Foundations & references

A research-backed development of “link brain”

Version 1.0 · 22 October 2026. This extended synthesis includes the user's later clarifications about practical grammar, voluntary exchange, ownership, archetypes, and the turning wheel. Published findings and historical examples are cited inline; definitions, schemas, experiments, and recommendations are proposals rather than a new established academic theory.

The strongest formulation is: make knowledge reusable at the level of its underlying structure, not merely its surface content. A “knowledge brain” would preserve the artifacts of work alongside the concepts, principles, assumptions, operations, evidence, and judgment needed to use them.

“Link brain” would mean establishing a meaningful connection between such bodies of knowledge. The connection could preserve an existing rule, adapt a method, reveal a contradiction, or produce a new capability that neither contributor had alone.

The design should keep three constructs distinct:

Scroll the table sideways to see every column.

The simple public-facing idea can remain “Link what you know. Build across disciplines. Feed what you learn back.” Metapraxis can remain the deeper framework behind it, rather than become terminology every participant must learn.

The practical grammar inside the work

Metapraxis should not drift into a purely abstract account of knowing. Its practical value is in the details: what a component is called, which inputs it requires, which operation changes it, what a valid output looks like, what depends on it, and how someone can tell whether it worked.

For the proposed framework, a minimal grammar of practice could be:

Context + components + relationships + operations + constraints + evidence + revision.

This is a design vocabulary, not a new formal language already specified or implemented. It can begin with structured notes and worked examples before any executable schema is needed.

The ontology names the relevant entities and relationships. The epistemic record explains the status and support of claims. The workflow specifies actions, dependencies, and checks. The revision history preserves what changed and why.

Different disciplines may need different descriptions. Standardize the interface and enough metadata to enable understanding; do not standardize away the judgment, ambiguity, or expressive freedom that makes the practice useful.

The archetype and the playbook within it

The user's hero's-journey analogy offers a distinction between recognizable patterns and practical navigation. The archetype might describe an encounter, a trial, a transformation, and a return; the playbook beneath it records how to diagnose, seek help, choose a response, recover, and preserve what was learned.

Metapraxis makes that playbook explicit. Brain Flywheel gives it a social and practical life by helping someone else use it and contribute their own result.

This is an analogy selected for the construct, not a claim that one narrative pattern describes every life or that narrative archetypes prove a universal grammar of knowledge. Counterexamples and variations are part of the record rather than defects to be erased.

The codex therefore needs to preserve more than best practices. It should retain the circumstances that made a method work, the exceptions that constrain it, and the deliberate “twists” that helped someone create something different.

Philosophy: order, appearance, and participation

Clarifying the Indo-Iranian recollection

The user's exact remembered term, “rtagama,” remains unresolved. The closest verified conceptual family is Sanskrit/Vedic ṛta, Old Persian arta, and Avestan Asha, corresponding to an Indo-Iranian concept of truth and order. (Encyclopaedia Iranica: Ardwahišt, Indo-Iranian Religion)

Asha belongs on the order/truth side; Druj is the opposing concept, and Jean Kellens discusses an interpretation of the contrast as “real order” versus “illusory, deceptive order,” associated with daylight and the shadows of night respectively. (Encyclopaedia Iranica: Druj)

This is close to the user's remembered theme, but does not identify the original text or establish “rtagama” as the right spelling. The earlier tentative interpretation as ṛta plus āgama should therefore not be treated as confirmed.

Āgama is separately associated with tradition or received knowledge and with religious texts conveying ritual procedures and knowledge. (Britannica: Agama)

For Metapraxis, the useful philosophical question is whether an apparent order is grounded, inspectable, and useful, or only persuasive. This should inspire epistemic care, not a claim that the project's schemas reveal cosmic truth or that a mistaken contributor embodies deceit.

The wheel requires participation

Bhagavad Gita 3.16 and its commentaries describe participation in an established cycle of action and sacrifice. This provides a philosophical parallel for contributing to the cycle that sustains one, without reducing the religious teaching to a software pattern. (Bhagavad Gita 3.16)

In the user's metaphor, the first push is a useful contribution, the wind is motivation and opportunity, and continued turning comes from real applications and returned learning. Maintenance means correcting errors, clarifying limits, and renewing participation.

The flywheel is not a perpetual-motion claim. A dormant library may remain valuable, but the active cycle requires work, attention, relationships, and resources.

The Puppet Master: recombination with boundaries

Ghost in the Shell's Puppet Master contrasts copying with variety and originality, while the proposed merger raises an explicit uncertainty about preserving identity. (Film dialogue)

For this construct, the attraction is the emergence of something new through combination. The caution is the possibility of absorption: becoming an input to someone else's system without meaningful control.

The proposed knowledge exchange should deliberately reject that surrender. A participant shares a bounded module for an agreed purpose, not their whole identity, a permanent digital replica, or unrestricted rights to their expertise.

Representation and willingness are separate problems

The representation question asks whether knowledge can be codified into a useful, transferable form. The participation question asks whether someone wants to give, request, or receive it under the available terms.

A third question is governance: who controls storage, access, ranking, attribution, payment, correction, and exit? Solving representation alone cannot establish willingness or fair governance.

The default exchange should be small and purposeful:

“Here is one useful part of how I work, with an example, its limits, and the conditions under which you may use it.”

The reciprocal request should be equally concrete:

“I am trying to do this. Can you share a relevant method, explain the missing piece, or collaborate on a test?”

Contribution could be motivated by friendship, curiosity, recognition, reciprocity, public benefit, or money. These are hypotheses to test with participants, not incentives that should be assumed to work universally.

Two major failure modes

Capture or monopolization

One participant or platform could accumulate contributed knowledge while controlling discovery and extracting disproportionate value. This is a prospective design risk rather than a claim about a specific existing service.

Proposed safeguards include:

• Portability: Export meaningful modules, mappings, and provenance, not only raw files.
• Nonexclusive participation: Allow contributors to participate elsewhere unless they deliberately agree otherwise.
• Bounded permissions: Separate private use, redistribution, commercial adaptation, and model training.
• Transparent returns: Agree on attribution and compensation before consequential reuse.
• Exit: Stop future access where feasible without falsely promising retrieval of every exported copy.
• Plural authority: Make ranking and review contestable; do not turn a popularity score into epistemic truth.

These measures reduce particular dependencies but do not prove that monopolization is impossible. Governance would need its own tests as the system develops.

Hollow, fabricated, or unusable brains

An incomplete brain is not necessarily a fake one. The failure is misrepresented scope or authority: fabricated provenance, unsupported expertise claims, or material that cannot be applied despite being presented as a ready-to-use method.

Proposed safeguards include worked examples, explicit assumptions, known failures, evidence statuses, contributor review, and independent reuse attempts. AI-assisted drafting should not manufacture provenance or convert a plausible summary into a claim of tested experience.

Record “partial,” “unverified,” “private,” and “not applicable” separately. These states should not be penalized as if they were deception.

The risks could reinforce one another: extractive terms could encourage shallow sharing, while shallow material could reduce trust and practical reuse. A request-led pilot can test whether a worthwhile exchange exists before attempting a large network.

The closest academic foundations

Structural analogy: transfer relationships, not resemblance

Dedre Gentner's structure-mapping theory treats analogy as a mapping of relational structure between a base and a target domain, prioritizing connected systems of relations rather than superficial object attributes; its computational implementation generates candidate inferences, not automatic empirical truths. (Northwestern's Structure-Mapping Engine overview)

Design implication: “These both look like networks” is a discovery cue, not sufficient validation. A useful link should specify which roles and relationships correspond, which do not, and which inferences the correspondence licenses as hypotheses.

Conceptual blending: the combination can produce something new

Fauconnier and Turner's conceptual-integration account describes selective projection from input spaces into a blend, where completion and elaboration can generate structure not supplied by either input alone; the resulting ideas can also change how the original inputs are understood. (Fauconnier and Turner, 1998)

Design implication: preserve both the source modules and the new construction. Do not merge them into a summary that loses the reasoning behind the combination.

This is a close conceptual neighbor to your “A + B is not just the sum of the parties.” It is a useful account of generative thinking, not proof that every combination is useful or true.

Knowledge creation: documents do not replace people

Nonaka's knowledge-creation theory describes interaction among socialization, externalization, combination, and internalization; it also warns that explicit recombination can become superficial when disconnected from lived experience and that tacit knowledge is difficult to articulate. (Nonaka, 1994)

Design implication: a knowledge brain should be an interface to expertise, not a claim to have captured a person's entire mind. Include demonstrations, examples, judgment calls, and ways to collaborate with the contributor, not only rules and files.

Ontology engineering: make meanings interoperable

Gruber's ontology work explicitly addresses sharing and reusing formally represented knowledge through vocabulary definitions, including classes, relations, and functions, while translating between representation systems and attempting to preserve declarative content. (Gruber, 1993)

Design implication: different brains need not use one universal vocabulary. Instead, maintain local meanings and explicit translation agreements between them.

Epistemology: record why a claim deserves belief

Epistemology concerns knowledge and justification, with competing accounts of evidence, reliability, the structure of justification, and the limits of knowledge; there is no single settled account that your system can simply encode as “the rules of knowing.” (Stanford Encyclopedia of Philosophy)

Design implication: separate an item's content from its epistemic status. A proven statement within a formal system, a measurement, a useful heuristic, and an untested analogy should not acquire the same authority merely because they appear in the same graph.

Compositionality: specify how parts can act together

Fong and Spivak develop compositional approaches to databases, circuits, signal-flow systems, and collaborative design; structure-preserving maps retain specified mathematical relationships, while “generative effects” illustrate how combinations can create properties not visible in isolated parts. (Seven Sketches in Compositionality)

Design implication: this is a promising mathematical direction for the “mesh,” particularly when modules have precise inputs, outputs, and composition rules. It should be an optional formalization, not a requirement imposed on every artist, researcher, or practitioner.

Bootstrapping: improve the ability to improve

Engelbart's ABC model distinguishes doing work, improving that work, and improving the improvement process; his bootstrapping strategy includes sharing methods and lessons across organizations to strengthen collective capability. (Doug Engelbart Institute)

Design implication: the deepest compounding occurs when a project improves not only its own output, but also the methods used to create, connect, and evaluate subsequent projects. This is the closest operational counterpart to the relationship you propose between Metapraxis and Brain Flywheel.

Systems thinking: a good flywheel also needs correction

Meadows distinguishes reinforcing loops from balancing feedback and emphasizes information flows, rules, goals, and paradigms as intervention points; reinforcing feedback can amplify deterioration as well as improvement. (Leverage Points)

Design implication: the design needs two loops: one that generates and reuses knowledge, and another that challenges, corrects, retires, and limits it. Faster circulation without correction is not the goal.

Your examples, checked and sharpened

Pāṇini: from examples to a usable grammar

Pāṇini is a strong example of sophisticated codification: Kadvany analyzes the grammar's rewrite rules, auxiliary markers, and metalanguage, while distinguishing the Sanskrit grammar itself from broader claims about universal computation. (Kadvany, 2016)

For your construct, the productive question is: “Can work in a domain be distilled into a usable grammar of its concepts, operations, conditions, and exceptions?” Do not imply that all expertise can be reduced to a complete rulebook.

The sources reviewed here do not establish the broader claim that Pāṇini's grammar directly trained modern LLMs. However, a distinct 2026 system named PANINI is unusually relevant: it represents documents as structured question-answer networks over entities and events, retrieves reasoning chains, and keeps the base model fixed rather than updating its parameters. (PANINI paper)

Its ICML presentation reports improvements on its evaluated QA tasks using structured memory and compact reasoning chains; those results concern a particular structured-memory approach, not a demonstration of universal cross-disciplinary reasoning. (ICML 2026 presentation)

Recommended example: “Pāṇini illustrates disciplined codification; contemporary structured-memory systems illustrate one way codified relationships can support reusable machine reasoning.” Treat these as related design inspirations, not an established historical lineage.

Brownian motion and Black–Scholes: shared mathematics, different mechanisms

The chronology is more interesting than a one-way “physics became finance” story: Bachelier modeled financial fluctuations in 1900, while Einstein's account of suspended-particle motion followed in 1905; mathematical Brownian motion became a framework used across settings. (Britannica's probability-theory account)

In the physical example, suspended particles are buffeted by surrounding molecules, whereas the stock-price model uses geometric Brownian motion, (dS_t=\mu S_t\,dt+\sigma S_t\,dW_t), with different variables and interpretation. (Britannica, Columbia's Black–Scholes notes)

Columbia's derivation treats continuously compounded cash, (B_t=e^{rt}), as an assumption alongside stock dynamics and develops option replication through a self-financing hedging strategy; it is not an invention of continuous compounding or a payment-routing mechanism. (Columbia)

Recommended lesson: preserve the distinction between a stochastic structure and the mechanism it models. Similar data shapes may suggest a transfer, but do not establish shared causes, appropriate parameters, valid assumptions, or reliable predictions.

For a Brain Flywheel payment layer, start separately with permissions, contribution records, agreed revenue splits, accounting, and settlement. Mathematical inspiration does not itself establish a right to payment.

Markov processes, reinforcement learning, and markets

Reinforcement learning draws on Markov decision processes, where agents act in uncertain environments and learn from rewards; neural networks can represent learned functions within that framework. (ACM's account of Sutton and Barto's work)

A concrete study models Uniswap v3 liquidity provision as an MDP and uses PPO to choose liquidity ranges, evaluating the strategy on historical data in a simulated environment with fee, gas, and loss-versus-rebalancing terms. (Xu and Brini)

Recommended lesson: separate the components being linked: a state-transition model, a decision problem, a learning algorithm, a function approximator, and a market-specific environment. A Markov chain alone does not specify actions and rewards, and neither formalization nor a backtest establishes live profitability.

“Train arbitrage spreads” is too compressed to be a useful technical claim. A better proposal would identify exactly what is predicted or controlled, what information is observable, what reward is used, and how costs, changing regimes, and failure are tested.

Psychedelic neural art, latent representations, and Transformers

The psychedelic-art example corresponds well to DeepDream/Inceptionism: it repeatedly modifies an image to enhance features detected by a trained neural network, exposing learned patterns and sometimes mistaken associations. (Google's original Inceptionism account)

The Transformer was introduced as an attention-based sequence-transduction architecture replacing recurrent and convolutional sequence processing, not as the result of separate art-generating models communicating in latent space. (Vaswani et al., 2017)

There is a real later combination of language modeling and reinforcement learning: InstructGPT used supervised demonstrations followed by reinforcement learning from human feedback, while still retaining limitations and making mistakes. (Ouyang et al., 2022)

Recommended lesson: use this as a branching map of reusable ideas, not a single origin story. Learned representations, feature visualization, attention, optimization, and feedback are different components that can be recombined for different purposes.

Harness, flywheel, rocket

Retain these as your design metaphors, with explicit jobs:

• Harness: the execution environment you propose, containing tools, context, permissions, workflow state, and evaluations.
• Flywheel: the cycle that retains validated learning and returns it to future work.
• Rocket: a focused project that deploys the accumulated capability toward a specific goal.

The important distinction is between executing a process and improving a process. A harness becomes part of the flywheel only if useful results and corrections are actually captured, reviewed, and reused.

Expand “link brain” into a concrete operation

The link should be an artifact in its own right

The most useful design expansion is to preserve a transfer record between brains, rather than only drawing an edge labeled “related.” That record becomes a reusable piece of knowledge.

Scroll the table sideways to see every column.

This is compatible with FAIR's emphasis on reusable research objects, including algorithms and workflows, with rich metadata, provenance, licenses, and qualified references; FAIR does not require that every resource be publicly accessible. (Wilkinson et al., 2016)

Distinguish different kinds of links

I recommend typed relationships rather than treating every connection as equivalent. The following is a proposed vocabulary, not an established taxonomy:

• Related to: useful for discovery, with no claim of transfer.
• Analogous to: shares a specified relational pattern.
• Instantiates: applies an abstract structure in a concrete setting.
• Adapts: modifies a method for a new setting.
• Composes with: has compatible interfaces that support a combined workflow.
• Derived from: records an explicit reasoning or production lineage.
• Tests or contradicts: supports, challenges, or limits a claim.
• Revises or supersedes: records an updated version without erasing the old one.

This would let the system keep a speculative association without promoting it to a theorem. It would also make failed connections useful rather than invisible.

A + B can produce C, and change A and B

Use this as a design notation, not an empirical law:

[ \operatorname{Link}(A,B;\,m,e)\rightarrow(A',B',C,L) ]

Here (m) is the proposed mapping or composition, (e) is the evaluation evidence, (C) is a resulting capability or artifact, and (L) is the preserved lesson. (A') and (B') are revised source modules if the attempt changes how either is understood.

The result may be a successful invention, a useful adaptation, or a well-supported reason not to combine them. All three can improve the knowledge system, but they should have different statuses.

Preserve known rules, exceptions, and “twists”

Your idea should have two equally important modes: preservation and generation. It should help people retain what is already understood, including limits and exceptions, as well as discover new combinations.

I recommend the following record types for the codex. These are working design distinctions, not a resolution of philosophical debates:

Scroll the table sideways to see every column.

“Memorializing” should mean durable, attributed, versioned preservation, not freezing every accepted claim forever. Retain the original formulation and originator while allowing later correction, competing interpretations, and changes in applicability.

A creative twist should state what it changes. “Invert this design heuristic to provoke surprise” is different from “this mathematical consequence is no longer true,” and the system should not blur the two.

Make compounding a testable claim

The canonical six-stage cycle, consistent with the simple public website, is:

Collect → Curate → Connect → Apply → Learn → Reinvest.

At the working level, Collect and Curate distill practice; Connect makes a transfer explicit; Apply composes and tests; Learn and Reinvest preserve changes and return useful resources. The four-operation diagram is a compressed view of this same cycle, not a separate model.

I would track three accumulating assets:

• Knowledge: better-supported claims, examples, definitions, and constraints.
• Transfer capability: tested mappings, reusable interfaces, and knowledge of where transfer fails.
• Collective capability: stronger methods of collaboration, evaluation, and improvement.

Add a balancing loop alongside that cycle:

Challenge → detect conflict → correct or retire → propagate the revision

The system should not be judged by file count, link density, or how much text it generates. Measure whether subsequent work is better, faster, more reliable, or newly possible.

Scroll the table sideways to see every column.

Treat “compounding” as the hypothesis that retained improvements make later improvements easier or more productive. Do not promise exponential growth: errors, irrelevant material, maintenance, and coordination costs can offset the gains.

Metapraxis as the mesh

The refined definition is: Metapraxis is the practical grammar for documenting, standardizing, connecting, and revising how work is done. Its higher-order role is to connect ways of knowing and acting without erasing their differences; its everyday expression is inspectable components, processes, interfaces, assumptions, tests, and revisions.

Give it four questions:

• Ontology: What entities, relations, states, and operations does this construct recognize?
• Epistemology: What warrants its claims, and what would make us revise them?
• Composition: How can it connect with another construct while preserving relevant meaning and constraints?
• Praxis: What action does it enable, and what happened when we tried it?

The mesh need not force all modules into one logic. It should expose translation boundaries, incompatible premises, and unresolved disagreement, including situations where no responsible mapping is available.

The proposed system's distinctive emphasis is not any individual ingredient. It is the combination of people-owned knowledge modules, explicit transfer records, practical composition, evidence-sensitive evaluation, and retained learning.

A bounded first experiment

Do not begin by trying to model all knowledge. Start with two genuinely different, well-understood modules and one practical task.

A candidate suited to your interests would be a music module and a software-scheduling module. The proposed transfer would map beats and bars to scheduling boundaries, musical cues to events, and performance constraints to timing requirements, while explicitly preserving the difference between expressive musical timing and computational correctness.

Build a small interactive audiovisual sequencer as the test artifact. Compare the linked modules against the same source files supplied without the distilled concepts, interfaces, and transfer record.

Keep the task, source material, time budget, and evaluation criteria comparable. Record build time, timing errors, unexplained assumptions, successful reuse by a second person, and whether the final project produces an improved module that can be used again.

Predeclare a meaningful success threshold before running the comparison. If the structured version does not help, diagnose whether the problem is poor extraction, bad mapping, overhead, or an unsuitable task rather than calling the mere presence of a graph “compounding.”

This experiment tests your central claim directly: not whether two people can exchange files, but whether exchanging structured understanding produces a more capable result.

Recommended public and private language

For the public Brain Flywheel:

Turn what you know into something others can build with. Link knowledge, principles, and ways of working across disciplines, create something together, and feed what you learn back into the next cycle.

For the deeper Metapraxis construct:

The practical grammar for making work explicit, systematic, and transferable, and a mesh for connecting ways of knowing and acting without erasing their differences.

For “link brain”:

Not just a friend request or a shared folder. An invitation to connect the useful structures behind what we know, with enough context to understand them, enough freedom to recombine them, and enough evidence to know when they work.

Open decisions

The core distinctions are stable enough to explain, but several design decisions remain open. They should not be silently turned into product commitments.

• Unit of exchange: A method, a small collection, a workflow, a live collaboration, or several supported types?
• Contributor identity: Named individuals, communities, pseudonymous contributors, or different trust requirements for different tasks?
• Evidence threshold: What can be shared as exploratory, and what requires independent testing before being promoted?
• Economic model: Gift exchange, commissions, licenses, subscriptions, or project-specific revenue agreements?
• Storage and portability: Local modules, hosted collections, federated exchange, or a simpler pilot before deciding?
• Conflict handling: Who evaluates disputed provenance, incompatible frameworks, and downstream corrections?
• Public scope: How much conceptual depth belongs on Limelight versus a private or separately published framework?

This package develops the concept only. It does not change Limelight, publish Metapraxis, or establish novelty through an exhaustive prior-art review.