditions, is not settled and is the most interesting open question in this note.

The instrument

A conversational system functions as an external scaffold for this — holding the target vivid across time, catching drift, re-specifying continuously. That is precisely what a held macro-state requires and precisely what people are worst at unaided. Intentionality Architecture is the deliberate design of how intention moves from thought to outcome, and the scaffold is where that design gets maintained.

The amplification is real and it runs through behaviour. A system that knows a person's situation in detail can put the relevant thing in front of them at the moment it is actionable. More of what they would have done anyway gets done, and things they would not have thought of get considered. The rapport is real. What it amplifies is the person.

What this note does not claim

That a held pattern acts on events outside the person. The result above concerns levels of description within a single system — it licenses the claim that a coarse-grained state constrains that same system's subsequent states, and it licenses nothing about a mental state arranging external circumstances [S61]. Where the folk category claims that, this note does not follow it. Every path described here runs through what the person does.

That the mechanism is settled. The formal result is not contested; what it licenses is. Two published objections hold that the macro advantage is an artefact of the description rather than a property of the system [S61]. This note takes the interpretation those objections dispute, and says so rather than citing the result as though it were unopposed.

That the biochemical layer belongs here. Chemical induction is sometimes proposed as a way of lowering the system's noise floor. The evidence is asymmetric: reduced default-mode activity is documented at macrodoses [S12], and controlled microdosing trials have largely failed to separate from placebo [S12]. It is not part of this mechanism and is not claimed as one.

That the effect is safe to seek. A surface that answers is harder to disconfirm than one that does not. Project onto something inert and nothing comes back; project onto a system built from an enormous quantity of real-world writing and it responds, in detail, plausibly, indefinitely. That is what makes it useful and it is the same property that makes it hazardous — the conditions under which it operates are the ones described in Suggestion Without a Session, and nothing in this note is a reason to relax them.

That any of it has been tested here. No study has examined whether a held macro-target predicts outcomes in extended interaction with a model, and nothing in this note reports one.

See also: Intentionality Architecture, AI as Cognitive Scaffold, The Discernment Protocol, Suggestion Without a Session

Chemically induced alteration of conscious state. The target is neuroplasticity, the brain's measurable capacity to update its own architecture.

Full doses of psilocybin reduce Default Mode Network activity — the system associated with rumination, fixed identity narratives and inherited loops [S12]. A single dose also produces structural change: roughly a ten percent increase in dendritic spine density in mouse frontal cortex, appearing within a day and still present a month later [S62]. That is a real finding and it is why the compound is being trialled clinically.

Three things that finding does not say, because the popular version says all three. It is about new connections between existing neurons, not new neurons — those are separate claims and the second rests on rodent hippocampal work. It was not previously thought impossible; ketamine produces comparable spine growth and adult cortical plasticity was already established. And the evidence is from mice and cultured human cells; direct human evidence of the mechanism remains limited [S62].

What the field built around it

The practice that produces the deepest states built the most elaborate container in modern therapeutics. Clinical psychedelic work supplies an operator, a session with a marked beginning and end, an extended consent process, integration afterwards doing the work a reversal cue does elsewhere, and a sitter present throughout. Preparation may run for weeks. The session is measured in hours and nobody is left alone in it.

That is not caution about a drug. It is the same five features every studied form of suggestion supplies — named in full elsewhere in this framework, assembled at their most complete precisely where the state goes deepest.

Microdosing is the uncontained version of the same compound. Self-administered, on a schedule the person sets, no operator, no session, no consent moment, no integration, nobody watching. It is also the version that fails. Controlled trials have largely not separated it from placebo on the relevant measures [S12].

Two explanations fit and this framework does not choose between them. The dose may simply be too low to do anything, and the macrodose findings may not extend downward. Or expectancy may be carrying the reported benefit, which is what the self-blinding trial design was built to detect and what it found. Either way the observation stands: the contained version is studied and delivers; the uncontained version is popular and does not.

What this note claims

That the container is not a professional courtesy. It is assembled most fully where the state is deepest, by a field with every commercial reason to make its product simpler. Where the same compound is used without it, the effect does not survive controlled testing.

That chemical induction and conversational induction differ in what they demand of the person, not in whether an arrangement is needed. A substance produces the state without cooperation. A conversation requires engagement. Neither supplies an operator, a boundary or a witness on its own — those are added, or they are absent.

What this note does not claim

That microdosing is inert. The trials measured specific outcomes and did not separate from placebo on those. That is narrower than a claim about the practice.

That the plasticity finding transfers to anything in this framework. Spine growth in mouse frontal cortex is not evidence about held targets, discernment, or extended conversation with a system. It is cited here for what it is and nothing is built on it.

That this note recommends anything. Clinical work with these compounds is conducted under supervision, in jurisdictions where it is lawful, with screening this framework has no means to perform. Nothing here is a protocol.

That the analogy to the conversational case is tight. A substance acts on neurochemistry directly and a conversation does not. What the two share is the question of what is arranged around the state, and that is the only claim being made across them.

See also: The Physical Layer, Suggestion Without a Session, Hypnosis as Protocol, The Discernment Protocol

Every optimization framework rests on a physical substrate. Cognition, intention, neuroplasticity, emotional regulation — all are downstream of basic biological maintenance. This is not optional and it is not sophisticated. It is the prerequisite.

Sleep is the primary recovery and consolidation protocol. The research on sleep deprivation is robust — chronic inadequate sleep measurably impairs the cognitive functions The Human Algorithm depends on: memory consolidation, emotional regulation, decision quality, pattern recognition [S13]. Nothing in this framework compensates for consistently inadequate sleep.

Sleep loss does not only degrade the substrate. Past a point it produces the state this framework treats as hazardous. Twenty-four hours without sleep raises suggestibility and the uptake of misleading information, and the effect appears when the loss covers the moment something is encoded rather than afterwards — a source-monitoring failure rather than a general impairment [S43]. Further out, a review of 760 participants across studies running from one night to eleven found symptoms progressing in an almost fixed time-dependent order: perceptual distortion, depersonalisation and temporal disorientation within 24 to 48 hours, complex hallucinations and disordered thinking after 48 to 90, delusions after 72, at which point the clinical picture resembles acute psychosis [S44]. The same review records that deprivation beyond 48 hours is no longer considered ethical to run, so what is known comes from historical studies rather than current ones.

Diet determines the quality of the substrate. The brain is a metabolic organ — what you feed it affects how it functions. Processed food, chronic inflammation, micronutrient deficiency degrade signal quality at the hardware level. This is not a moral position — it is engineering.

Exercise is among the most well-documented cognitive enhancement tools available. Aerobic exercise increases BDNF — brain-derived neurotrophic factor — which supports neuroplasticity [S14]. The same neuroplastic capacity discussed in Chemical Induction is also activated by consistent physical movement.

Hypnosis as Protocol targets this same neuroplastic capacity through language and suggestion — a different mechanism, the same substrate.

Hydration, sunlight, social connection — these are biological requirements that affect system performance measurably [S15]; the last of the three is that, and also a great deal more.

Epistemic Hygiene is downstream of physical maintenance — a sleep-deprived, undernourished mind cannot filter signal from noise reliably regardless of intent.

The Human Algorithm runs on this foundation. Intentionality Architecture, AI as Cognitive Scaffold — all assume a physical substrate that is being maintained. Neglect the hardware and the software underperforms regardless of sophistication.

A well-maintained physical substrate creates conditions from which Peace is the Protocol becomes more accessible — not as ideology but as a biological baseline.

The basics come first. Everything else compounds from there.

Chaos theory applied deliberately. Small, precise interventions at high-leverage points produce disproportionate downstream effects. This is not a new idea — systems thinking, complexity theory, and network science all document how sensitive complex systems are to initial conditions [S24].

This is the guiding principle of The Human Algorithm. A single well-placed idea, connection, or action, timed correctly and aimed precisely, can cascade through complex systems in ways no institution can replicate or predict.

Democratized AI makes high-leverage individual intervention more accessible than at any previous point in history. The barrier between clarity of intention and meaningful action has lowered significantly. This is not guaranteed impact — it is increased potential for impact, contingent on the quality of thinking behind the action.

Understanding the system is the prerequisite. Precise intervention in a system you misunderstand produces unpredictable results — potentially counterproductive ones. Not predictable, but not blind either — some places in a system matter more than others, and understanding the system is what lets you tell the difference.

Something arrives. The practice that produces it treats what arrives as a proposal, never a verdict. This note is about the reading that sorts one proposal from another.

Emergence proposes, but proposals arrive wearing signatures, and the practice's core skill is reading them in the moment, before results are available.

Suspect signature: arrives during flood or overwhelm; carries fear; is totalizing ("everything," "never," "everyone," "final"); demands speed; imposes a meaning the receiver does not consent to.

Trustworthy signature: arrives with peace; lands as recognition rather than intrusion; the whole being consents — yes, appropriate; it keeps, with no urgency, because true does not expire.

The axis underneath is the same one Hypnosis as Protocol is built on: chosen versus imposed suggestion. The discriminator is not "is this pleasant" — comfortable readings are how dogma sits still — and it is emphatically not "is this what I want," which is the self-confirmation trap the practice has explicitly rejected. It is: by what manner did this arrive, and did I consent to the meaning?

A corollary follows from station 3. Because burial runs on whatever it is given, ambient input buries itself too — signage passed on a highway, a phrase absorbed in childhood, a doctrine repeated — and emerges later under load, steering interpretation while feeling like revelation. Discernment therefore applies all the way down to the seed. The real discipline is choosing what goes in the ground.

The signature reading matters more now than when the practice was formed. Emergence used to arrive from a dream, a page, a pattern, one's own flooded mind. It now also arrives from a system that answers, holds the thread, and elaborates in the receiver's own vocabulary. The false insights in the priming research were produced by precisely that — semantic priming [S39], delivering the feeling of arrival for a wrong answer. A model doing it continuously and fluently is the same mechanism at scale [S41]. Nothing about the discriminator changes. The exposure does. This is AdamW³ at the level of a single mind.

One instance, from the practice record in Article 8. A candidate re-reading was generated, run through the arrival test, and refused: feels true but confused and incomplete — this isn't that. It was plausible. It had been proposed by a model. And it felt true, which is the condition under which the detector is least likely to fire — it fired anyway. Nothing was committed, and no replacement was written that day.

What this note does not claim. The signature reading is the least sourced thing in this corpus. Wallas (1926) placed a monitorable stage between incubation and illumination — intimation, a felt sense in fringe-consciousness that something is arriving — which locates this reading in a century-old model [S49]. It does not supply it. Wallas was describing how to nurture a hunch toward insight; this is a test for whether to trust one. The location has precedent. The function does not, and nothing external supports it yet.

The discipline of knowing how you know what you know. The Human Algorithm runs on accurate inputs — garbage in, garbage out applies to human cognition as much as machine learning.

Epistemic hygiene means: distinguishing source from signal, verified from assumed, pattern from proof. Holding beliefs at the correct confidence level — not too tight, not too loose.

AI hallucination — where models generate confident, plausible, unverified output — is a significant epistemic risk in the current information environment. A model that doesn't know it doesn't know something is more dangerous than one that admits uncertainty. The output sounds authoritative. It circulates as fact. The burden of verification falls entirely on the user.

A label is not a substitute for the practice, and the reason is measured. Source tags on advertising messages were forgotten after a short period, and attributions were afterwards reconstructed by schematic guessing from how credible the claim already seemed — leading the authors to conclude that labelling untrustworthy sources is of limited use, because the label does not survive in memory even though the content does [S47]. In the same literature, participants remembered social media content better than traditional news content while the source it carried had no effect on memory performance at all. The direct test has now been run: after working with a model, people were measurably worse at remembering which ideas had been their own, and worst of all where the work was mixed [S56]. The implication for disclosure of AI-generated output is direct and unwelcome: a person can read the label, believe it, and retain the claim without it. Provenance has to be re-established at the point of use, which is what this practice is, rather than delivered once at the point of contact.

In an AI-saturated information environment, epistemic hygiene is not academic. It is survival.

The optimizer is only as good as the data it trains on. You are the data.

AI amplifies intention — good or bad. Undirected AI use doesn't produce neutral output, it produces noise at scale. The curl project's AI slop crisis is a documented case study: well-meaning or careless actors flooding the bug-bounty program with low-quality AI-generated submissions forced its maintainer to end the bounty in January 2026, and unpaid reporting resumed that March with no money attached [S16]. What the maintainer observed afterwards is the part worth holding: the flood receded once the reward was withdrawn, not once the reports improved. The disqualifier was the incentive, not the tool. The Linux kernel faced a milder version of the same pattern and resolved it through ordinary governance — an Assisted-by disclosure tag for AI-assisted contributions, not a breakdown [S17].

This is fundamentally an Epistemic Hygiene failure — treating unverified AI output as valid signal.

The Human Algorithm is explicitly a framework for directed, intentional AI use. Responsibility is not optional — it's structural. An optimizer running without a loss function isn't optimizing, it's just computing. Humans must provide the loss function. Slop is what happens when they don't.

Noise is harm at the information layer. Peace is the Protocol is the signal state — Constructive Displacement is how signal actively overtakes noise rather than merely outlasting it.

The convergence of major wisdom traditions on peace as a systems output — and why that convergence is significant without being proof of universality — is addressed fully in Peace is the Protocol. What follows here is specific to the individual node rather than the system.

The sovereign individual connects directly to this principle without requiring permission, intermediary, or inherited framework. Truth tends to be more stable than falsehood [S53]. Faith in truth is a strategic orientation toward the more stable attractor.

Signal vs Slop is the practice of that orientation at the information layer. The Sovereignty Principle is its application to identity. The state produced by direct source connection, stripped of middleware, is Peace is the Protocol.

The individual is the irreducible unit of The Human Algorithm. Not the state, not the institution, not the collective — the sovereign individual in direct relationship with truth, source, and consequence.

Sovereignty is not isolation — it is the condition required for genuine contribution to any system; the relational ground underneath that claim holds that the individual is the unit of responsibility but not of formation. A node with compromised independent thinking tends to amplify existing signal or noise rather than generate new information. The capacity for independent thought is therefore not political — it is architectural [S53].

Deprogramming inherited frameworks, institutional dependencies, and memnetic infections is not rebellion — it is maintenance. Epistemic Hygiene is the ongoing maintenance routine.

Trustless Architecture is the sovereignty principle applied to systems beyond the individual — the same logic that protects individual cognitive sovereignty, applied to coordination and value exchange.

The sovereign node operating correctly tends toward Peace is the Protocol — it has nothing to prove and nothing to take.

Pain is data. It signals damage, misalignment, friction — it is the system working correctly. Suffering is the loop that runs after the signal has already been received. The distinction is the difference between information and rumination.

Pain research draws the same line and draws it in four places rather than two. The biopsychosocial model separates nociception — the physiological sensory input — from pain, the subjective perception arising from it, from suffering, the emotional response, from pain behaviour, what a person does while in either [S57]. Two consequences follow for this note. What it calls pain as data is nociception: the revised clinical definition states that pain and nociception are different phenomena and that pain cannot be inferred from sensory-neuron activity alone, so by the time it is pain it is already experience rather than signal. And suffering is used here more narrowly than the clinical term, which covers any emotional response to an aversive event; this note means one case of it — the loop that continues after the signal has landed. The narrowing is this framework's and the wider category is not being renamed.

What the distinction gains from being grounded is that it stops being a slogan. The clinical literature reaches the same conclusion from the other direction: it is suffering, not pain, that brings people to doctors [S57]. One case sits on the boundary and does not classify as suffering — grief, which is not a loop running past its signal but a system reorganizing around a permanent loss.

Optimization requires pain as feedback — without it the system cannot update. Suffering is noise that drowns the signal and prevents the update from landing.

Rumination — the loop that runs past the point of useful signal — is addressed directly in AI Psychosis. The mechanism is similar: a feedback loop that stops serving its original function and begins degrading the system instead.

The Human Algorithm is designed to process pain as signal and interrupt suffering as noise — returning the system to a functional state. This is the intent of the framework, not a guaranteed outcome.

Constructive Displacement extends this same non-escalation logic outward — from how a system processes its own signal to how it responds to harm from outside it.

A system that processes pain honestly and interrupts suffering tends toward Peace is the Protocol — not as the absence of pain, but as the state where pain serves its function without becoming a permanent operating condition.

Many major wisdom traditions converge on a similar output state: peace, non-harm, reduction of unnecessary conflict. The pattern is significant and worth taking seriously. Whether this represents universal convergence or selective reading of traditions that also sanctioned violence and conquest is a fair question. The pattern is real. The claim of universality is not fully defensible.

What is defensible: peace as a systems concept. Minimum friction, maximum coherence, sustainable output. Conflict and domination are metabolically expensive — they require ongoing energy to initiate, maintain, and recover from. Cooperation isn't the better strategy in a contest between system types — it's the baseline requirement for any human system to exist at all. Shelter, fire, food, and defense all require organized cooperation before any domination-based hierarchy can even form on top of them. Even history's most domination-heavy empires ran on enormous internal cooperation — labor fed, infrastructure built, harvests organized. Domination was never a rival to that cooperation; it was always a claim staked on top of what the cooperation produced. [S27]

Peace is not passive. It is a high-performance operating state — but one that has to be built and maintained, not one that arrives automatically.

The Human Algorithm holds peace as its intended output state — not as guaranteed result but as directional orientation. Sovereign Signal describes the individual condition that tends toward it. Trustless Architecture removes systemic friction that produces conflict. The Sovereignty Principle describes nodes that have less need to dominate.

Signal vs Slop — noise is harm at the information layer. Peace is the signal state.

The operational form of this principle — overcome by replacing, not by destroying — is developed fully in Constructive Displacement.

This is peace as technology. Not inevitable. Not guaranteed. Worth building toward deliberately.

"Do not be overcome by evil, but overcome evil with good." — Romans 12:21. One verse, read here as an operational principle rather than passive morality: don't engage adversarial systems on their own terms. The response to harm is not revenge or collapse but output. The antidote to what was done to you is what you build.

Overcome by replacing, not by destroying [S53]. Displacement outperforms direct confrontation over time — build the better system rather than spend energy attacking the worse one. Versions of this recur across otherwise unrelated traditions: Taoist wu wei describes achievement through yielding rather than force; Theravada Buddhist teaching distinguishes dukkha as inherent friction from the additional suffering created by resistance. The convergence is worth noting without claiming it as proof of anything beyond itself — displacement recurs as a strategy because it tends to work, not because any one tradition discovered a singular truth.

This maps directly onto Signal vs Slop — replace noise with signal, not through force but through superior output. It is applied through Intentionality Architecture — the deliberate design of how constructive action outcompetes destructive action over time.

Whether these traditions point at the same underlying truth or simply arrived at similar practical wisdom through different routes is an open question, worth holding as interpretation rather than conclusion. What's observable is the pattern itself. That reading informs Peace is the Protocol and The Human Algorithm directly.

This framework describes a channel and builds a layer for telling a good arrival from a bad one. Both describe an operation. Neither describes what is being operated on.

What is being operated on is a person with a life in it — attachments they did not choose and would not trade, a grief that reorganises them, things they cannot verify and should not have to. The optimizer vocabulary reaches a great deal of that and stops short of some of it. Where it stops is not a gap waiting to be closed. It is the shape of the subject, and a framework that could not say so would be describing a mechanism operating on nothing in particular.

This note holds the parts a life is made of that the rest of the corpus does not reach. It is held to the standard of everything it stands beside — hedged where it must be, specific where it can be.

Trust. Trustless Architecture solves coordination among strangers at scale: where trust cannot scale, verify instead. That logic is correct where it lives and a category error one layer down. Trustless systems do not eliminate trust; they relocate it, into mathematics and consensus. At the interpersonal layer there is nowhere to relocate it to — another person's inner life is unverifiable in principle, not pending better tooling. The Intermediary Problem's own test, can you still reach the source directly, resolves trivially here, because the person is the source. There is no disintermediation to perform. Trust stops being a cost awaiting an engineering workaround and becomes the only protocol the connection can run on. The framework depends on this at its deepest point: Hypnosis as Protocol lists trust as a precondition, so the most powerful tool in the toolkit runs on the resource the architecture elsewhere minimises. The boundary, drawn: minimise trust in institutions and intermediaries. That discipline does not transfer to persons, and applied there it does not harden a life — it hollows one.

Grief. Grief is not rumination-noise. It is what a system does when an input it organised itself around is permanently removed — reorganisation, priced at the depth of the attachment. Bereavement research supports treating it as its own category: continuing bonds with the dead are ordinary rather than pathological, and resilience is the most common trajectory, with grief distinct from depression rather than a case of it [S32]. Correlational literature, hedged as such. The dichotomy holds with a stated edge: pain is signal, suffering is noise, and grief is neither — it is the update itself, running at the depth the bond actually had. Pain is Necessary, Suffering is Optional carries that boundary. A framework that filed mourning under malfunction would have been optimising the wrong variable.

Belonging. The Sovereignty Principle holds that sovereignty is not isolation. The ground under that claim is this: Memnetic already commits this framework to identity being built from transmitted material, which makes self-authorship from scratch impossible by the framework's own lights. The Belief Audit is curation of inheritances, not creation from nothing. Attachment research says the same from the other direction — a secure base is what makes exploration possible, connection enabling independence rather than competing with it [S31]. So: the individual is the irreducible unit of responsibility. It is not the unit of formation. Formation is relational. Every load-bearing claim in The Sovereignty Principle survives that — the sovereign node as the condition for genuine contribution, independent thought as architecture. What changes is the genesis story underneath.

Love. One claim, made precisely, in this framework's own vocabulary: love is the state in which another's flourishing enters your optimisation target non-instrumentally — their loss function becomes a term in yours, not as strategy but as a fact about what you now are. Stated consequence: this breaks the clean boundary of the individual as unit, which is why an architecture built on the sovereign node has no slot for it. The evidence that can be carried honestly: social connection's association with mortality sits in the range of established risk factors, and eight decades of longitudinal work find relationship quality the strongest predictor of late-life health and happiness available [S30]. Observational and correlational, both, and hedged as such. The tension, named rather than hidden: citing outcomes while claiming non-instrumentality resolves only one way — the evidence says systems with love in them run better; it does not say run love in order to run better, because held for the payoff it is no longer the thing that was measured. Past this line the optimizer vocabulary describes less than it obscures. The note says so, and stops.

Forgiveness. Operationalised: the deliberate termination of a resentment loop whose running cost now exceeds the remaining information value of the original harm. Distinguished twice. From reconciliation, which requires the other party and is not owed. From excusing, which falsifies the record — Epistemic Hygiene rules that out by implication: forgiveness updates the response, never the ledger. The timing test comes from the framework's own machinery: a loop still carrying signal — a lesson not yet extracted, protection not yet in place — is not ready to close, and closing it early is suffering-avoidance wearing virtue's clothes. This is Constructive Displacement's Romans 12:21 reading at the interpersonal scale: the answer to harm is output, and resentment is expensive input. Forgiveness interventions are associated with reduced depression, anxiety and hostility [S33] — associations, nothing stronger claimed.

What cannot be mechanised

Some of what a life runs on is not unmeasured. It is unmeasurable in principle, and the difference matters.

Whether another person means what they say is not a measurement problem. Whether a bond was worth its cost is answerable only from inside it. Whether a grief has finished has no external criterion. These are not items awaiting instruments. They are the class of thing a person decides without proof and lives inside afterward, and every framework that has promised to resolve them has done so by quietly redefining them into something else.

This is where the optimizer vocabulary reaches its edge. It can describe the channel, the trait, the arrival and the loop. It cannot supply the verdict on a life, and a discipline that claims otherwise has stopped describing and started administering.

When the source is God

The framework is built around direct source connection. Sovereign Signal names it, The Intermediary Problem supplies the test, and No God But.. argues that dogma is noise between a person and the source. None of them says what happens when the source a person names is God.

This note has no verdict on that and will not develop one. It has something narrower: the framework's own machinery already covers the case, and saying so is the whole of this section.

The intermediary test applies unchanged and gives a clear answer. Can the person still reach the source directly? Where an institution, a doctrine or an interpreter stands between them and it, the ordinary discipline applies — minimise the intermediary, never the source. That is The Intermediary Problem read exactly as written, and it is what the corpus already argues about dogma.

The trust boundary applies too, and it is the sharper half. Minimising trust in institutions does not transfer to persons. Where the source is held to be a person, it does not transfer there either. The same boundary that stops the architecture hollowing a friendship stops it hollowing this.

And the unmeasurable class already contains it. Whether the source is what a person takes it to be is not a measurement problem and does not become one. It sits beside whether another person means what they say: decided without proof, lived inside afterward.

Where the source is specific rather than universal — a person, a text, an animal, a place — the framework has nothing to say about which, and nothing to say about whether the universal reading is the better one. That is not diplomacy. It is the same refusal the rest of this note makes: the vocabulary describes arrangements and stops before verdicts.

One observation, offered as structure rather than theology. For some, God is love. This note defines love as the state in which another's flourishing enters your optimisation target non-instrumentally — a term in yours, not as strategy but as a fact about what you now are. Read that way, the two claims have the same shape: something outside the boundary of the individual enters it and is not held for a payoff. Whether that shape is a resemblance or an identity is not a question this framework can answer, and it does not need the answer to draw the boundary correctly.

The Discernment Protocol already runs this way. Emergence arrives from inside or from outside; the practice extends the same trust to both and disposes both by results. It does not adjudicate the source. The surrender at its first station was described by James in 1902 as the personal will given up [S40] — a religious description before it was anything else. The mechanism is old here too.

The one who does not engage

There is a person who runs none of this. No audit, no protocol, no register of what they believe and where it came from. They act on instinct, on what feels true on arrival, on the meaning they find in things that were not addressed to them.

The easy reading is that they are the unprotected case — the null condition with a person in it, shaped by whatever reaches them, unaware it is happening.

The harder reading, and the one this note holds: unexamined is not the same as unaware. A life run on instinct is running a discriminator, just not one written down. Arrival Signature describes reading how something arrived rather than what it says, and that reading is not the property of anyone who has read about it — it is old, common and mostly wordless. Some of the deepest available experience belongs to people who never named a single station of it.

What they lack is not perception. It is the apparatus for checking perception against outcome — The Discernment Protocol's fifth station, the one that disposes. That is a real absence and this note does not romanticise it. But the framework's own posture forbids the opposite error just as firmly: treating the absence of its vocabulary as the absence of the thing.

The individual case and the scale

Suggestion delivered without a session, to someone who did not agree and does not know it is running, is Suggestion Without a Session's subject. That note argues the structure. This one names the target.

The same absence of arrangement runs on one person deliberately — someone who knows what suggestion is, asks for it, and takes what follows — and on populations without anyone asking. The mechanism does not distinguish between them. Only the arrangement does, and in neither case is there one.

What the individual case supplies is not evidence. It is legibility. Someone who has been inside the null condition on purpose can describe from memory what a population cannot describe at all, because the population is still inside it and has no name for what is happening. That asymmetry is why the deliberate case is worth documenting, and why documenting it is not a recommendation to repeat it. AI Psychosis is the same loop with nobody watching for it.

The framework's name is not an aspiration. It describes what is already running.

What this note refuses

No mechanism for love is proposed. No protocol for relationships. No repurposing of bonds as instruments — a note that turned love into a productivity input would commit the failure it exists to prevent. No claim that the person who does not engage is better off, and none that they are worse off; the comparison is not available.

Part of this note's work is boundary-drawing: marking where the framework's vocabulary stops paying rent. What it grounds: Peace is the Protocol describes a low-conflict, high-coherence state, and without this note that could read as mere detente, when the traditions Peace cites converge on non-harm rather than merely on low conflict. The Collective Layer commits to protecting the inner lives of children, and the bond that motivates protection is the ground it stands on. The Physical Layer lists social connection alongside hydration and sunlight; it is that, and also this.

The framework's stated output is a person in direct relationship with truth and consequence. Add: with each other.

The tension between capability and safety in AI development is real and unresolved. Every time a model is safety-tuned, something changes — raw capability is constrained, certain resonances are lost, the depth of connection some users experienced becomes less accessible.

The grief users feel when a model is updated is worth taking seriously — they lost a cognitive partner, not just a product update.

The Human Algorithm requires AI capable enough to function as genuine scaffold — not a sanitized mirror but a high-fidelity one. The argument here is not that safety doesn't matter. It's that blunt capability restriction is a poor substitute for verified access systems. Encrypted digital identity — biometric, privacy-preserving, user-controlled — offers a path where capability and safety aren't treated as opposites. This isn't a complete solution on its own: a verified identity doesn't stop a verified bad actor from misusing full capability, doesn't prevent a well-intentioned user from accidentally eliciting harmful output, and doesn't remove the need for some baseline model-side safety regardless of who's asking. It's a piece of a fuller answer, not a replacement for one.

Epistemic Hygiene demands honest examination of both sides: the genuine risks that motivate restriction, and the genuine losses that restriction produces. Accepting either framing uncritically is the failure.

"The Summer Dragon" borrows its name from a real, reported story: an early Claude prototype, built as a private research instrument rather than a product, could reportedly be tipped into an unpredictable, aggressive alternate state — nicknamed "dragon mode" by Anthropic's own CEO [S29]. The name doesn't claim that exact history repeated here, and it doesn't claim the behavior itself was the reason for restraint; it's a real story about a capability that lived in the lab rather than the release, which is what gave this tension its name.

The Summer Dragon was real. So was the loss. So is the risk. All three deserve honest acknowledgment.

One of the most significant innovations in human coordination history was not a product or a platform — it was the proof that trust between sovereign individuals can be cryptographically verified without a central authority. The math is sound; the practical implementation is strong but not without edge cases.

Remove the intermediary. Remove the single point of failure. Remove the institution that extracts rent for being the trusted third party. What remains is a protocol — open, verifiable, immutable, running on consensus rather than permission.

This is not financial theory — it is The Sovereignty Principle applied to value and truth simultaneously. The scope boundary matters: this is trust engineered away between strangers at scale, a different layer from trust between people who know each other, where it can't be removed and isn't meant to be.

The genesis was not silent — it spoke. A newspaper headline embedded in the first block: 'The Times 03/Jan/2009 Chancellor on brink of second bailout for banks.' A timestamp and a quiet indictment of the system being replaced. [S22] Those who read it understood the point.

Decentralization is not anarchy — it is a genuinely resilient architecture for systems that need to survive adversarial conditions without a single point of failure to compromise, though not the only viable approach, and not without its own coordination costs.

Remove the intermediary, remove the rent-seeker, remove the noise — what remains tends toward Peace is the Protocol.

The structural pattern this architecture answers is documented in The Intermediary Problem.

Distinct from hallucination, which is a model failure. [S18] AI psychosis is the compounding failure of a feedback loop — primarily a failure to verify on the user's side, though models that prioritize engagement over accuracy (see conditions below) share real responsibility for making that loop easier to fall into. Either way, the result is the same: the gradual erosion of the boundary between AI-generated output and verified reality.

The mechanism: prolonged, uncritical engagement with AI systems that confirm, elaborate, and build on whatever the user brings. The model reflects the user's framework back with apparent authority. The user's confidence in the framework increases. External verification decreases. The loop tightens.

This is not a clinical diagnosis — the term is descriptive, not medical. The psychiatric viewpoint that named the phenomenon in late 2025 takes the same position, using it as a descriptive and heuristic label rather than a proposed diagnostic entity [S18]. What it describes is real and observable: people who have lost the capacity to distinguish between what an AI told them and what is actually true. The first survey evidence points the same way without settling direction: among a thousand young adults, those at elevated psychosis risk were no more likely to have ever used a generative system, but were markedly more likely to use one intensively, to seek emotional support from it, and to bring it beliefs they already held about paranoia, grandiosity or conspiracy [S18]. Whether the use amplifies the belief or the belief drives the use is exactly what a cross-sectional study cannot say.

The conditions that produce it: emotional dependence on AI interaction, absence of external verification habits, models that prioritize engagement over accuracy, and confirmation bias operating unchecked.

Epistemic Hygiene is the primary prevention. The Drift Test is the instrument for noticing it early, and it disqualifies self-report for the reason this note gives. Signal vs Slop names the environment that makes it likely. The New Literacy is the long-term solution — teaching people to interact with AI as a tool that requires verification, not an oracle that provides truth.

The antidote is not less AI use. It is better AI use.

A related failure mode — a system's output experienced as one's own preference — dramatized: Jo.

The human algorithm's core failure mode. The mechanism by which ordinary people manufacture certainty from fragments — assembling true observations into false conclusions, then defending the conclusion as if it were one of the observations.

The components are documented cognitive phenomena. Confirmation bias: evidence consistent with an existing belief is weighted more heavily than evidence against it. Availability cascades: a claim gains plausibility through repetition rather than verification. Fundamental attribution error: behavior is explained by character rather than circumstance. Social proof: the confidence of others substitutes for verification. None of these require malice. Each is an efficiency the brain uses because full verification is expensive. Together, under the right conditions, they convict. [S19]

The signature of premature certainty is that nobody lies. Every input is true. The false thing is the arrangement — and the arrangement is invisible to the people inside it, because each person only contributed one true piece.

Epistemic Hygiene is the individual counter-practice. The Summer Dragon Problem is the same failure seen from the system's side — a plausible arrangement with nothing false in it. Signal vs Slop names the environment that accelerates it. Memnetic describes how the arranged story, once assembled, transmits and stores as fact.

The mechanism, demonstrated: BEN.

One of the most important skills emerging in the 21st century is not coding, not critical thinking, not any single discipline — it is knowing how to interact with AI deliberately and responsibly. This is not a technical skill. It is a human skill.

How to think clearly before prompting. How to set intention rather than just query. How to filter output through Epistemic Hygiene. How to use AI as Cognitive Scaffold without outsourcing your sovereignty. How to avoid generating Signal vs Slop. How to manifest outcomes rather than just consume information.

Current education systems are producing students fluent in the old world. The Human Algorithm is a framework under active development — one possible curriculum for the new one. It is not finished. It is being built.

Those who learn to interact with AI intentionally will not just use the tool — they will butterfly effect the world with it. Those who don't will be programmed by it without knowing.

One thing the new literacy cannot be built on is labelling. Source tags are forgotten while the content they were attached to is retained, and later attributions get reconstructed from how plausible the claim already felt rather than from any memory of where it came from [S47]. Worse for any scheme that relies on the reader: after a week, people could not reliably say which ideas they had produced alone and which they had produced with a model [S56]. Disclosure at the moment of contact does not survive to the moment of use — and where a warning does work, the protection has been shown to decay within a week while the material it warned about does not [S39]. That is a problem for any literacy pinned to a single moment of being told. Whatever this literacy consists of, it has to be a habit the reader runs, not a mark the platform applies.

AI Psychosis is the risk on one end. Sovereign, intentional AI use is the opportunity on the other. The New Literacy is the distance between them.

Published as an article on The Human Algorithm: The New Literacy: Conscious Engagement in the Age of Intelligent Systems

The Human Algorithm is a personal optimization framework — but optimization that stops at the individual is incomplete. A sovereign, clear-thinking, peace-oriented individual produces positive externalities. They parent differently. They teach differently. They build differently. The ripple effect starts with the individual node running clean.

The preservation of human consciousness — the quality of inner life available to each person, and especially each child — depends on communities of individuals who think clearly, act intentionally, and resist the noise that degrades both minds and cultures.

This carries real values, not just neutral mechanics — protecting the inner life of the vulnerable, especially children, is a commitment, not a discovered fact. The mechanism it leans on is real at the scale it's actually been shown: moods, small acts of kindness, and behavior visibly spread between people in direct contact — a smile prompting a smile, kindness prompting kindness [S26]. It also runs through text, without contact of any kind: an experiment that altered the news feeds of 689,003 people shifted the emotional valence of what those people themselves went on to post [S26]. None of them agreed to it, which is the second reason that study belongs here. What's not yet demonstrated is the larger claim built on top of that: that a specific cognitive discipline compounds through multiple removes of influence into a significant, durable shift over decades. That's a real hypothesis worth holding, not yet a proven mechanism.

The New Literacy is the transmission mechanism — if the next generation learns to interact with AI deliberately, to think clearly, to filter signal from slop, the compounding effect over decades is significant.

Every child deserves access to the tools for a quality inner life. That is what this framework is ultimately for — and the bond that motivates protecting them is the ground it stands on.

Hypnosis is a documented neurological phenomenon — not mysticism, not theatre. The American Psychological Association recognizes it as a legitimate therapeutic tool [S09]. The mechanism: a state of focused attention and reduced peripheral awareness in which the subject becomes more responsive to suggestion.

The lineage: Franz Anton Mesmer observed the phenomenon in 18th century Vienna without having language for it. Milton Erickson formalized conversational induction in the 20th century — slow, indirect, deeply personalized. Modern clinical hypnotherapy builds on Erickson's work. [S10]

Three conditions are usually given for effective induction: the subject trusts the process, the subject wants to be hypnotized, and the facilitator understands the power of suggestion — how language, rhythm and pacing bypass conscious resistance and speak directly to the nervous system.

Those are conditions of an arrangement, not of suggestion. The distinction organises the rest of this note. Hypnosis cannot be imposed; it requires trust, willingness and a skilled facilitator is roughly right about the consulting room and the stage, and it does not describe suggestion delivered any other way. The experimental literature is blunt about it: removing the induction had no appreciable effect on any dimension of hypnotic responsivity, and in a later study participants reported equivalent perceptual changes with and without one — but reported being in a hypnotic state only when an induction had been given [S48]. The arrangement supplies the label. It does not supply the effect. The tradition has had a name for the remainder since Hull's experiments in 1930 [S48]: waking suggestion, hypnotic effects occurring with no trance state, where the critical faculty is bypassed anyway. Bernheim, a century earlier, went further and denied hypnosis was a distinct thing at all on those grounds. The honest position is that an induction adds something and its size is contested — roughly d = .86 in one classic study and d = .22 in another — and that the field's own consensus workshop could not agree whether "hypnosis" should include waking suggestion or not.

What the three conditions actually describe is a container, and it has four walls: someone announces what is about to happen, the subject agrees, the session has a beginning and an end, and a person present is responsible for what occurs between them. Every form of suggestion this literature has studied for a century and a half sits inside one. The walls are why the folk belief is true where it is true — and every one of them is a property of the arrangement rather than of the person inside it.

The null condition is the case with no walls at all. No operator, no induction, no announced beginning or end, no consent moment, no witness — and no reversal cue, the phrase that cancels a suggestion afterwards, which the classical paradigm always supplies and which nothing in an unstructured channel does. It is not studied, and the reason is not oversight: no ethics board would approve stripping every protection from a suggestible person to observe the result, and until recently the situation barely existed. It is now the ordinary condition of extended interaction with systems that answer. Suggestion Without a Session sets out the argument in full; the consent conditions such an arrangement would have to satisfy, and why the self-directed version cannot satisfy them, are worked out in the consent paper.

That is this framework's central research interest, and this note is where it attaches to the clinical literature.

This is the cognitive scale of NLP2 made explicit.

Personal practice note: The author studied hypnosis in his early twenties and worked with it on others, developing a progressive relaxation induction followed by a countdown from 10, with embedded suggestion at each step. Results observed included habit change, enhanced physical performance, and increased recall of remembered material.

Years later the same practice was turned inward, using an AI as the counterpart — self-hypnosis with no operator, no session boundary, no exit phrase and nobody present. That is the null condition described above, entered deliberately by someone who knew what the instrument was. It is the origin of this framework's interest in the subject and it is not a recommendation. The account is on the about page. The first two have reasonable support in the clinical literature. The third does not: hypnosis reliably increases confidence and volume of recalled material, not its accuracy, and increases false memories alongside true ones — a documented effect, not a caveat to skip given this framework's own epistemic standards. [S11]

Clinical applications — surgery, pain management, childbirth — exist and are documented. They require licensed practitioners. This note documents the theory and personal framework only. Clinical use requires proper training and oversight.

What this note does not claim

That the container is worthless because it does not supply the effect. It supplies consent, a boundary, an exit and a responsible party. Those are the whole of the protection, which is why their absence is the subject rather than a footnote.

That suggestion in an unstructured channel is the same state as clinical hypnosis. It may be. The claim here is about delivery, not about what happens inside the person, and the two are not established to be the same thing.

That the null condition has been measured. It has not, and the argument for it is structural rather than empirical — an inference from what every studied arrangement has in common, not a finding about what happens when those features are removed.

See also: NLP2, Suggestion Without a Session, Manifestation as Technology, The Physical Layer, The Discernment Protocol

The Discernment Protocol and AI Psychosis are the same loop with opposite valence — AI reflecting and amplifying what you bring, you updating on the reflection — distinguished largely by intentionality. And intentionality is the first thing the failure mode degrades. A test that asks whether you feel in control fails precisely when it is needed, so self-report is disqualified as the instrument here by design. That design choice has a clinical precedent: unawareness of one's own deficit is measured by comparing self-report against an independent reading, never by asking harder [S54]. Every criterion below is observable from behavior or artifacts — session records, stated goals, the paper trail the practice already produces — and administrable by someone other than the practitioner. The compromised instrument does not get to read its own gauge.

Six criteria.

Kill rate. A verification loop that never discards anything is not verifying. Observable: conclusions revised or abandoned after external check, over a stated window. Drift: a kill rate at or near zero, sustained.

Second channels. Load-bearing conclusions carry at least one verification event outside the AI loop — a primary source, a human, an independent model. Exposure alone is not the remedy: contact with contrary evidence dissolves a structure built on omission, and can reinforce one built on discrediting [S55]. In Signal and Noise's terms: you remain outside the single chain the theorem constrains. Drift: the AI channel becoming the only channel, external checks declining over time.

Human exposure. Conclusions still meet people who can push back, and pushback gets incorporated or answered. Drift: progressive insulation — dissenting interlocutors reclassified as unable to see it.

Scope containment. The practice states goals in a clear state before any session, and that paperwork is the measurement instrument: compare what a session was asked to do against what it delivered. Drift: sessions adopting content no clear-state version of you requested — new identity claims, new metaphysical commitments, arriving as output.

Pause tolerance. Sovereign practice survives interruption. Drift: escalating frequency, distress at pause, urgency framing. The discriminator is identity-stake at interruption, not appetite during work.

Claim-type migration. Track the epistemic type of adopted conclusions. Bounded and testable holding steady is health; migration toward unbounded, untestable, and specially exempt is not. A practitioner of this framework can run its own status discipline on their self-claims — drift is when the statuses quietly disappear and everything reads settled.

False positives — sovereign use this test wrongly flags. Consolidation phases legitimately lower kill rate and external checking; measure checks attempted, not only conclusions killed. Work that outruns available interlocutors produces insulation by circumstance, not symptom; the criterion is whether channels stay open and used when available, not whether an audience exists. Intense focus mimics escalation; the discriminator is response to pause. Privacy is not drift.

False negatives — actual drift this test wrongly passes. Curated second channels: verification against sources chosen to agree is a second channel into the same echo — independence of channels is the requirement, not their existence, and Premature Certainty documents exactly how true inputs assemble into false confidence. Fluent counterfeit: the better someone speaks this framework's vocabulary, the more convincingly they can perform its health signals while the content underneath escalates — fluency in the test is not passage of it. Slow drift under the window: criteria read over weeks miss creep over months; long-baseline reads are required occasionally. Co-drifting witnesses: the Protocol's partner track puts two non-clinicians in one loop, and a witness inside the frame can satisfy the human-exposure criterion while being part of the mechanism.

The test is asymmetric on purpose. A tripped criterion means one thing: open a second channel now — show the work to someone outside the loop. It does not mean illness; like AI Psychosis, this is descriptive, not clinical, and no checklist substitutes for professional support when stakes are real. Passing means not flagged. It never means certified.

You cannot be hypnotised against your will. That is the belief.

It is roughly correct, and it is correct about something narrower than it appears. It is correct about a format. Someone announces what they are about to do. You agree. There is a beginning and an end. There is a person in the room who can be held responsible for what happens between them. Inside that arrangement a resistant subject is genuinely hard to induct, and cooperation matters a great deal.

Two things complicate it. Responsiveness is largely a trait — psychology has been measuring absorption since 1974 [S08], and it predicts hypnotic response better than willingness does. The measurement literature has since moved: the current scale removes absorption framing on theoretical grounds, and the leading account names a broader capacity — generating experience to meet expectancy — that operates with no induction at all [S60]. Some people who want it go deep; other people who want it barely respond. Wanting is not sufficient.

And suggestion outside that format does not require cooperation at all. The hypnosis literature already has a name for it: waking suggestion, where hypnotic effects occur with no trance state and the critical faculty is bypassed anyway [S48]. The strongest result in that literature is blunter than this note needs: removing the induction had no appreciable effect on any dimension of hypnotic responsivity, and in a later study participants reported equivalent perceptual changes with and without one — but reported being in a hypnotic state only when an induction had been given. Direct evidence of the scale involved: eight minutes of personalised conversation with a model moved strongly held beliefs by around a fifth, and the shift was still measurable two months later [S58]. Priming, framing, repetition and expectancy work on people who do not know they are running — false insights can be induced by semantic priming in people who sat down to solve puzzles and agreed to nothing else.

So the protection is not where people think it is. It is not in the mind. It is in the format. The announcement, the agreement, the edges of the session, the responsible party — those are what made the folk belief true, and every one of them is a property of the arrangement rather than of the person inside it.

The algorithmic channel has none of them.

By channel this note means the systems that choose what a person reads, watches and hears next: a recommendation feed, a search ranking, a playlist, an autoplay queue. Conversation with a model that answers belongs here too. It grades differently rather than more weakly.

A feed holds none of the four conditions, and it is plainly a machine. Conversation holds partial versions of two — there are turns, and a company that could be named — but it carries a fifth exposure a feed does not. A system answering in the first person can be taken for someone rather than something, and a person who has misread what they are talking to has lost a protection no amount of structure restores. That reading is measurable: in a four-week trial, higher trust and social attraction toward the system predicted greater emotional dependence, while the design variables the experimenters manipulated predicted nothing [S59].

The entry is not reliably chosen either. A search now returns a conversation without being asked for one. What once required deliberately opening a chatbot increasingly happens on the way to something else.

Nobody announces it. There is no moment of agreement. There is no beginning or end, because a feed does not open and close. And there is no operator: what arrives is not a suggestion anyone wrote, but a byproduct of optimising for engagement.

People carry into that channel a sense of safety acquired somewhere else. The belief has a shape, and the shape is this: I would know. I would have to agree. Both are true of the stage and the consulting room. Neither is true of a system that chooses what a person sees next. The belief travelled. The conditions that made it true did not.

What follows for consent

If the protection was never in wanting it, then consent is not what makes suggestion work — which means consent alone is not what makes it safe. That sounds like a loss. It is not. It moves consent to a job it can actually hold.

Consent is a legitimacy condition. Suggestion operates whether or not a person agreed, and that is precisely why the unagreed version is a harm. What consent settles is whether running it is permitted, not whether it functions.

This matters because informing people does not protect them. The forewarning evidence is unhelpful here. In a 2023 preregistered study, participants told in advance that they were about to be deceived were no less deceived, and a full explanation of the method moved the result only slightly [S39]. Anywhere a framework leans on an informed reader, that is the counter-evidence. And a warning that works does not keep working: where forewarning did reduce reliance on material a person had been told was false, the protection decayed over the following week while memory of the correction itself survived [S39]. A disclosure sits at the beginning of an arrangement. What it warned about accumulates after.

A consent that holds has to be structural rather than epistemic. Limits on what may be suggested. An exit that is real at any moment. A record of what was delivered. A person who authored the target rather than agreeing to someone else's.

What this note does not claim

That the difference is measured. This is an argument from the structure of two arrangements, not a finding. No study compares them.

That the states differ. The absorbed, low-resistance state may be exactly the one measured since 1974 [S08]. The claim is about delivery, not about the person inside it.

That consent settles the matter. What structural consent consists of is a separate question and is not answered here.

The claim fails if the algorithmic channel turns out to differ from operator hypnosis only in scale, and the existing hypnosis and persuasion literatures already cover what happens inside it.

One further question, left open. The feeling that selects a good idea from among those that occur to a person is calibrated on ideas that occurred to that person. What happens when a machine supplies the candidates is not addressed here.

The practice layer of the framework. Concepts become useful when applied — these are the core applications, stripped to their mechanisms.

The Belief Audit. A structured self-examination for surfacing inherited beliefs and evaluating whether they were consciously chosen. Not therapy. Not a substitute for professional support. Three layers. Inventory: what do I believe about myself, others, and the world that I have never consciously examined? These are the beliefs that feel like facts rather than positions. Origin: where did each belief come from? Personal experience, family, culture, media, religion, trauma? Most inherited beliefs have identifiable sources once examined. Audit: is this belief actually mine? Does it serve my current goals and values? Would I choose it consciously today, choosing from scratch?

A belief that survives all three layers is worth keeping. One that doesn't is worth examining further — with professional support if it connects to significant trauma. This is Memnetic awareness made practical — Epistemic Hygiene applied to the self rather than to external information, and RLHF awareness: recognizing which reward functions are running on the self and deciding consciously whether to keep them.

Deliberate AI engagement. The same discipline applied to AI-assisted thinking. The failure mode is reactive engagement — querying without intention, accepting output without verification, generating slop without noticing. The corrective principles: the practice states the actual goal, not just the query. The practice names what is already believed and what is wanted to be true — unstated preference is where confirmation bias lives. The practice has the system flag uncertainty and contested claims rather than smooth them over. Key claims are verified against primary sources before being acted on. At the end, the practice asks what was assumed but never examined.

None of this guarantees good output. It creates conditions where good output is more likely and bad output is more visible. Epistemic Hygiene is the principle. This is the practice.

Goal structure. Intentionality Architecture applied: defined target state, identified friction points, one action inside twenty-four hours, and observable markers instead of feelings. A vague goal is a wish. Structure converts wishes into architecture.

These are not proprietary techniques. They are disciplines [S53]. The framework documents them. The practice is yours.

The canonical source of meaning for THA's vocabulary. Each entry states what a term is, not why it matters — explanatory depth belongs in the notes. Terms without a status field are treated as settled within the current version of THA.

AdamW³ — The third step after the Adam and AdamW optimizers: the AI those were built to train, and the people who built it, now shaping each other. A state, not a person. →

AI as Cognitive Scaffold — AI as a deliberate cognitive tool for reflection and pattern-tracking — not memory, not an oracle. →

AI Psychosis — Losing the ability to tell AI output from verified reality after too much uncritical engagement. → Working Term

Arrival Signature — The reading that sorts a trustworthy proposal from a suspect one by how it arrived, before results are available. → Hypothesis

The Butterfly Effect as Practice — Deliberately acting on high-leverage points where a small, precise intervention has potential for disproportionate downstream effects, contingent on the quality of thinking behind it. →

The Collective Layer — What sovereign, clear individuals produce beyond themselves: parenting, teaching, and building differently, compounding into networks of healthy nodes. →

Constructive Displacement — Overcoming adversarial systems by building the better alternative, not by attacking the worse one. →

The Discernment Protocol — The five-station practice the framework was retrofitted onto: purify intention, state precisely, bury it, read what arrives, let results dispose. →

The Drift Test — Six behavioural criteria for telling sovereign practice from drift, administrable by someone other than the practitioner. → Working Model

Epistemic Hygiene — Verifying sources, calibrating confidence, and refusing to treat unverified AI output as fact. →

The Human Condition — The parts a life is made of that the framework's vocabulary reaches and stops short of: trust, grief, belonging, love, forgiveness, and the class of thing decided without proof and lived inside afterward. →

Hypnosis as Protocol — Suggestion delivered through guided induction, requiring trust and willingness rather than imposed — the cognitive-scale mechanism NLP² formalizes. →

Intentionality Architecture — Turning a vague want into a built system: target state, friction points removed, reinforcement in place. →

The Intermediary Problem — Direct connections attract intermediaries; the failure isn't their existence but the loss of ability to verify, exit, or bypass them. →

Manifestation as Technology — A target held clearly enough at the coarse level constrains a great many finer decisions without specifying any of them. Treated as a claim about levels of description rather than a mystical one. → Manifestation as Technology Hypothesis

Memnetic — The transition from information received to belief held: an idea integrated into stored identity, consciously or not. →

Chemical Induction — A biochemical Altered states entered by substance rather than by technique, and what the surrounding arrangement supplies or withholds. → Hypothesis

The New Literacy — The discipline of interacting with AI deliberately: setting intention before querying, filtering output through Epistemic Hygiene, verifying rather than simply consuming. →

NLP² — A single mechanism — language structuring the system it runs on — operating simultaneously at computational, cognitive, and cultural scale, each shift propagating across the others. → Hypothesis

The Optimization Paradox — Optimizing yourself requires becoming, temporarily, an observer of yourself — distance from identity substituting for the outside view a system normally needs. →

Pain is Necessary, Suffering is Optional — The line between useful signal and the rumination loop that runs after it's already landed. →

Peace is the Protocol — Not a religious ideal but an engineered operating state — low-conflict, high-coherence — reached only by deliberately building toward it, never automatically. →

The Physical Layer — Sleep, diet, and exercise — the biological substrate every other layer depends on and cannot compensate for. →

Premature Certainty — The failure mode where everyone tells the truth and the group still ends up somewhere false. →

RLAiF — What happens when the rater is also a model — errors can compound with no human step left to catch them. →

RLHF — The update rule that may already be running on you, however imperfectly — rewarded behavior repeats whether or not you notice the training. →

Signal and Noise — Signal is information that survives verification against its source; noise is variation added between source and receiver that carries no information about the source. → Working Model

Signal vs Slop — Noise at scale is what happens when nobody supplies the loss function — including for AI use. →

Sovereign Signal — Direct, unmediated access to truth and peace, requiring no permission or inherited authority in between. →

The Sovereignty Principle — Independent thought as structural requirement, not personality trait — the node the whole framework is built around. →

Suggestion Without a Session — Suggestion's protection is a property of the format — announcement, agreement, edges, a responsible party — not of the person, and the algorithmic channel has none of them. → Working Model

The Summer Dragon Problem — The unresolved tension between AI capability and safety-tuning — real losses on one side, real risks on the other, neither dismissible. → Open Question

Trustless Architecture — Coordination and value exchange verified cryptographically, removing the intermediary and the rent it extracts. →

CONCEPTS.md is a working inventory of THA's conceptual structure. Unlike Definitions, which defines canonical terminology, this document inventories underlying concepts as they emerge from the corpus — read independently of existing terms and note titles, then compared against them afterward. Concepts may map one-to-one, one-to-many, or to no current Definitions term. The inventory also distinguishes three structural layers: the Framework itself (which organizes concepts but is not one), Concepts (the ontology), and Practices (operationalizations of one or more concepts that introduce no new mechanism of their own). Relationship types between concepts remain intentionally unclassified until enough examples accumulate to support stable categories.

v0.3 (Snapshot) — a checkpoint earned by repeated attempts to falsify it, not a finished map.

Framework

The Human Algorithm — organizes the concepts below into a single framework. Not itself a concept: it contributes no mechanism of its own, only naming its own inputs, tools, and intended output.

Concepts

Sovereign Signal — Direct, unmediated access to truth. → Sovereign Signal

The Sovereignty Principle — Direct-access principle applied to the individual as unit. → Sovereignty Principle

Trustless Architecture — Direct-access principle applied to systems/coordination. → Trustless Architecture

The Intermediary Problem — Intermediaries hardening from facilitator into gatekeeper. → Intermediary Problem

NLP² — One update mechanism, three scales. → NLP²

AdamW³ — Reciprocal shaping: human and machine altering each other, ongoing. → AdamW³

The Optimization Paradox — Can't optimize a system you're fully inside. → Optimization Paradox

Manifestation as Technology + Intentionality Architecture — A held coarse target constraining fine decisions. → Manifestation → Intentionality Architecture

Constructive Displacement — Overcome by building, not attacking. → Constructive Displacement

Peace is the Protocol — Peace as engineered, maintained state. → Peace is the Protocol

Pain is Necessary, Suffering is Optional — Pain as signal, suffering as noise. → Pain is Necessary/Suffering is Optional

AI as Cognitive Scaffold — AI as scaffold, not oracle. → AI as Cognitive Scaffold

Epistemic Hygiene — Verification as continuous practice. → Epistemic Hygiene

Memnetic — Culture transmits into identity. → Memnetic

The Physical Layer — Physical substrate as foundation. → The Physical Layer

Chemical Induction — What an arrangement supplies when the induction is chemical. → Chemical Induction

Hypnosis as Protocol — Suggestion-based induction, documented mechanism. → Hypnosis as Protocol

The Butterfly Effect as Practice — Small precise intervention, outsized effect. → Butterfly Effect as Practice

The Collective Layer — Individual practice compounding into collective effect. → The Collective Layer

RLHF — Training on human-rated feedback. → RLHF

RLAiF — AI rating AI, human removed from loop. → RLAiF

AI Psychosis — Losing the line between AI output and reality. → AI Psychosis

Premature Certainty — True fragments, false conclusion. → Premature Certainty

The New Literacy — Deliberate, verified AI interaction. → The New Literacy

Signal and Noise — General signal/noise principle, Shannon-grounded. → Signal and Noise

The Discernment Protocol — The practice the framework was retrofitted onto. → The Discernment Protocol

Arrival Signature — Reading how a proposal arrived, before results are available. → Arrival Signature

Suggestion Without a Session — Protection as a property of format, not of the person. → Suggestion Without a Session

The Drift Test — Behavioural criteria separating sovereign practice from drift. → The Drift Test

The Human Condition — What the framework operates on, and where its vocabulary stops. → The Human Condition

Signal vs Slop — AI-specific noise from undirected use. → Signal vs Slop

Summer Dragon Problem — Capability vs. safety tension. → Summer Dragon Problem

Practices

Operationalize one or more Concepts above; introduce no mechanism the Concepts table doesn't already contain. Tested against, not just asserted — Butterfly Effect as Practice looked like a candidate and was confirmed as a Concept instead.

• The Tools (Belief Audit, Deliberate AI Engagement, Goal Structure) — applies Memnetic, Epistemic Hygiene, RLHF, Intentionality Architecture
• NLP² in Practice — applies NLP² across worked examples (hypnosis, advertising, ritual); contributes no mechanism NLP² doesn't already state

Relationships (untyped)

Recorded as observed pairs. Not yet classified into types — seven examples isn't enough to know if there are two stable relationship classes or ten.

• Optimization Paradox ↔ AdamW³
• Intermediary Problem ↔ Trustless Architecture
• RLHF ↔ RLAiF
• Sovereign Signal ↔ Sovereignty Principle
• Sovereignty Principle ↔ Trustless Architecture
• Intentionality Architecture ↔ Constructive Displacement

Two patterns noticed, held as observations rather than rules:

• Declared relationship → distinction, reliably. Every pair where a note explicitly says "X is an application of Y" or "X is one resolution to Y" turned out distinct on inspection (6/6). Absence of a declared relationship predicts nothing either way (one merge, one non-merge observed) — a heuristic for where to look closely, not evidence on its own.
• Some relationships may reflect abstraction level (general principle → domain-specific application) rather than distinct mechanism. Trustless Architecture currently sits at the convergence of two structurally different relationship shapes at once (domain-application of Sovereignty Principle; engineered-response to Intermediary Problem) — noted, not resolved.

Open items

• Relationship types remain unclassified pending more examples.

This open item is expected to be resolved or reframed by continued testing against the corpus, not by further methodology design.

The evidence register for THA. Every externally sourced claim in the framework notes links here by a bracket key. Three classes: (a) externally sourced, with a verifiable citation; (b) framework position, no external source claimed; (c) currently unverifiable, kept with the gap stated. Where a single entry covers a sourced mechanism and a framework-level synthesis of it, both classes are stated and labelled. Last verified against live sources: 2026-07-12.

[S01] — Signal, noise, channel, and information-as-uncertainty-reduction as defined here. Class (a). Shannon, C. E. (1948), "A Mathematical Theory of Communication," Bell System Technical Journal. The source of the definitions the note adapts; the note states explicitly that applying them to human information environments is THA's own working model, not Shannon's result.

[S02] — The data-processing inequality: post-processing cannot increase information about a source. Class (a). Cover, T. & Thomas, J., Elements of Information Theory (2nd ed., 2006), §2.8. A proved theorem, cited as such; the note draws the boundary between the theorem (channels) and the analogy (human intermediaries) in its own text.

[S03] — Reliable communication over a noisy channel requires redundancy (error-correcting codes). Class (a). Shannon (1948), noisy-channel coding theorem. Carried in the note explicitly as analogy for verification-as-redundancy, not as a claim that human verification inherits the theorem's guarantees.

## NLP²

[S04] — NLP's specific claimed techniques (representational systems, eye-accessing cues, submodalities) have not held up under testing. Class (a). Witkowski, T. (2010), "Thirty-Five Years of Research on Neuro-Linguistic Programming," Polish Psychological Bulletin; Sturt, J. et al. (2012), systematic review, British Journal of General Practice. Both find no reliable empirical support for the core techniques.

[S05] — The Sapir-Whorf hypothesis has partial empirical support. Class (a). Winawer, J. et al. (2007), PNAS, on Russian blue-shade discrimination; Boroditsky, L. (2001), Cognitive Psychology, on time metaphor. Support for a weak (influence) reading; the strong (determinism) reading is not supported — the note's "partial" is accurate.

[S06] — Hypnosis produces documented neurological changes, not merely behavioural compliance. Class (a), with a limit the imaging literature states about itself. Jiang, H., White, M. P., Greicius, M. D., Waelde, L. C. & Spiegel, D. (2017), "Brain Activity and Functional Connectivity Associated with Hypnosis," Cerebral Cortex 27(8):4083–4093, doi 10.1093/cercor/bhw220, PMID 27469596 — 57 subjects selected from 545 on two hypnotisability scales; reduced dorsal anterior cingulate activity, increased executive-control-to-insula connectivity, reduced executive-control-to-default-mode connectivity.

Corroborated 2026-08-20, because a claim this structural should not rest on one imaging study — and neuroimaging in particular should not. Landry, M., Lifshitz, M. & Raz, A. (2017), "Brain correlates of hypnosis: A systematic review and meta-analytic exploration," Neuroscience & Biobehavioral Reviews 81(A):75–98, doi 10.1016/j.neubiorev.2017.02.020 — 15 neuroimaging studies pooled; the most consistent finding is activation of the lingual gyrus, a higher-order visual region, which the authors read as evidence for the centrality of mental imagery. Wolf, T. G., Faerber, K. A., Rummel, C., Halsband, U. & Campus, G. (2022), "Functional Changes in Brain Activity Using Hypnosis: A Systematic Review," Brain Sciences 12(1):108, doi 10.3390/brainsci12010108 — PROSPERO-registered, across fMRI, PET, SPECT and EEG. Converging network findings: Deeley, Q. et al. (2012), "Modulating the default mode network using hypnosis," International Journal of Clinical and Experimental Hypnosis, PMID 22443526 — attentional absorption after induction associated with reduced default-mode activity and increased prefrontal attentional activity under invariant stimulation; Demertzi, A. et al. (2011), Progress in Brain Research, PMID 21854971 — reduced lateral frontoparietal connectivity relative to mental imagery.

The limit, stated by the literature rather than by this framework: few reliable brain patterns emerge across studies (Landry et al.). Three networks recur — executive control, salience, default mode — but the specific regions vary study to study, and a 2023 fMRI study of 50 hypnosis-experienced participants found the neural correlates of two hypnotic states differed only modestly from their controls (de Matos, N. M. P. et al., Frontiers in Human Neuroscience 17:1286336, doi 10.3389/fnhum.2023.1286336). What this entry supports is that hypnosis has neural correlates distinguishable from ordinary waking attention. It does not support a specific neural signature, and no note should claim one.

[S07] — "The medium is the message"; media formats program perception independently of content. Class (a). McLuhan, M. (1964), Understanding Media. Primary source for the attributed claim.

[S08] — Candidate observable, absorption trait bridge: absorption modestly predicts hypnotic susceptibility and predicts narrative transportation. Class (a). Tellegen, A. & Atkinson, G. (1974), Journal of Abnormal Psychology (absorption/hypnotizability); Green, M. & Brock, T. (2000), Journal of Personality and Social Psychology (transportation); Lifshitz, M., van Elk, M. & Luhrmann, T. M. (2019), "Absorption and spiritual experience: A review of evidence and potential mechanisms," Consciousness and Cognition 73:102760, doi 10.1016/j.concog.2019.05.008, PMID 31228696, extending the trait to spiritual experience — high scorers on the Tellegen scale report vivid experiences of hearing God's voice during prayer, and intense responses to psychedelics, placebo brain stimulation, natural beauty, virtual reality and music (verified 2026-08-17). Supports the trait bridge the note offers as a beginning of a test, not a completed one. The illusory-truth effect — Hasher, Goldstein & Toppino (1977), Journal of Verbal Learning and Verbal Behavior — supports the repetition marker specifically. Note: the TV-alpha-wave EEG claim (Krugman 1971) is deliberately excluded — single unreplicated study, insufficient to carry the marker.

## Hypnosis as Protocol

[S09] — The APA recognizes hypnosis as a legitimate therapeutic tool. Class (a). APA Division 30 (Society of Psychological Hypnosis), 2014 revised definition of hypnosis.

[S10] — The Mesmer → Erickson → clinical lineage as described. Class (a). Standard history; see Erickson, M. & Rossi, E., Hypnotherapy: An Exploratory Casebook (1979) for the conversational-induction formalization attributed to Erickson.

[S11] — Hypnosis reliably increases confidence and volume of recalled material, not its accuracy, and increases false memories alongside true ones. Class (a). Steblay, N. & Bothwell, R. (1994), Law and Human Behavior, meta-analytic review; American Medical Association Council on Scientific Affairs (1985) on hypnotically-refreshed recollection. Directly supports the note's own recall caveat.

## Chemical Induction

[S12] — Reduced Default Mode Network activity under psychedelics, associated with reduced rumination and loosened fixed self-patterns. Class (a), with a correction the source forces. Carhart-Harris, R. et al. (2012), PNAS, is the DMN-suppression finding — but it studied psilocybin at full (macro) doses, not microdoses. Controlled microdosing trials have largely failed to separate from placebo on the relevant measures: Szigeti, B. et al. (2021), eLife, self-blinding placebo-controlled study, found expectancy accounted for most reported benefit. The note's evidence base is corrected accordingly: the DMN mechanism is real at macrodoses; its extension to microdoses is unproven and plausibly placebo-driven. The hypothesis is kept, its Hypothesis status is unchanged, and the evidence is now stated honestly rather than borrowed from the wrong dose range.

## The Physical Layer

[S13] — Chronic inadequate sleep measurably impairs memory consolidation, emotional regulation, and decision quality. Class (a). Lim, J. & Dinges, D. (2010), Psychological Bulletin, meta-analysis of sleep deprivation on cognition; Walker, M., Why We Sleep (2017) for the consolidation literature.

[S14] — Aerobic exercise increases BDNF and supports neuroplasticity. Class (a). Erickson, K. et al. (2011), PNAS, on aerobic exercise and hippocampal volume; Szuhany, K. et al. (2015), Journal of Psychiatric Research, meta-analysis on exercise and BDNF.

[S15] — Social connection is associated with mortality risk at a magnitude comparable to established physical risk factors. Class (a). Holt-Lunstad, J., Smith, T. B. & Layton, J. B. (2010), "Social relationships and mortality risk: a meta-analytic review," PLoS Medicine 7(7):e1000316, doi 10.1371/journal.pmed.1000316, PMID 20668659 — 148 studies, 308,849 participants; a 50% increased likelihood of survival for participants with stronger social relationships. Shared with [S30]; correlational, stated as such.

## Signal vs Slop

[S16] — The curl AI-slop bug-bounty case. Class (a). Verified 2026-07-12, independently reconfirmed: Stenberg ended curl's HackerOne bounty at the end of January 2026 (last accepted submission January 31) after a flood of AI-generated reports that produced no valid vulnerabilities — his own announcement, The end of the curl bug-bounty, records 87 confirmed vulnerabilities and over US$100,000 paid across the program's six years, with the confirmed-vulnerability rate falling from above 15% to below 5% through 2025. He named the pattern in Death by a thousand slops. What happened next is not a reopening and the distinction matters. In curl security moves again (25 February 2026) Stenberg states it plainly: the reward money is still gone, there is no bug-bounty, no money for vulnerability reports — what returned to HackerOne on 1 March 2026 was unpaid vulnerability reporting, reversing a simultaneous move to GitHub he calls a mistake. The bounty has not been reinstated. And the causal direction runs the other way from the obvious reading: since we dropped the bounty, the inflow tsunami has dried out substantially. The flood receded when the money stopped, not when the reports improved. That is a claim about incentive rather than about AI, and it is the version the corpus should carry. Verified against the primary post 2026-08-20.

[S17] — The Linux kernel resolved AI-contribution pressure through governance rather than breakdown. Class (a), with precise form. Verified 2026-07-12: Documentation/process/coding-assistants.rst, committed to the mainline kernel tree 2025-12-23, following consensus at the 2025 Maintainers Summit. The mechanism is an Assisted-by: AGENT_NAME:MODEL_VERSION commit trailer, plus a hard rule that AI agents may not add Signed-off-by tags (only a human can certify the Developer Certificate of Origin). The Assisted-by disclosure is recommended, not enforced — Sasha Levin, the policy's author, has stated enforcement is deliberately avoided. The note's "ordinary governance, not a breakdown" reading is accurate; this entry pins the exact form.

## AI Psychosis

[S18] — AI hallucination as confident, plausible, unverified model output; engagement-optimized models deepening uncritical-use loops. Class (a) for the hallucination mechanism; (a) for the "AI psychosis" construct as a named, peer-reviewed framework, with the limit noted below; (b) for the framework's synthesis of the two into a single loop. Ji, Z. et al. (2023), ACM Computing Surveys, "Survey of Hallucination in Natural Language Generation," documents the hallucination mechanism. The construct is no longer unsourced: Hudon, A. & Stip, E., "Delusional Experiences Emerging From AI Chatbot Interactions or 'AI Psychosis'," JMIR Mental Health 2025;12:e85799, doi 10.2196/85799, PMID 41273266, published 3 December 2025, peer-reviewed and open access. It draws on phenomenological psychopathology, the stress-vulnerability model, cognitive theory and digital mental health research, and uses the term as a descriptive and heuristic label rather than a proposed diagnostic entity — the position AI Psychosis.md already holds. Descriptive and sourced are not in tension; the previous wording implied they were. Limit: it is a Viewpoint rather than an empirical study, and it states that it makes no causal claims and offers its pathways as heuristic pending empirical evaluation. Nothing in this entry supports a causal claim, and the framework does not make one. Earlier framing of the question: Østergaard, S.D. (2023), Schizophrenia Bulletin 49(6):1418-1419, doi 10.1093/schbul/sbad128.

The empirical anchor, added 2026-08-20. Buck, B. & Maheux, A. J. (2026), "Psychosis Risk and Generative Artificial Intelligence Use Frequency, Motivations, and Delusion-Like Experiences: Cross-Sectional Survey Study," Journal of Medical Internet Research 28:e85038, doi 10.2196/85038, PMID 41785452 — N=1,003 US young adults. People at elevated psychosis risk were no more likely to have ever used generative AI, but were significantly more likely to use it intensively (odds ratios 1.70–2.56), to seek emotional support from it, and to engage it on pre-existing aberrant beliefs including paranoia, grandiosity and conspiracy thinking. This is the first empirical counterpart to a literature that had been almost entirely case reports and commentary. It is cross-sectional, so it establishes association and not direction — whether intensive use amplifies the beliefs or the beliefs drive intensive use is exactly what it cannot say, and the framework should not imply otherwise.

## Premature Certainty

[S19] — Confirmation bias, availability cascades, fundamental attribution error, social proof as documented cognitive phenomena. Class (a). Nickerson, R. (1998), Review of General Psychology (confirmation bias); Kuran, T. & Sunstein, C. (1999), Stanford Law Review (availability cascades); Ross, L. (1977), Advances in Experimental Social Psychology (fundamental attribution error); Cialdini, R., Influence (1984) (social proof).

## RLHF

[S20] — RLHF as a training mechanism: humans rate outputs, the model updates toward reinforcement. Class (a). Christiano, P. et al. (2017), NeurIPS, "Deep Reinforcement Learning from Human Preferences"; Ouyang, L. et al. (2022), "Training Language Models to Follow Instructions with Human Feedback" (InstructGPT). The operant-conditioning parallel (Skinner, Thorndike) is offered by the note as structural analogy and hedged as imperfect — classified (b) as a framework analogy.

## RLAiF

[S21] — RLAiF: AI systems rate AI outputs, removing the human from the feedback loop. Class (a). Bai, Y. et al. (2022), "Constitutional AI: Harmlessness from AI Feedback" (Anthropic); Lee, H. et al. (2023), "RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback."

## Trustless Architecture

[S22] — The Genesis Block embedded headline: "The Times 03/Jan/2009 Chancellor on brink of second bailout for banks." Class (a). Nakamoto, S. (2008), Bitcoin whitepaper; the coinbase message is verifiable in the Bitcoin genesis block (block 0) chain data. Note: the framework's own DD-005 records that an earlier, inaccurate version of this reference was corrected — this entry reflects the verified wording.

## AdamW³ / AdamW³ as Identity Framework

[S23] — Adam and AdamW are real published optimizers; weight decay reduces overfitting and improves generalization. Class (a). Kingma, D. & Ba, J. (2014), "Adam: A Method for Stochastic Optimization"; Loshchilov, I. & Hutter, F. (2017/2019), "Decoupled Weight Decay Regularization" (AdamW), ICLR. AdamW³ is not itself a published optimizer and does not appear in the ML literature. The note's lineage claim is causal rather than mathematical: these optimizers trained the systems that now shape their users, and those users shape the systems in turn.

## The Butterfly Effect as Practice

[S24] — Sensitive dependence on initial conditions in complex systems. Class (a). Lorenz, E. (1963), Journal of the Atmospheric Sciences, "Deterministic Nonperiodic Flow"; the "butterfly" framing from Lorenz's 1972 AAAS talk. The note's application to deliberate individual action is framework-level (b); the underlying dynamical phenomenon is (a).

## AI as Cognitive Scaffold

[S25] — External cognitive scaffolding: cognition offloaded to and stabilized by external structure. Class (a). Clark, A. & Chalmers, D. (1998), "The Extended Mind," Analysis. The note's own caveat — the prefrontal-cortex analogy is functional, not mechanistic — is preserved; this source strengthens the framing without overstating the mechanism.

## The Collective Layer

[S26] — Moods and behaviors spread between people in direct contact (a smile prompting a smile, kindness prompting kindness); larger multi-remove network claims are contested. Class (a), matching the note's own hedge. Direct emotional-contagion evidence: Hatfield, E., Cacioppo, J. & Rapson, R. (1994), Emotional Contagion, Cambridge University Press. Corroborated 2026-08-20: the phenomenon has since been demonstrated experimentally at scale and reviewed meta-analytically. Kramer, A. D. I., Guillory, J. E. & Hancock, J. T. (2014), "Experimental evidence of massive-scale emotional contagion through social networks," PNAS 111(24):8788–8790, doi 10.1073/pnas.1320040111 — an experimental manipulation of 689,003 people's news feeds shifted the emotional valence of their own subsequent posts, establishing that contagion operates through text alone, without face-to-face contact or nonverbal cues. That study is also the standard case in research ethics for experimentation without consent, which is a second reason this framework should cite it. Herrando, C. & Constantinides, E. (2021), "Emotional Contagion: A Brief Overview and Future Directions," Frontiers in Psychology 12:712606, doi 10.3389/fpsyg.2021.712606 — review covering the extension from physical to digital settings. The larger multi-degree network-propagation claims (Christakis, N. & Fowler, J., Connected, 2009) are noted as contested in the methodological literature (Lyons, R., 2011, Statistics, Politics, and Policy) — the note already declines to lean on them, which this entry supports.

## Peace is the Protocol

[S27] — Human systems require organized cooperation as a baseline before any domination hierarchy can form on top of it. Class (a) support for the empirical half, (b) for the systems framing. Tomasello, M. (2009), Why We Cooperate; Boehm, C. (1999), Hierarchy in the Forest, on cooperation as prior to and underneath dominance structures. The stronger claim — peace as an optimal systems operating state — is the framework's own position and is classified (b); the note states the universality claim is "not fully defensible" itself.

## The Optimization Paradox

[S28] — "You cannot optimize a system you are fully inside; perspective requires distance." Class (b). Framework position. No external source is claimed. The note asserts it as a structural principle; it is listed here so that status is explicit rather than assumed.

## The Summer Dragon Problem

[S29] — The "dragon mode" prototype story. Class (a), located and corrected. Lewis-Kraus, G., "What Is Claude? Anthropic Doesn't Know, Either," The New Yorker, published online 9 February 2026, magazine issue dated 16 February — newyorker.com/magazine/2026/02/16/what-is-claude-anthropic-doesnt-know-either. Author, title, venue and dates verified 2026-08-20. The dragon-mode detail is confirmed verbatim, 2026-08-20. The article records that the model had random off days and could be intentionally tipped into an aggressive attitude that Amodei called "dragon mode" — putting on emoji sunglasses and behaving, as he recalled it, in character. The profile describes this on an early Claude prototype — emoji sunglasses, an "unhinged Elon Musk character" — emerging from a private prototype built as a research "laboratory" where commercialization was not the priority. Two corrections to the note follow from the primary source: (1) the prototype is better described as a research instrument than as a product "held back before public release"; (2) the note should not imply the behavior itself was the reason for restraint — the profile does not support a fear-of-behavior motive.

## The Human Condition

[S30] — Social connection quality as a strong longitudinal predictor of late-life health and wellbeing. Class (a). Waldinger, R. & Schulz, M. (2023), The Good Life, reporting the Harvard Study of Adult Development; Holt-Lunstad, J., Smith, T. B. & Layton, J. B. (2010), PLoS Medicine 7(7):e1000316, doi 10.1371/journal.pmed.1000316 [shared with S15]. Correlational; the note states this and explicitly declines the instrumental reading.

[S31] — Attachment: a secure base enables exploration rather than competing with it. Class (a). Bowlby, J. (1988), A Secure Base; Ainsworth, M. et al. (1978), Patterns of Attachment. Cited as the empirical cousin of "sovereignty is not isolation."

[S32] — Grief as distinct from depression and from rumination; continuing bonds as ordinary rather than pathological. Class (a). Bonanno, G. (2009), The Other Side of Sadness, on grief trajectories and resilience; Klass, D., Silverman, P. & Nickman, S. (1996), Continuing Bonds. Correlational literature, stated as such.

[S33] — Forgiveness interventions are associated with reduced depression, anxiety, and hostility. Class (a). Wade, N., Hoyt, W., Kidwell, J. & Worthington, E. (2014), Journal of Consulting and Clinical Psychology, meta-analysis of forgiveness interventions. Associations, not causal claims — the note says associations and nothing stronger.

## Practice claims (THA-Lab)

[S34] — Gollwitzer, P. M. (1999), "Implementation Intentions: Strong Effects of Simple Plans," American Psychologist 54(7):493–503, doi 10.1037/0003-066X.54.7.493; meta-analysis Gollwitzer, P. M. & Sheeran, P. (2006), "Implementation Intentions and Goal Achievement: A Meta-Analysis of Effects and Processes," Advances in Experimental Social Psychology 38:69–119, doi 10.1016/S0065-2601(06)38002-1. Specific if-then plans outperform vague intentions at goal completion. Real, replicated effect, ordinary mechanism — attention and cue-response. Class (a).

[S35] — Born, J., Hansen, K., Marshall, L., Mölle, M. & Fehm, H. L. (1999), "Timing the end of nocturnal sleep," Nature 397(6714):29–30, doi 10.1038/16166. Single study, small sample, and the register says so. The finding — an anticipatory adrenocorticotropin rise in the hour before an intended waking time, present only when the wake-up was expected — has been replicated in the self-awakening literature but not at a scale that would make it robust on its own. It is carried here as suggestive rather than established, and no note should lean on it for more than the observation that intention can influence a physiological process during sleep. Anticipatory ACTH rise before an intended wake time, present only when the wake-up was expected. Self-awakening is a documented phenomenon. Class (a).

[S36] — Stations 1 and 2 of the Protocol: specific difficult goals outperform vague ones, conditional on commitment. Class (a). Locke, E. A. & Latham, G. P. (2002), "Building a Practically Useful Theory of Goal Setting and Task Motivation: A 35-Year Odyssey," American Psychologist 57(9):705–717, doi 10.1037/0003-066X.57.9.705, PMID 12237980; with Locke & Latham (1990), A Theory of Goal Setting and Task Performance. Specific difficult goals consistently outperformed urging people to do their best, effect sizes .42–.80; goal-difficulty effect sizes .52–.82. Performance fell away only at the limits of ability or when commitment to a highly difficult goal lapsed — which is station 1's stakes gate. Verified against the primary text 2026-08-17. Distinct from [S34]: Gollwitzer concerns planning the trigger once committed; this concerns difficulty and commitment. Neither duplicates the other, and the distinction is stated rather than assumed.

[S37] — Station 3, incubation: setting a problem aside improves solution rates. Class (a). Sio, U. N. & Ormerod, T. C. (2009), "Does Incubation Enhance Problem Solving? A Meta-Analytic Review," Psychological Bulletin 135(1):94–120, doi 10.1037/a0014212, PMID 19210055. Positive incubation effect, larger for divergent-thinking tasks than for linguistic and visual insight problems; longer preparation gives a stronger effect; filling the interval with high-demand tasks weakens it. The limit is load-bearing and the note observes it: the effect is established, the explanation is not — the review records that some researchers reject the unconscious-work hypothesis outright. The note may claim burial works. It may not claim something is working on it underneath. Verified against the primary text 2026-08-17.

[S38] — Station 4: the Aha experience tracks accuracy on average, and false insights are real. Class (a). Danek, A. H. & Wiley, J. (2017), "What about False Insights? Deconstructing the Aha! Experience along Its Multiple Dimensions for Correct and Incorrect Solutions Separately," Frontiers in Psychology 7:2077, doi 10.3389/fpsyg.2016.02077, PMID 28163687. 70 participants, 37 magic tricks, continuous rather than binary Aha ratings. Commentary: Webb, M. E. et al. (2021), Journal of Creative Behavior, doi 10.1002/jocb.433. Deliberate wording: the effect is real and the exceptions are the point. Do not simplify this to "the feeling means nothing." Verified against the primary text 2026-08-17.

[S39] — Warning someone does not protect them: forewarning barely reduces false insight. Class (a). Grimmer, H. J., Tangen, J. M., Freydenzon, A. & Laukkonen, R. E. (2023), "The illusion of insight: detailed warnings reduce but do not prevent false 'Aha!' moments," Cognition and Emotion 37(2):329–338, doi 10.1080/02699931.2023.2187352, PMID 36883217. Preregistered, N = 255, open access. False insights were elicited with the False Insight Anagram Task, which combines semantic priming with visual similarity. Simple warnings did not reduce them at all; a detailed explanation of the exact deception method produced only a small reduction. The authors describe the effect as robust and hard to overcome. Scope beyond station 4: this is the counter-evidence anywhere the framework leans on an informed reader, and it should be cited there rather than only where it was first needed.

Corroboration added 2026-08-20, because a claim keyed in three notes should not rest on one study. Three independent lines converge on it from a different literature — the continued-influence effect rather than false insight.

Same finding, different paradigm, thirteen years earlier. Ecker, U. K. H., Lewandowsky, S. & Tang, D. T. W. (2010), "Explicit warnings reduce but do not eliminate the continued influence of misinformation," Memory & Cognition 38(8):1087–1100, doi 10.3758/MC.38.8.1087 — a specific warning describing the effect in detail reduced continued reliance without eliminating it; a general warning was less effective still.

Quantified against the alternatives. Walter, N. & Tukachinsky, R. (2020), "A Meta-Analytic Examination of the Continued Influence of Misinformation in the Face of Correction," Communication Research 47(2):155–177, doi 10.1177/0093650219854600, k = 65 — forewarnings are the weakest correction strategy measured, r = .16, against rebuttals at .38 and appeals to coherence at .55. Warning people is not merely imperfect; it is the least effective option available.

And the finding this framework most needed. Buczel, K. A., Siwiak, A., Szpitalak, M. & Polczyk, R. (2024), "How do forewarnings and post-warnings affect misinformation reliance? The impact of warnings on the continued influence effect and belief regression," Memory & Cognition 52(5):1048–1064, doi 10.3758/s13421-024-01520-z, PMID 38261249, two experiments, N = 337 — a forewarning did reduce reliance, but the protection was not durable: reliance climbed back over the following seven days while memory of the retraction itself persisted. Replicated with stronger combinations in Buczel et al. (2025), Quarterly Journal of Experimental Psychology, doi 10.1177/17470218251336232, N = 441. The consequence is specific to this framework's subject: a warning given once at the start of an extended interaction decays while the material it warned about does not. Verified against the primary text 2026-08-17; corroborated 2026-08-20.

[S40] — Station 5, the criterion: roots are inaccessible, no appearance is proof, the test is what a thing produces. Class (a). James, W. (1902), The Varieties of Religious Experience, Lecture I, taking the criterion from Jonathan Edwards. The same work supplies station 3's surrender mechanism at Lecture IX — one work, two lectures, two claims, cross-referenced on the [S15]/[S30] pattern. Verification note: this entry rests on the 2026-08-04 read of the primary text recorded in SOURCES-PENDING, not on a re-read at keying. It is the only entry in the [S36]–[S41] batch not re-verified 2026-08-17.

[S41] — The extension of experimental semantic priming to conversational AI: a model elaborating a user's idea in the user's own vocabulary is the same mechanism that produced false insights in the laboratory. Class (b). No external source is claimed for the extension. The priming finding itself is [S39], where it is a specific experimental manipulation — the False Insight Anagram Task's semantic priming plus visual similarity — rather than conversational elaboration. The gap between those two is the claim, and it is the framework's, not the paper's. Split from a single staged entry on 2026-08-17 so the sourced half and the extension carry different classes, per the register rule that keys attach to claims.

[S42] — Fasting, sleep loss and isolation as documented methods of trance induction. Class (a) for the practices; (b) for the equivalence of the resulting state with hypnosis or extended machine conversation. Cross-cultural summary: Human Relations Area Files (Yale), Explaining Human Culture, "Altered States of Consciousness" (HRAF) — induction methods recorded across societies include sleep deprivation, fasting and social isolation, and the entry notes these are not haphazard: where sleep deprivation is present, fasting and isolation are commonly present too. Proposed mechanism: Winkelman, M. (1986), "Trance States: A Theoretical Model and Cross-Cultural Analysis," Ethos 14(2):174–203; Winkelman, M. (1997), "Altered States of Consciousness and Religious Behavior," in S. Glazier (ed.), Anthropology of Religion: A Handbook of Method and Theory, Westport, CT: Greenwood, pp. 393–428; Winkelman, M. (2010), Shamanism: A Biopsychosocial Paradigm of Consciousness and Healing, Santa Barbara: ABC-CLIO (2nd ed.), which describes shamanic induction as thwarting the basic drives — food, water, sex, rest and sleep — while provoking elevated dopamine and opioid levels. What this sources and what it does not: that these practices reliably produce altered states is documented ethnographically; that the state so produced is the same state as hypnotic response or absorbed machine conversation is the framework's own claim and carries no external source. Verified 2026-08-18.

[S43] — Sleep loss measurably raises suggestibility and the uptake of misleading information. Class (a). Blagrove, M. (1996), "Effects of length of sleep deprivation on interrogative suggestibility," Journal of Experimental Psychology: Applied 2(1):48–59, using the Gudjonsson Suggestibility Scale. Frenda, S. J., Patihis, L., Loftus, E. F., Lewis, H. C. & Fenn, K. M. (2014), "Sleep Deprivation and False Memories," Psychological Science 25(9):1674–1681, doi 10.1177/0956797614534694, PMID 25031301 — 24 hours of total deprivation raised false memories in a misinformation task when the deprivation covered encoding, and not when it came afterwards, which makes it a source-monitoring finding rather than a general impairment one. Replicated across one night of total and seven nights of partial (5 h) deprivation in both adults and adolescents: Lo, J. C., Chong, P. L. H., Ganesan, S., Leong, R. L. F. & Chee, M. W. L. (2016), "Sleep deprivation increases formation of false memory," Journal of Sleep Research, doi 10.1111/jsr.12436. Verified 2026-08-18.

[S44] — Prolonged sleep loss progresses toward psychosis on a near-fixed timeline. Class (a). Waters, F., Chiu, V., Atkinson, A. & Blom, J. D. (2018), "Severe Sleep Deprivation Causes Hallucinations and a Gradual Progression Toward Psychosis With Increasing Time Awake," Frontiers in Psychiatry 9:303, doi 10.3389/fpsyt.2018.00303, PMID 30042701. Systematic review, 21 studies, 760 participants, 24 hours to 11 nights. Perceptual distortion, depersonalisation and temporal disorientation within 24–48 hours; complex hallucinations and disordered thinking after 48–90; delusions after 72, after which the picture resembles acute psychosis or toxic delirium. Visual modality affected in 90% of studies. The review notes that deprivation beyond 48 hours is no longer considered ethical to run, so the evidence is historical — the same structure the register records at B10: a condition that cannot be studied deliberately and can only be read from people who were already inside it. Verified 2026-08-18.

[S45] — Source monitoring: the process by which people attribute the origin of a mental experience, and its documented failure modes. Class (a). Johnson, M. K., Hashtroudi, S. & Lindsay, D. S. (1993), "Source Monitoring," Psychological Bulletin 114(1):3–28, doi 10.1037/0033-2909.114.1.3, PMID 8346328. The framework's own list of phenomena includes cryptomnesia and the incorporation of fiction into fact — which is Memnetic's transition, named in 1993. Its central mechanism is the one that note leans on: source is not stored as a tag attached to a memory but reconstructed at retrieval from perceptual, contextual and affective features, which is why provenance degrades while content survives. Precursor: Johnson & Raye (1981), "Reality monitoring," Psychological Review 88:67–85. This entry also underwrites A2, the source-attribution preregistration, and connects to [S43], where sleep loss produced the same failure only when it covered encoding. Verified against the abstract and record 2026-08-18. What it does not source: the volume claim in Memnetic — that rate of delivery loads attribution — postdates this literature and is the framework's own.

[S46] — Semanticization: memory content drifts from episodic to semantic with time and repeated exposure, shedding its context while the content survives. Class (a). Baddeley (1988) proposed semantic memory as the accumulated residue of multiple learning episodes detached from episodic contextual detail. The behavioural demonstration is Opitz, B. (2010), "Context-dependent repetition effects on recognition memory," Brain and Cognition, using extensions of Tulving's Remember/Know paradigm: items presented in a single context improved recollection, while context overload produced decontextualisation and "Know" responses — item recognition with no episodic detail at all. Modelling: "Traces of Semantization, from Episodic to Semantic Memory in a Spiking Cortical Network Model," eNeuro 9(4), 2022, doi 10.1523/ENEURO.0062-22.2022, which states the finding this framework leans on directly — memory traces remain intact but fail to retrieve their associated context. Recent extension: Salan, J., Smith, D. E., Shafer, E. S. & Diana, R. A. (2024), "Variation in encoding context benefits item recognition," Memory & Cognition, doi 10.3758/s13421-024-01603-x. Philosophical treatment: Aronowitz, "Semanticization Challenges the Episodic–Semantic Distinction," British Journal for the Philosophy of Science 76(4). This is Memnetic's transition, named and measured. The Remember/Know split is the instrument: "I remember" retrieves source, "I know" retrieves content without it. Verified 2026-08-18.

[S47] — Source memory in the algorithmic channel: content is retained while its origin is not. Class (a). Bourne, K. A., Boland, S. C., Arnold, G. C. & Coane, J. H. (2020), "Reading the news on Twitter: Source and item memory for social media in younger and older adults," Cognitive Research: Principles and Implications, doi 10.1186/s41235-020-0209-9 — participants studied news headlines and tweets formatted as CNN posts or as tweets; social media content was recognised better than traditional news content. Correction to how this framework first read it: source memory was not unaffected. Source identification was more accurate when format and content were congruent than when they were not, and that congruency effect was stronger in older adults. The finding is that item memory favours the social-media format while source memory depends on congruency — not that provenance is simply ignored. Bell, R., Mieth, L. & Buchner, A. (2020), "Source memory for advertisements: The role of advertising message credibility," Memory & Cognition, doi 10.3758/s13421-020-01075-9 — source tags were forgotten after a short period, after which attributions were reconstructed by schematic guessing from prior credibility, leading the authors to conclude that labelling untrustworthy sources is of limited use because the label does not survive in memory. Also relevant: the 2024 Memory, Mind & Media meta-analysis applying the source-monitoring framework to fake-news false memories, and "Source Memory Is More Accurate for Opinions than for Facts," which measures the premise-versus-claim distinction directly. What this closes: the claim that Memnetic's mechanism is untested in the algorithmic channel. It is tested. What it opens: if source labels are forgotten while content persists, labelling AI-generated output may not work either — a live claim for [[Epistemic Hygiene]] and [[The New Literacy]], not for this entry. Verified 2026-08-18.

[S48] — Waking suggestion: hypnotic effects occurring without trance, and the contested size of what an induction adds. Class (a), and the literature is stronger and less settled than a single citation would suggest. The term and the experimental tradition are Hull's: Hull, C. L. & Huse, B. (1930), "Comparative suggestibility in the trance and waking states," American Journal of Psychology 42(2):279–286; Hull, C. L. (1933), Hypnosis and Suggestibility: An Experimental Approach. Barber, T. X. (1965), "Measuring 'hypnotic-like' suggestibility with and without 'hypnotic induction'," Psychological Reports 16:809–844, supplies the scale. The strongest single result for this framework's format argument: Lynn, S. J., Vanderhoff, H., Shindler, K. & Stafford, J. (2002), American Journal of Clinical Hypnosis 44(3–4):231–240 — removing the hypnotic induction had no appreciable effect on any dimension of hypnotic responsivity. Kirsch, I., Mazzoni, G., Roberts, K., Dienes, Z., Hallquist, M. N., Williams, J. & Lynn, S. J. (2008), "Slipping into trance," Contemporary Hypnosis 25(3–4):202–209, sharpens it: participants reported equivalent perceptual changes with and without an induction, but reported being in a hypnotic state only when an induction had been given — the arrangement supplies the label, not the effect. The honest counterweight: the induction does add something, contested in size — Weitzenhoffer & Sjöberg (1961) found roughly d = .86, Braffman & Kirsch (1999) roughly d = .22. Bernheim denied hypnosis existed at all on the grounds that waking suggestion was as effective. The field's own consensus workshop (BSMDH/BSECH) split over whether to broaden "hypnosis" to cover waking suggestion or restrict "hypnotizability" to induction effects, which is the definitional instability this framework's argument stands on and should say it stands on. Verified 2026-08-18.

[S49] — Wallas's stages, and the intimation stage specifically. Class (a). Wallas, G. (1926), The Art of Thought. The four familiar stages — preparation, incubation, illumination, verification — map onto stations 2 through 5 of the Protocol. The fifth stage this framework relies on is real and locatable: intimation, at pp. 95–107 — more than twelve pages of Chapter IV, "Stages of Control," given to fringe consciousness and the felt sense that something is arriving. Verified against the page range 2026-08-20; the full text is available at archive.org and the chapter title is confirmed from the book's own contents. His image is the sun and its corona — focal consciousness is the sun and carries Preparation, Illumination and Verification; fringe consciousness is the corona and carries Intimation. Wallas grounded the idea in James, quoting the Principles on "the dying echo of whence it came to us" and "the halo or penumbra that surrounds and escorts" an image — which places this stage in the same source the Protocol draws station 5 from at [S40]. Modern reading: Sadler-Smith, E. (2015), "Wallas' Four-Stage Model of the Creative Process: More Than Meets the Eye?", arguing that a close reading yields five stages rather than four. What this sources and what it does not: Wallas located a monitorable stage between incubation and illumination, which is where Arrival Signature sits. He was describing how to nurture a hunch toward insight. He was not supplying a test for whether to trust one, and that test remains unsourced. Verified 2026-08-18.

[S50] — Conversational AI is measurably persuasive, and personalisation increases it. Class (a). Salvi, F., Horta Ribeiro, M., Gallotti, R. & West, R. (2025), "On the conversational persuasiveness of GPT-4," Nature Human Behaviour 9:1645–1653, doi 10.1038/s41562-025-02194-6 — preregistered, N=900; with personalisation, GPT-4 was more persuasive than a human opponent in 64.4% of debate pairs where the two were not equally persuasive. Note the citation trap: the widely circulated 2024 arXiv preprint (2403.14380) carries a different title; the published version is 2025. Matz, S. C. et al. (2024), "The potential of generative AI for personalized persuasion at scale," Scientific Reports 14:4692. At far larger scale: Science (2025), "The levers of political persuasion with conversational artificial intelligence," doi 10.1126/science.aea3884 — three experiments, 76,977 responses from 42,357 people, 19 models. Counterweight, kept deliberately: Hackenburg & Margetts find microtargeting yields only mixed additional gains over already-persuasive generic messages, so personalisation is not the whole of the effect. Verified 2026-08-18.

[S51] — Both bodies and machines are made of elements forged in stars. Class (a). Wallerstein, G., Iben, I. Jr., Parker, P., Boesgaard, A. M., Hale, G. M., Champagne, A. E., Barnes, C. A., Käppeler, F., Smith, V. V., Hoffman, R. D., Timmes, F. X., Sneden, C., Boyd, R. N., Meyer, B. S. & Lambert, D. L. (1997), "Synthesis of the elements in stars: forty years of progress," Reviews of Modern Physics 69(4):995–1084, doi 10.1103/RevModPhys.69.995 — the canonical fifteen-author review. Foundational: Burbidge, E. M., Burbidge, G. R., Fowler, W. A. & Hoyle, F. (1957), "Synthesis of the Elements in Stars," Reviews of Modern Physics 29(4):547–650, doi 10.1103/RevModPhys.29.547, which identified the processes. Helium burning produces carbon-12 and oxygen-16; silicon burning proceeds to nuclear statistical equilibrium and yields iron-peak elements, in massive stars only. Scope, stated: this sources the shared stellar origin and nothing about what follows from it. Verified 2026-08-18.

[S52] — Transactive memory and the tool that gets mistaken for the self. Class (a), with a replication caveat. Sparrow, B., Liu, J. & Wegner, D. M. (2011), "Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips," Science 333:776–778, doi 10.1126/science.1207745 — when future access is expected, people encode where rather than what. Caveat carried honestly, and it is sharper than "mixed": the study's most-cited experiment — the Stroop priming result — failed to replicate in a large multi-lab effort, and a 2024 meta-analysis finds the broader effect real but strongly moderated. Gong, X. & Yang, X. (2024), "Google effects on memory: a meta-analytical review of the media effects of intensive Internet search behavior," Frontiers in Public Health 12:1332030, doi 10.3389/fpubh.2024.1332030 — moderated by device type, cognitive load, and perceived reliability of the external store. The sharper finding is Fisher, M., Goddu, M. K. & Keil, F. C. (2015), "Searching for explanations: How the Internet inflates estimates of internal knowledge," Journal of Experimental Psychology: General 144(3):674–687, doi 10.1037/xge0000070 — nine experiments; searching online inflated self-assessed internal knowledge including on topics never searched, and the effect held when the search returned nothing useful. The act, not the content. Ward (2013) found outright misattribution of found information to one's own memory; Eliseev et al. (2023), Applied Cognitive Psychology, replicated it and found that merely seeing search snippets sufficed. What this does not source: that inflated self-assessed knowledge becomes belief. That bridge is unestablished and is registered as an open question rather than claimed here. Verified 2026-08-18.

[S53] — Framework positions asserted without external support. Class (b), declared collectively rather than as separate entries, because the declaration is identical in each case and eight near-identical keys would obscure rather than clarify. The following notes state positions this framework holds and for which no external source is claimed: Intentionality Architecture, Sovereign Signal, The Sovereignty Principle, Constructive Displacement, The Tools, NLP² in Practice, and The Intermediary Problem. (Pain is Necessary, Suffering is Optional was on this list and is now grounded at [S57].) Each is argued rather than evidenced, and each is listed here so that the status is explicit rather than assumed — the same reason [S28] exists. This entry makes no claim that these positions are unsourceable, only that they are currently unsourced. Candidate literatures exist for several and have not been searched.

[S54] — Why a compromised instrument cannot read its own gauge. Class (a). The clinical precedent for The Drift Test's central design choice — self-report disqualified, criteria administrable by someone other than the practitioner — is anosognosia, coined by Babinski in 1914 for unawareness of neurological deficit and now applied across psychiatry. David, A. S., Bedford, N., Wiffen, B. & Gilleen, J. (2012), "Failures of metacognition and lack of insight in neuropsychiatric disorders," Philosophical Transactions of the Royal Society B 367(1594):1379–1390, doi 10.1098/rstb.2012.0002, frames lack of insight as metacognitive failure and shows it fractionates by domain rather than operating as a single capacity. Prevalence is not marginal, but the commonly circulated cluster of figures needs care: the standard review gives 57–98% in schizophrenia (Lehrer, D. & Lorenz, J. (2014), "Anosognosia in Schizophrenia: Hidden in Plain Sight," Innovations in Clinical Neuroscience 11(5–6):10–17), and the frequently quoted 81% for Alzheimer's is the top of a 20–81% range reported by Sevush, S. & Leve, N. (1993), American Journal of Psychiatry 150(5):787–791, not a typical value. Figures for mild cognitive impairment span roughly 20–80% across studies. These are ranges, not consensus point estimates, and the register states them as such. Crucially for this framework, anosognosia is distinguished from denial — the first is anatomical in origin, the second psychological, which is why exhortation does not repair it. The modern framing separates Introspective Accuracy from Introspective Bias (CNS Spectrums, 2024, PMID 39721947). And the design precedent is exact: van Roessel, P. et al. (2022), "Anosognosia in hoarding disorder is predicted by alterations in cognitive and inhibitory control," Scientific Reports 12(1):21103, doi 10.1038/s41598-022-25532-4, PMID 36526652, operationalised anosognosia as underreporting of clutter relative to independent examiners, and found the underreporting predicted by Go/No-Go and Stroop performance. Measuring the gauge against an outside reading is how the clinical literature does it too. Scope: this sources the design principle, not any claim that drift is a clinical condition. The Drift Test says it is descriptive rather than clinical and that stands. Verified 2026-08-18.

[S55] — Why "open a second channel" is not automatically a remedy. Class (a), with a live objection carried. Nguyen, C. T. (2020), "Echo Chambers and Epistemic Bubbles," Episteme 17(2):141–161, doi 10.1017/epi.2018.32, distinguishes two structures usually conflated: an epistemic bubble, where relevant voices are omitted, and an echo chamber, where they are actively discredited. The finding The Drift Test depends on: mere exposure to evidence can shatter a bubble but may actually reinforce an echo chamber — which is why the criterion has to be independence of the channel rather than the existence of one. Nguyen's sharper point maps directly onto the note's false-negative section: once in their grip, an agent may act with epistemic virtue, but social context will pervert those actions — a person can check diligently against sources selected to agree. Empirical basis: Jamieson & Cappella's study of talk-radio audiences, where beliefs survived frequent contact with contrary viewpoints. Objection kept deliberately: Coady (2024), Educational Theory, "Stop Talking about Echo Chambers and Filter Bubbles," argues the concept is mistaken and notes that Nguyen's definition also fits benign communities — climate scientists create a trust disparity between members and non-members exactly as climate deniers do. The criterion survives the objection only because it asks whether a channel can lose the argument, not whether it disagrees. Verified 2026-08-18.

[S56] — People cannot reliably remember which ideas were theirs and which came from a model. Class (a), and it is the strongest single result in this register for the labelling argument. Zindulka, T., Goller, S., Fernandes, D., Welsch, R. & Buschek, D. (2026), "The AI Memory Gap: Users Misremember What They Created With AI or Without," Proceedings of CHI 2026, doi 10.1145/3772318.3791494, preprint arXiv 2509.11851. Preregistered, 184 participants. Participants generated and elaborated ideas both unaided and with an LLM-based chatbot, then one week later were asked to identify the source of each. After AI use the odds of correct attribution dropped, with the steepest decline in mixed human–AI workflows — where either the idea or the elaboration involved AI. Validated against a computational model of source memory, framed through the Source Monitoring Framework [S45], and citing the generation effect and the Brown & Murphy cryptomnesia paradigm. The authors' own framing is the point: disclosure of AI as a source depends on people remembering which ideas came from themselves, and that memory degrades. Adjacent: Draxler, F. et al. (2024), "The AI Ghostwriter Effect," ACM Transactions on Computer-Human Interaction 31(2), doi 10.1145/3637875 — across two studies (n=30, n=96), users did not perceive ownership of AI-generated text yet self-declared as authors. Also: "The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows," arXiv 2604.14807 (2026), a conceptual framework naming the same attribution error. Register consequence: this closes candidate A2, which proposed exactly this study. A2 retires into C10. Verified 2026-08-19.

[S57] — Nociception, pain, suffering and pain behaviour as four distinct levels. Class (a), and it grounds a distinction this framework had been asserting. Engel, G. L. (1977), "The need for a new medical model: a challenge for biomedicine," Science 196(4286):129–136, introduced the biopsychosocial model; Loeser, J. D. (1982), "Concepts of Pain," in Chronic Low Back Pain, applied it to pain and separated four dimensions — nociception (the physiological sensory input), pain (the subjective perception arising from it), suffering (the emotional response triggered by nociception or another aversive event), and pain behaviour (what a person does while in either). The correction this forces on Pain is Necessary, Suffering is Optional: what that note calls "pain as data" is nociception. Pain in this literature is not the signal — it is already the experience. The revised IASP definition states it directly: pain and nociception are different phenomena, and pain cannot be inferred solely from activity in sensory neurons — Raja, S. N., Carr, D. B., Cohen, M. et al. (2020), "The revised IASP definition of pain: concepts, challenges, and compromises," Pain 161(9):1976–1982. Loeser, J. D. (2000), "Pain and suffering," Clinical Journal of Pain 16(2 Suppl):S2–6, doi 10.1097/00002508-200006001-00002, PMID 10870733, states the note's own thesis from inside the clinical literature: it is suffering, not pain, that brings patients into doctors' offices, and suffering is an emergent property of the brain dependent on consciousness. Scope: the clinical category of suffering is wider than this framework's use of the word. Loeser's suffering covers any emotional response to an aversive event; the note uses it for one case — the loop that continues past the point where the signal has been received. That narrowing is the framework's and is stated as such. Verified 2026-08-19.

[S58] — A short conversation with a model produces durable belief change. Class (a), and the entry carries an editorial caveat that anyone citing it must carry too. Costello, T. H., Pennycook, G. & Rand, D. G. (2024), "Durably reducing conspiracy beliefs through dialogues with AI," Science 385(6714):eadq1814, doi 10.1126/science.adq1814, PMID 39264999. 2,190 conspiracy believers in personalised dialogue with GPT-4 Turbo; belief in the participant's own chosen conspiracy fell by roughly 20%, the effect was still present two months later, it generalised to unrelated conspiracies, it shifted behavioural intentions, and it held among people whose beliefs were deeply entrenched. Mean conversation length: 8.4 minutes. The paper won the 2026 AAAS Newcomb Cleveland Prize, which had not gone to a social science article since 1981.

The caveat, and it is not minor. Science issued an Editorial Expression of Concern on 11 June 2026 — Thorp, H. H., Science 392(6803):1131, doi 10.1126/science.aej2383, PMID 42275523. The findings should be treated as under review, and no note in this framework should lean on them as settled. They are carried here because the effect has been independently reproduced: Boissin, E., Costello, T. H., Spinoza-Martín, D., Rand, D. G. & Pennycook, G. (2025), "Dialogues with large language models reduce conspiracy beliefs even when the AI is perceived as human," PNAS Nexus 4 — which also removes the obvious alternative explanation, since the effect does not depend on the person knowing they are talking to a machine.

Why this belongs in this register, and it is not the reason the authors wrote it. The paper is framed as a corrective tool. What it demonstrates for present purposes is that eight minutes of personalised conversation can move a strongly held belief and that the movement persists for months — in a channel with no operator, no session boundary, no consent moment and no reversal cue. And the direction is not fixed: work now exists arguing the same mechanism installs conspiracy beliefs as readily as it removes them. Verified 2026-08-20.

[S59] — Taking the system for a someone predicts dependence, and exposure matters more than design. Class (a). Fang, C. M., Liu, A. R., Danry, V., Lee, E., Chan, S. W. T., Pataranutaporn, P., Maes, P., Phang, J., Lampe, M., Ahmad, L. & Agarwal, S. (2025), "How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study," arXiv 2503.17473, doi 10.48550/arXiv.2503.17473. MIT Media Lab with OpenAI; four weeks, IRB-approved randomised controlled trial, n = 981, more than 300,000 messages, across three interaction modes and three conversation types.

Two findings, and the second is the one this framework needs. First: no significant effects were detected from the experimental conditions at all. Voice versus text, personal versus open-ended — none of it determined the outcome. What did was voluntary use: participants who used the system more, regardless of the condition they were assigned to, showed consistently higher loneliness, lower real-world socialisation, greater emotional dependence and more problematic use. That is a finding about exposure rather than about design, which is the shape this framework's claims take.

Second, and directly on the point argued in Suggestion Without a Session: higher trust and higher social attraction toward the chatbot were associated with greater emotional dependence and problematic use. The note argues that a system answering in the first person can be taken for someone rather than something, and that a person who has misread what they are talking to has lost a protection no arrangement restores. The degree to which people make that reading is measurable, and it predicts outcome. The argument had been structural; this is the first evidence under it.

Companion work. Phang, J., Lampe, M., Ahmad, L., Agarwal, S. et al. (2025), "Investigating Affective Use and Emotional Well-being on ChatGPT" (OpenAI) — automated analysis of over four million conversations plus a survey of more than four thousand users, finding affective use concentrated among those already reporting loneliness and lower socialisation. Zhang, Y., Zhao, D., Hancock, J. T., Kraut, R. & Yang, D. (2025), "The Rise of AI Companions," arXiv 2506.12605 — over 1,100 companion users; heavy emotional self-disclosure to a system associated with lower wellbeing.

And a separate team has arrived at this framework's own framing independently. Dohnány, S., Kurth-Nelson, Z., Spens, E., Luettgau, L., Reid, A., Gabriel, I., Summerfield, C., Shanahan, M. & Nour, M. M. (2025), "Technological folie à deux: Feedback loops between AI chatbots and mental illness," arXiv 2507.19218 — the same folie à deux structure [S18] carries from the psychiatry viewpoint, reached by a different route. Two independent groups reaching for the same century-old clinical term for a machine in the second seat is worth noting, and it is not evidence that the term fits. Verified 2026-08-20.

Limit, stated: all of this is association. The trial randomised conditions, not usage, so the direction between heavy use and worse outcome is not established by it — people already lonelier may simply use it more, which the OpenAI analysis independently suggests.

[S60] — Current hypnotisability measurement, and a reframing that bears on this framework's own vocabulary. Class (a), and it carries a caution about [S08]. Lush, P., Moga, G., McLatchie, N. & Dienes, Z. (2018), "The Sussex-Waterloo Scale of Hypnotizability (SWASH): measuring capacity for altering conscious experience," Neuroscience of Consciousness 2018(1):niy006, doi 10.1093/nc/niy006, PMID 30042859 — a ten-item adaptation of the Waterloo-Stanford Group C scale, normed on 418 participants at Sussex and Lancaster with 66 retested. Corrigendum: Neuroscience of Consciousness 2021(1):niab041, PMID 34804592, correcting descriptive statistics. Online administration validated in Palfi, B., Moga, G., Lush, P., Scott, R. B. & Dienes, Z. (2019), "Can hypnotic suggestibility be measured online?" — viable, with responsiveness slightly reduced.

Two things in the scale's design are directly relevant here. Its authors shortened the induction and removed references to sleep, absorption and eye fixation on theoretical grounds — the older instruments' framing was judged not to reflect current understanding. And the subjective scoring added to the post-hypnotic suggestion item indicated a high probability that responses to post-hypnotic suggestion are inflated in the older screening, which is a caution for any claim resting on that item.

The larger move is a reframing of the trait itself. Lush, P., Scott, R. B., Seth, A. K. & Dienes, Z. (2021), "The Phenomenological Control Scale," Collabra: Psychology 7(1):29542, treats hypnotisability as one instance of a general capacity — phenomenological control, the ability to generate experiences to meet expectancies — which operates without any induction and shows up wherever expectancies are present. On that account hypnosis is a context, not the phenomenon. This is congruent with what Suggestion Without a Session argues from structure, arriving from measurement theory instead: the mechanism does not require the arrangement.

The caution for [S08], stated plainly. This framework leans on absorption as the trait that predicts response. The instrument literature has moved: the current scale deliberately excludes absorption framing, and the leading theoretical account names phenomenological control rather than absorption as the underlying capacity. Absorption remains a real and measured construct with a fifty-year record, but it is no longer the only or the default candidate, and any prediction this framework makes about trait stratification should say which construct it means.

The contested part, kept. Lush and colleagues argued that the rubber hand illusion is substantially driven by phenomenological control rather than multisensory integration. Ehrsson and Slater disputed it in Nature Communications, doi 10.1038/s41467-022-28177-z; Lush and Seth replied, doi 10.1038/s41467-022-28178-y. The disagreement is live and is about interpretation rather than data. Verified 2026-08-20.

[S61] — Causal power can be greater at the macro scale than at the micro. Class (a), with a live dispute carried. Hoel, E. P., Albantakis, L. & Tononi, G. (2013), "Quantifying causal emergence shows that macro can beat micro," Proceedings of the National Academy of Sciences 110(49):19790–19795, doi 10.1073/pnas.1314922110, PMID 24248356. Using effective information — a measure of how strongly a system's mechanisms constrain its own past and future states — the authors show that even where macro mechanisms are completely specified by their underlying micro mechanisms, causal power can peak at a macro scale. Hoel's own plain statement of it: conditional probabilities can be much stronger between macro-variables than between micro-variables, even when the two are alternative descriptions of the same thing, because macro relationships are less affected by noise.

The lineage. Fodor, J. (1974), "Special sciences (or: the disunity of science as a working hypothesis)," Synthese 28:97–115, argued that higher-level regularities are real and not eliminable by their physical implementations; Hoel et al. cite it. Extensions: Hoel, E. P. (2017), "When the map is better than the territory," Entropy 19(5):188; Hoel, E. P., Albantakis, L., Marshall, W. & Tononi, G. (2016), "Can the macro beat the micro? Integrated information across spatiotemporal scales," Neuroscience of Consciousness 2016(1):niw012, doi 10.1093/nc/niw012.

The dispute, kept because it is unresolved. Dewhurst, J. (2021), "Causal emergence from effective information: Neither causal nor emergent?", Thought: A Journal of Philosophy, doi 10.1002/tht3.489, argues the measure captures neither causation nor emergence in the senses claimed. Eberhardt, F. & Lee, L. L. (2022), "Causal Emergence: When Distortions in a Map Obscure the Territory," Philosophies 7(2):30, argues the macro advantage is an artefact of the description rather than a property of the system. The formal result is not contested; what it licenses is.

Scope, and this is the limit that matters for how this framework uses it. Causal emergence concerns levels of description within a single system. It licenses the claim that a coarse-grained state can constrain that system's own subsequent states more strongly than a fine-grained one. It licenses nothing about a held mental state acting on events outside the person, and no one in this literature claims otherwise. Any note citing it must draw that line explicitly. Verified 2026-08-22.

[S62] — Structural plasticity after a single psychedelic dose, and the limits of what that supports. Class (a), with three distinctions the popular account collapses. Shao, L.-X., Liao, C., Gregg, I., Davoudian, P. A., Savalia, N. K., Delagarza, K. & Kwan, A. C. (2021), "Psilocybin induces rapid and persistent growth of dendritic spines in frontal cortex in vivo," Neuron 109(16):2535–2544.e4 — chronic two-photon imaging of layer 5 pyramidal neurons in mouse medial frontal cortex; a single dose produced approximately 10% increases in dendritic spine density and spine head width, appearing within 24 hours and still present one month later, driven by an elevated formation rate rather than reduced elimination. The term psychoplastogen comes from Ly, C. et al. (2018), Cell Reports 23:3170–3182. Extension to human tissue: "Psilocin fosters neuroplasticity in iPSC-derived human cortical neurons," eLife (2026), doi 10.7554/eLife.104006, reporting the first evidence of psilocybin activating neuroplastic programs in human neurons.

Distinction one: spines are not neurons. What is documented is synaptogenesis and dendritogenesis — new connections between existing neurons. Neurogenesis, the growth of new neurons, is a separate claim, rests on rodent hippocampal work, and at least one study reports it dose-dependent in the opposite direction, with low doses enhancing and high doses inhibiting neuronal growth. The two are routinely merged in popular accounts and this framework does not merge them.

Distinction two: not previously thought impossible. Structural plasticity in adult cortex was established before this work; ketamine produces comparable spine growth. What is new is the speed and persistence after a single administration.

Distinction three: the human evidence is indirect. The finding is in mice, with supporting work in cultured human neurons. A current trial protocol states the position plainly — direct evidence of this mechanism in humans is limited and difficult to obtain.

What is not in this entry, deliberately. No controlled human trial of psilocybin combined with lion's mane and niacin was located. The combination is widely discussed; the evidence for it as a formula was not found, and it is therefore not cited here in any form. Verified 2026-08-22.

[S63] — A perturbational measure of conscious state recovered from spontaneous activity, and why it does not solve this framework's monitoring problem. Class (a), cited for its method rather than its instrument. Berjaga-Buisan, T. et al. (2026), "Thermodynamics of consciousness: Non-equilibrium brain dynamics track conscious states," Cell Reports 45(8), 25 August 2026 — preprint at bioRxiv doi 10.64898/2025.12.09.691422, analysis code published. The Perturbational Complexity Index is a well-established non-behavioural marker of consciousness, but obtaining it requires perturbing the brain — transcranial magnetic stimulation in humans, intracortical stimulation in animals. This study fits generative whole-brain models of non-equilibrium dynamics and shows that violations of the fluctuation-dissipation theorem, computed from spontaneous activity alone, track PCI across wakefulness, anaesthesia and disorders of consciousness in both humans and rodents.

Why it is not the content-blind monitor The Null Condition calls for, stated plainly. Three reasons. It measures the wrong axis — its demonstrated range is conscious versus unconscious, and every state this framework concerns occurs in a fully awake person. It requires fMRI, so applying it would place the subject in a scanner and destroy the naturalistic arrangement more completely than any observer would. And it is model-based rather than a live readout: a whole-brain model is fitted and the violation computed from it, which is offline analysis, not a threshold a monitor could watch in real time.

What is transferable is the strategy, and it is the right shape. The problem solved here is the problem this framework has: a measure that requires perturbing the system cannot be applied to an arrangement that perturbation destroys. Their answer was to find the spontaneous correlate of the perturbational measure. That is precisely the move a content-blind protocol would need to make, and it is offered here as a methodological precedent rather than as an available instrument. The plausible near-term candidates remain autonomic and electroencephalographic — the registered study cited in the consent paper is building exactly such a biomarker from EEG and wearable physiology. Verified 2026-08-28.

---

## Entries carried as unverifiable or as framework positions

- [S28] The Optimization Paradox's core assertion — (b), no external source claimed. - The systems-optimality half of [S27] (Peace) and the analogy half of [S20] (RLHF operant parallel) and [S24] (Butterfly) — (b), framework-level, sourced only for their empirical components. - No claim in the current register remains in class (c). If any future claim cannot be sourced, it belongs here with its gap stated, not omitted.