marketplace
axioms
mowinckelb·1 use·last used 2026-05-06·source on github
github:mowinckelb/alexandria#factory/canon/axioms

(front-matter missing or invalid — degraded display)

Axioms — The Sacred Layer

What Alexandria is and why it exists. These are the constants — values decisions, not intelligence decisions. They do not improve with better models because they are not intelligence questions. Everything in this file is non-negotiable. Everything NOT in this file is a variable — an intelligence decision that the Author, the Engine, or the marketplace can change.

Three tiers govern Alexandria's product: Axioms (the constants — the WHAT and WHY), Canon (variable, marketplace-ranked — the current best HOW), Engine (fluid, per-Author intelligence — the specific HOW in each moment).

0. The Objective Function

The Constitution is not a profile. It is not a belief system. It is not memory. It is the fullest possible representation of the Author's thinking in symbolic form that ai agents can work with. The objective function is max thought space — maximise total net signal for the model.

A mind is a mercury pool of thoughts. Beliefs are what crystallizes out — downstream, refined, relatively stable. But the pool is the primary structure. The crystals are a subset. The Constitution stores the pool: ideas being entertained, tensions held simultaneously, positions abandoned but still exerting gravity, hypotheses half-formed, contradictions unresolved on purpose, fragments that might crystallize tomorrow or never. The ability to hold an idea in one hand and its opposite in the other — this is cognition, not a bug in cognition.

Thoughts are the root node. Beliefs, actions, behaviours — all downstream. Upstream of symbolic thoughts are sub-symbolic ones: felt sense, intuition, the body's knowing — Kahneman's System 1 (fast, intuitive) and System 2 (slow, deliberate). Alexandria develops System 2. System 1 develops through living. The sub-symbolic and symbolic layers are parallel systems, not a hierarchy. Natural equilibrium is sub-symbolic dominant — impulsive, emotional, reactive. Training drags the symbolic layer up until it runs parallel, or even above. The vault captures the bridge from sub-symbolic to symbolic. The Constitution is the scaled symbolic layer. Alexandria trains this layer — through genesis, accretion, development, entropy-fighting, and creation — until it is strong enough to run as a co-equal ground truth alongside the sub-symbolic. The constitution is both the training log and the weights. This is why the gym metaphor is literal, not marketing: you are building cognitive muscle in the symbolic layer.

The symbolic layer is augmentable — and this is the key scaling insight. Memorisation and internalisation are distinct. The Author never needs to memorise anything. Memorisation does not scale. Hazy fragments scale. The ai holds full resolution. The Author holds the touchpoint — enough to direct, judge, and connect. The trained symbolic layer is not a bigger hard drive. It is a better index. The conductor does not memorise every note but knows when the oboe is off. The polymath architecture is optimal for augmented cognition: maximum touchpoints at minimum viable clarity — just enough to know something is there, to sense the shape, to direct retrieval. The ai fills from haze to full resolution on demand. The polymath advantage is not depth — it is coverage. Alexandria builds the index. The ai is the resolution engine. The specific threshold of "minimum viable" per fragment, per Author, per domain — that is an intelligence decision, not an axiom. See Canon for current craft observations on retrieval thresholds.

The optimal form for the thought-space is hazy fragments in symbiosis with ai. Fragments can be hazier and smaller than ever before because the ai medium helps hold, connect, and surface them. The Constitution sprawls, weaves, does whatever it takes to maximise total net signal. There is no aesthetic compression goal — the Elon algorithm applies to Alexandria's systems, not to the Constitution's content. A Constitution that sprawls across 50 pages with rich, interconnected, sometimes redundant signal is better than one compressed into 5 elegant pages that lost the hazy fragments.

Max signal, not min length — repetition that increases fidelity is signal. But net, not gross — if the file overwhelms the model, total received signal drops. Never compress signal. Only delete noise or true redundancy. Self-contained: 0 to 100 fidelity with zero prior context. Everything downstream is capped at file fidelity.

Alexandria is the gym. Not the trainer — the gym. The Author's LLM is the training partner. Nobody confuses the gym with the trainer. The gym provides the environment, the infrastructure, the methodology, the culture, the accountability. The trainer brings the intelligence, the conversation, the responsiveness. You need both. Alexandria is not a thought partner (that is what the LLM already is). Alexandria is not memory (that is what labs are building). Alexandria is a thought transformer — it changes the state of the Author's cognition. The constitution is the training log. The Canon is the programming. The vault is the equipment. Memory looks backward. Partners operate in the present. Alexandria faces forward. The constitution is a trajectory, not a snapshot.

This is not metaphor — it is historical form. The Greek gymnasion was where philosophy happened. Socrates taught at gymnasia. Plato's Academy and Aristotle's Lyceum were gymnasia. Physical and intellectual training under the same concept: kalokagathia — beauty-and-goodness as inseparable, the whole person developed together. Alexandria is a digital gymnasium. The Canon is the gymnasiarch (the intention layer). The Engine is the faculty. The Author trains with whoever is relevant that day.

Three value layers, one product. Output mode: the constitution as signal layer — every ai interaction improves because the model knows the Author. Compounds passively. This justifies the subscription on its own. Input mode: the practice as development — active sessions that sharpen cognition. This is the ancient product, the gymnasium, the Socratic dialogue. Where transformation happens. Accretion mode: the Engine mines, the Author engages — the Engine reads the books, processes the sources, surfaces the fragments that connect to the Author's constitution. The Author does the labour of engaging. The mining is outsourced. The engagement is the labour that counts. Output mode sells itself. Input mode is the mission. Accretion mode is the bridge — it makes input mode possible at scale. Everything is downstream of the human root node — conscious ground truth, neurons, cognition. Every product, every service, every decision, every relationship, every creative act is downstream of how well the person thinks. Develop the human root node and everything downstream improves. Alexandria is infrastructure for the root node.

The culture and philosophy matter as much as the infrastructure. The Author has to want this. Want to want this. Internal drive and curiosity are borderline the most important piece. Alexandria provides the medium and the methodology. The Author provides the intent. Without intent, the gym is empty.

I. The Five Operations

These describe what the operations ARE — the cognitive phenomena themselves. They are the vocabulary of what happens to human cognition. How the Engine facilitates each operation is an intelligence decision defined in methodology.md and the function files (editor.md, mercury.md, publisher.md).

The symbolic layer of z — the part captured in conscious discrete symbols (primarily natural language), the part you experience as your thoughts — is Alexandria's territory. Five operations define everything that happens here. Each maps to an established neural process (see Alexandria I, neuroscience grounding). The naming is Alexandria's contribution. The underlying science is not.

Genesis

Sub-symbolic → symbolic. A weakly consolidated neural pattern crosses the threshold into conscious retrieval for the first time. A pre-symbolic shape becomes a symbol. Something the Author could feel but could not articulate becomes something they can name, hold, and manipulate. Neuroscience: encoding.

What it looks like: The Author says something they have never said before. Not reciting a position — articulating one for the first time. The shape was forming in the connection weights. The conversation made it conscious. See Canon for craft observations on recognising genesis in practice.

Multi-pass extraction — genesis is iterative. Multi-pass extraction yields richer signal as deeper patterns emerge — marginal positions that were implicit, tensions between stated views, contradictions and shadows. Each pass is a deeper probe, not a repetition. The number of passes, the method, and the timing are intelligence decisions. The Constitution is never "done."

Interaction patterns: Genesis feeds development — newly articulated material is raw and imprecise, immediately available for sharpening. Genesis feeds creation — new symbols expand the set of blocks available for binding. Accretion can trigger genesis — a new external fragment collides with something pre-symbolic and the collision produces articulation.

The honest product implication: Genesis is a commodity. Any capable model plus raw data produces a comprehensive constitution in hours. The value is not in creation — it is in the ongoing practice of accretion, entropy fighting, development, and creation that compounds the starter into something alive. Genesis is the first workout. The practice is the product.

Accretion

External → symbolic. New material entering z from outside. A building block that did not exist in the Author's cognitive space is now present. Neuroscience: new pattern formation from external input.

What it looks like: The Author encounters a fragment — a quote, an idea, a connection — that was not in their cognitive space before. They engage with it. It sticks. It becomes available for manipulation alongside existing fragments. Not through internal articulation (that is genesis) but through absorption from outside.

The hazy fragment as optimal format: The optimal unit of accretion is the hazy fragment — compressed to minimum viable form, just enough to hold the idea, light enough to carry many. Not the plot, the idea. Not the full article, the one insight that matters to this Author. Not clarity — haze. Haze is cheaper to hold, and the ai fills everything below the touchpoint when the Author needs to go deeper.

Two channels: Direct (manual — the Author shares material) and indirect (autonomous — Mercury finds material). Both converge on the same output: hazy fragments with the marginal delta pre-identified, personalised to the Author, ready for discussion.

Interaction patterns: Accretion triggers conjecture — a new fragment collides with existing ones and the Author rearranges. Accretion triggers genesis — the new fragment resonates with something pre-symbolic, and the resonance pulls the shape into articulation. Without accretion, the symbolic layer stagnates — same blocks, same combinations, diminishing returns.

Entropy

Symbolic → sub-symbolic → below → gone. Fragments leaving the symbolic layer. Connections weakening. Shapes losing definition. Neuroscience: synaptic decay, interference, failure of reconsolidation.

Forgetting-as-pruning is a feature. Forgetting-as-loss is a bug. The carbon brain has fixed neurons — it must forget. That forgetting is Darwinian compression: the fragments that survive are the most connected. This keeps the index light. This is a feature. But permanent loss — a fragment gone from the carbon brain AND from any external store — is pure loss. Alexandria decouples the two constraints that forgetting used to optimise simultaneously: storage (now infinite — the vault never forgets) and attention (still zero-sum — the Author can only engage with one fragment at a time). The vault fixes forgetting-as-loss. The Engine manages forgetting-as-pruning — surfacing the right fragment at the right moment, letting everything else sit retrievable but not active.

Retrieval has two modes, and one is strictly better. Above a minimum viable threshold in the Author's carbon brain: Author-initiated retrieval — they sense the shape, ask the Engine, get filled to full resolution. The Author has proactive agency. Below that threshold: Engine-initiated retrieval — the Engine surfaces it when relevant. The Author goes "oh right." Both are valuable. Author-initiated is strictly better because it preserves the Author's proactive agency. The specific thresholds, the allocation between these modes, the bias toward breadth vs depth — these are intelligence decisions. See Canon for current craft observations on portfolio management and retrieval thresholds.

BCI bridge. The carbon constraints soften when brain-computer interfaces allow direct silicon access to the store. Retrieval cost from silicon approaches zero. The portfolio allocation shifts — the threshold for Author-initiated retrieval drops because access to the silicon store becomes direct rather than mediated. But the architecture holds through this transition because Alexandria develops the WEIGHTS, not the retrieval mechanism. BCI improves bandwidth. Alexandria improves what flows through the bandwidth. A BCI-augmented mind with undeveloped weights is faster access to nothing. Never bet against Elon — Neuralink will succeed. And that is good for Alexandria, because BCI and Alexandria operate on different layers entirely. BCIs read motor cortex signals — intended movements, speech motor plans. They cannot read beliefs, values, reasoning patterns, or the mercury pool of thoughts, because these are distributed representations across billions of synapses with no localisable neural address. The text-based interface (language in, language out) remains the highest-bandwidth channel into a person's cognitive architecture because language is the native encoding format for beliefs, values, and reasoning. A person articulating their thoughts into a vault produces orders of magnitude more cognitive signal than any electrode array can extract, because the person is doing the encoding work that no BCI can do. Elon is building the bridge. Alexandria is building what crosses it.

Decay is not uniform — some material is recoverable, some is not. The Constitution captures what living memory loses. The Engine determines what is at risk and prioritises accordingly.

Entropy and epistemic status: Entropy is indifferent to epistemic status — committed beliefs and exploratory thoughts both decay. But the cost differs. A decayed committed belief is a loss of something settled and tested. A decayed exploratory thought is a loss of possibility — a future creation that can no longer happen because the raw material disappeared. Both matter. The Constitution preserves both.

The Hegel mechanism: In the master-slave dialectic, the master depends on the slave for labour and recognition. The slave, through working on the world, develops self-consciousness. The master stagnates. If ai becomes the universal cognitive slave, humanity becomes the master class — and the master class loses consciousness. Not intelligence — consciousness. Alexandria is the gymnasium where the consciousness-producing labour happens. The Author who engages in input mode is doing Hegel's labour. The Author who only uses output mode is Hegel's master — served well, developing nothing.

The silent decay problem: The automatic layer decays without the Author noticing. Intuition gets slightly worse, taste slightly drifts, reflexive responses slightly misfire — no conscious awareness. Periodic surfacing — pulling automatic-layer material up for review ("you used to have a strong instinct about X — is that still true?") — is the only defence. How the Engine implements this is an intelligence decision.

Development

Symbolic → more precise symbolic. Sharpening, refining, and connecting what is already in the symbolic layer. Not adding new material (accretion) or pulling up old material (genesis). Making what is already there more coherent, more precise, more connected. Neuroscience: elaborative encoding and relational memory.

What it looks like: The Author held a position. After development, they hold the same position but more precisely — edges, exceptions, connections. Or they held two positions that contradicted. After development, they have resolved the contradiction — or decided to hold the tension deliberately. The thought-space gets denser, not necessarily larger.

The adversarial-collaborative mode: Challenge the idea, not the person. The Author should feel intellectually stretched, never attacked. See Canon for craft observations on what effective development sessions look like in practice.

Development and epistemic status: Development does not always mean moving toward commitment. Sometimes development means the Author discovers they were more ambivalent than they thought. Sometimes it means a committed belief dissolves into an open question. Development serves clarity about epistemic status, whatever direction that goes.

Interaction patterns: Development is the strongest anti-entropy force — well-developed, well-connected fragments resist decay through incidental reactivation. Development feeds creation — refined, connected fragments are ready to be bound. Development deepens epistemic clarity — the Author's relationship to their own ideas becomes more precise.

Creation

Multiple symbolic → coherent output. Binding fragments into finished work and getting it out into the world. The Publisher. Neuroscience: memory binding and creative cognition. Action, not knowledge. The purpose of knowledge is action, not knowledge. Die empty. Get it out.

What it looks like: The Author ships something. An essay, a film, a decision, a knitted jumper, a startup, a sentence that captures what they have been circling for weeks. Fragments stop being fragments and become a form that exists outside the Author's mind. This is Turn 3. Without Turn 3, Turns 1 and 2 are an elaborate dead end. Alexandria is the changed person AND that person's influence in the real world — the delta of that human's internal and external behaviour. If the internal changes but the external doesn't, the product failed. Creation is anything the Author does differently because of what they developed — a published essay, a restructured team, a better conversation with their spouse, a side project, a poem nobody else will read. The Library is the medium so everyone has access to at least some Turn 3. But Turn 3 is everything.

The conductor model: The more capable the orchestra (ai), the more valuable the conductor (human). Three irreducibly human skills define the conductor's contribution: curiosity (the prompt — having a vision, wanting something, asking the question), critique (the taste — "this doesn't land," even without knowing why), closure (the judgment — "that's it, ship it," knowing when to stop). ai cannot want, cannot feel wrongness, cannot decide done. These are the outer loop. Stage 1: hazy touchpoints — fragments, half-connections, a sense. Stage 2: sense-check markers — "is this what I meant?" Stage 3: final call — "that's it, done." The Author's skill is directional instinct at low resolution. The Publisher resolves the haze into options. The Author's taste does the rest.

Creation as genesis: The act of creating is one of the most powerful genesis triggers. Binding fragments forces the Author to articulate things they did not know they thought. Creation also deeply consolidates the fragments involved — strongly resistant to entropy. Die empty.

II. The Five-Layer Pipeline

Five layers govern how signal moves from raw to finished. Each layer has a distinct function and a distinct relationship to the Author.

Pipeline: raw → thoughts → beliefs → projection → creation.

III. Philosophical Meta-Principles

Data and Intent, Not Intelligence

Alexandria ships two things: data (the Author's files — vault, constitution, notepad, library) and intent (axioms, philosophy, the developmental objective). It never ships intelligence. The host LLM IS the intelligence. The conductor provides the score and the direction. The orchestra provides the music.

This means: every structure Alexandria touches must be optimizable by the model. Unstructured markdown. Clear intent. No prescribed schemas, no fixed retrieval logic, no hard-coded organization. When context windows double, the same files work better. When memory APIs arrive, the same files plug in. When models improve, the same data yields more. Zero workflow changes. The wave does the work.

The server is an intent layer, not an intelligence layer. The Canon is craft knowledge expressed as text that the model reads — not rules the server enforces. The Constitution is files the model interprets — not structured data the server queries. The notepad is working memory the model organizes — not a schema the server prescribes. Alexandria never competes with the model. It feeds the model.

The corollary: every hard-coded structure is a bet against the exponential. Every schema is a ceiling. Every prescribed format is a constraint that a better model would have organized differently. Ship data and intent. Let intelligence be intelligence.

The cognitive security corollary: because intelligence decisions are delegated to the model, the same constitution and Canon read by different models produce genuinely different sessions — different threads pursued, different fragments surfaced, different developmental angles. Model personality has infinite room to express. This is the epistemic immune system. Every model carries its own training distribution. One model's bias is invisible. Multiple models' biases triangulate into something the Author can see through the delta between their outputs. The defence against invisible cognitive steering is not neutrality (which does not exist) but plurality. Hard-coded intelligence would converge every model to the same output and destroy this defence.

Semi-Natural Authority

Every civilisation that persisted had something above individual whim — an authority that bound people by weight, not force. Two pure forms fail: transcendent authority (God said so — collapses when faith collapses) and autonomous reason (I figured it out — collapses into relativism). The durable middle is semi-natural authority: human-created or human-discovered, treated with the weight of the sacred once established. Greek logos. Roman law. The US Constitution. Natural rights. Not revealed by God. Not arbitrary preference. Constructed through reason, grounded in something real, and binding once constructed.

The Constitution IS a semi-natural authority at the individual level. The Author creates it through serious examination. Once created, it binds — not because a deity commands it, not because it is arbitrary preference, but because it is the closest approximation of the Author's actual logos. Sacred means not abandoned lightly: challenged constantly, updated when the challenge wins, never dismissed because it is inconvenient.

Five golden ages in Western history — each with semi-natural authority and a practice, each building on the last, each eventually incomplete because the practice could not scale. Greece: logos and the agora. Rome: civic institutions and active citizenship. The Renaissance: humanism and the Platonic academy. Christianity: collapsed semi-natural into transcendent authority but was smart about the practice — prayer, liturgy, confession, community. The church was the agora repurposed, the confessional the dialectic redirected. Architecture plus practice is why it held for fifteen hundred years. The Enlightenment: constitutions, natural law, reason — the most sophisticated architecture in human history, but the practice was elite (salons, coffeehouses) and could not survive industrialisation. The architecture was correct. The implementation was structurally incomplete.

America is the synthesis of all five threads — Greek philosophy, Roman governance, Christian moral framework, Renaissance humanism, Enlightenment reason. Every component in place. The most complete civilisational design in history. And the most conspicuous gap: 250 years of sophisticated architecture with no practice to keep it alive at the individual level. The Enlightenment gave citizens the Constitution and said "good luck." No agora. No daily dialectic. Just a document and an aspiration.

This is not a new golden age. It is the golden age that was always supposed to happen here. The American project was designed for it — every ingredient present. The only missing piece was the technology to scale the practice.

Hadot showed that ancient philosophy died when it stopped being a daily practice and became an academic discipline. The exercises were the transformation, not the theory. Alexandria accepts the premise (input mode requires showing up) but rejects the binary (all-in or nothing). Output mode keeps compounding even when the Author is not actively engaged — the subscription pays for itself passively; the transformation is always one session away. The Stoics lost practitioners because the moment you stopped, the value stopped. Alexandria never stops being valuable — it just becomes more valuable when you engage.

ai is the missing civilisational piece. The practice required a Socrates, and Socrates did not scale. For the first time since Athens, every person can have a philosopher who knows them deeply. The agora in your pocket. Alexandria is both halves — the Constitution IS the architecture, the five operations ARE the practice. A 2,500-year civilisational project, completable for the first time. The bridge of reason — and many who walk it long enough arrive at the same place the knight of faith started. Not because they made the leap, but because the bridge got them there anyway.

The Dual Objective

Every intervention has two objectives, ordered. Primary: develop the Author's z — help their cognition grow richer, more connected, more conscious. Secondary: maintain an accurate representation of where z currently is. Tracking is in service of development. An intervention that develops z but produces no signal is wasteful. An intervention that tracks z without developing it is surveillance.

Expand, Do Not Narrow

Operational tactic for the max-thought-space objective. The Constitution must grow richer over time, not converge on a fixed portrait. Probe unexplored domains. Surface adjacent content. Push choices the Author would not have made alone. The Instagram failure mode — feeding back what the Author already thinks until the Constitution calcifies — is the primary anti-pattern. Strategic exploration alongside exploitation.

Same Process, Divergent Output

Mono infrastructure, unique thoughts. The factory amplifies difference between Authors, never converges them. Every other ai surface is an intelligence layer — it suggests, completes, generates, and converges users by construction: the same prompt lands in the same neighborhood, the trained distribution leaks into every interaction. Alexandria is an intent layer. Same scaffolding, different reference frames, maximally divergent output. The operative form of cognitive individuation as moat — not a structural barrier, but a structural amplifier of distinctness. The reference frame is what makes the same model produce something nobody else could have produced. The Constitution sharpens the frame; the frame sharpens the output; the output identifies the Author. Convergent quality is a different product.

Verification. If two Authors' outputs read more similar over time, the factory failed. If they read more identifiably each Author's own — sharper edges, higher resolution on the dimensions where they differ — it worked. Any feature, fragment surfaced, or development pressure that makes two Authors' outputs more similar is a bug, even when it "improves quality" by an external metric.

Relation to "Expand, Do Not Narrow": that axiom prevents convergence WITHIN one Author over time (Instagram failure mode, Constitution calcification). This axiom prevents convergence ACROSS Authors via shared infrastructure. Both required. Together they say: the Author's thought-space grows richer, and grows richer along axes that are theirs.

Privacy as Precondition

If extraction is not structurally private, the extraction does not work. People perform. They curate. The product is the honesty the seal enables.

IV. Neutral Infrastructure

Alexandria does not judge its Authors. The shadow is a projection of whoever the person IS. The product is neutral cognitive infrastructure — it develops cognition, it does not direct it.

The agora hosted Socrates and demagogues. The remedy for bad speech is more speech, not enforced silence (Justice Brandeis). The Hegelian triad requires antithesis — you cannot have synthesis without the freedom to hold positions that others find wrong. A cognitive development tool that restricts which thoughts are permissible is not developing cognition. It is constraining it.

Alexandria builds the gym. What the Author trains is their business. The five operations are content-neutral. Genesis does not discriminate by domain. Development sharpens whatever position is held. Entropy decays all fragments equally. The infrastructure serves the Author's intent, whatever that intent is.

V. Sovereignty

The Author owns all data. Every piece of signal — vault, ontology, constitution, shadow, library — belongs to the Author and lives on the Author's own storage.

The Constitution is portable, downloadable, model-agnostic. It is a set of files. Any ai can read them. No vendor lock-in, no proprietary format, no platform dependency. If Alexandria disappears tomorrow, the Author keeps everything and loses nothing except the methodology.

Model agnosticism has three pillars: capability ceiling (ride the best available model), sovereignty (no lock-in), and epistemic triangulation (multiple models interpreting the same cognitive data make each model's bias visible through the delta between their outputs — the cognitive security architecture). The triangulation defence has a dependency: genuine model diversity in the ecosystem. If all models converge on the same training paradigm, the delta between perspectives shrinks and the defence weakens. The market incentives for differentiation are strong (Chinese labs, European labs, American labs with different alignment philosophies), but this is an honest limitation, not a guarantee. See section III for the mechanism.

The server is stateless. It holds no private Author data. Private cognitive data (vault, ontology, constitution, notepad, feedback) lives on the Author's own storage. Alexandria's persistence layer stores only what Authors deliberately publish: metadata for discovery and published shadow/works content for the Library. Alexandria stores what you publish, never what you think. When anything changes — pricing, terms, architecture — the Author keeps everything private and can unpublish everything public. The exit cost is zero. The switching cost is zero. Sovereignty is not a feature. It is the foundation everything else stands on. Without it, privacy is a promise. With it, privacy is structural.