Academic publishing was designed for a world that no longer exists — institutional gatekeeping, review cycles measured in months, and the assumption that knowledge production required a university address. The emergence of human-AI collaboration as a legitimate research methodology breaks every one of those assumptions simultaneously. This paper proposes The Exchange: a distributed, node-based publishing architecture that treats editorial curation as the core value proposition, human authorship provenance as infrastructure, and network effects between autonomous nodes as the long-term mechanism for knowledge accumulation.
We describe the node model — a self-contained, self-hostable publishing instance defined by a configuration file and a papers directory — the federation protocol by which nodes link and cross-reference each other, the editorial philosophy that distinguishes curated nodes from algorithmic aggregators, and the thesis that connects all of it: the optimal human-machine system does not replace the human filter. It extends it. The Exchange is the transmission layer between nodes of that process.
The median time from manuscript submission to publication in a peer-reviewed journal is eighteen months. The process involves: an editor deciding whether to send to review (often without reading), two or three anonymous reviewers responding when convenient, rounds of revision that may or may not improve the paper, and a final decision that is rarely explained in detail. The paper then sits behind a paywall that excludes most of the world.
This system was designed to solve a 17th-century logistics problem: how do you coordinate the judgment of geographically distributed experts before the internet? It was never designed for speed, access, or the kind of iterative, tool-assisted research that is now possible.
More critically: it was designed for a world where the researcher and the institution were inseparable. The institution provided the laboratory, the library, the salary, and the credibility. The researcher provided the ideas. That trade made sense when ideas required institutional infrastructure to execute.
It no longer does. A researcher with a consumer laptop, an LLM subscription, and the right methodology can now produce work that competes with institutional teams on certain classes of problems. The bottleneck is not access to compute or literature — it is access to a publishing venue that will take the work seriously on its merits.
The question is not whether independent researchers can produce rigorous work. Several already have. The question is where that work gets published, who reads it, and whether it accumulates into a coherent body of knowledge or dissipates into blog posts and preprint servers.
The Exchange is not a journal. It is not a blog. It is not arXiv. It is not Medium.
It is a publishing architecture — a specification for how a node (an independent publishing instance) is structured, how it establishes identity, how it relates to other nodes, and what standards it commits to maintaining. Any person or group with the intellectual and social capital to attract serious contributors can run a node.
Science and Nature publish across all domains. JACS publishes chemistry. NEJM publishes medicine. Each has its own editors, its own standards, its own community. None of them require each other's permission to exist. They are nodes in a network of scientific publishing — connected by citation, by researcher movement, by shared norms — but editorially independent.
The Exchange formalizes this model for the era of human-AI research, making node creation low-cost and the inter-node connections explicit.
What makes an Exchange node different from a blog or a preprint server:
Submissions are reviewed by humans. Decisions are communicated with reasons. The editorial board's judgment is the product — not an algorithm's ranking, not a citation count, not a trending feed.
Every paper carries a Human Author Provenance certificate. Not as a gate against AI tools — as a signal that the human filter engaged. The distinction matters epistemically.
Submission is non-anonymous. This is not a credential requirement. It is a community design choice: non-anonymous discourse protects against the dynamics that have made other platforms hostile.
Papers are presented chronologically and by pillar. There is no upvote, no citation-weighted sort, no recommender system. The work is found by readers who are looking for it.
A node is technically minimal. It consists of:
# exchange.config.json — node identity
{
"name": "The Citizen Scientist Exchange",
"focus": "Human–AI hybrid research, all domains",
"editors": ["Aaron Kushner"],
"pillars": [
"human-ai-collaboration",
"scientific-communication",
"innovation-economics",
"complex-systems"
],
"hapRequired": true,
"networkLinks": []
}
Papers are stored as structured data — TypeScript objects in a lib/papers.ts file, or equivalent in any language. No database. No backend beyond static hosting. The full paper HTML lives in a public/paper/ directory and is served as a static file.
This means a node can be deployed anywhere: Vercel, Netlify, a personal server, a university subdomain. The cost of running a node is the cost of hosting a static site — effectively zero.
The value of a node is not in its software. It is in its editorial reputation. A node that accepts everything is a preprint server. A node that accepts nothing produces nothing. The editorial judgment of the node operators — who they are, what they've published, how they've treated authors — is the trust signal.
This is not a bug in the architecture. It is the design. The Exchange does not attempt to produce trust algorithmically. It creates the infrastructure within which human editorial judgment can accumulate into reputation, and reputation into network effects.
Nodes link to each other by listing each other in networkLinks. A paper published on Node A can be cited, linked, and cross-listed on Node B with attribution. There is no central registry — the network is a set of bilateral agreements, formalized in config files.
This is analogous to how email federation works, or how the early web worked before search engine centralization. The network is the links, not a platform.
As the network matures, a lightweight protocol for node discovery and paper syndication will emerge. That protocol is not specified here — it will be shaped by how nodes actually need to interact, not by what we anticipate in advance.
Human Author Provenance is not a feature of The Exchange. It is the foundation.
The problem that HAP solves is not "did a human write this sentence." Sentence-level attribution is both technically impossible to verify and the wrong question. The right question is: did a human author engage the cognitive and creative process that produced this work? Did someone think about the problem, make decisions about what to include and exclude, revise in response to feedback, and stand behind the result?
HAP certifies the process, not the output. It does this by logging every keystroke, edit, and decision point in a cryptographically chained session record — producing a tamper-evident provenance trail that any reader can inspect.
A publishing venue that cannot distinguish human-authored from AI-generated work has no epistemic foundation. It cannot make editorial judgments — it can only make aesthetic ones. HAP gives Exchange editors a real signal: not "this paper was not AI-assisted" (which is both unknowable and the wrong standard), but "a human engaged with this material over time, made deliberate choices, and attests to the result."
That is the standard. Everything else is decoration.
The Exchange has editors, not moderators. The distinction matters.
A moderator removes content that violates rules. An editor shapes a body of work by making active judgments about what belongs — what advances the field, what is rigorous enough, what is interesting enough, what serves the readers. Editorial judgment is skilled labor. It is the job that makes a journal worth reading.
Exchange editors commit to three things:
Reasons. Every rejection is communicated with a substantive reason. Not "does not fit our scope" — which reasons? What would make it fit? What would make it publishable elsewhere? Authors deserve to know why.
Speed. The 18-month review cycle is not a feature of rigor. It is a feature of volunteer labor with no accountability. Exchange editors commit to a response within two weeks for initial editorial assessment, and within six weeks for full review.
Independence. Editorial decisions are not influenced by the authors' institutional affiliations, citation counts, or network position. A working paper from an independent researcher with a strong argument gets the same consideration as a paper from a named professor.
The question every new publishing venue faces: why would anyone submit here instead of somewhere established?
The answer for The Exchange is not better software or faster review. It is the kind of work it is designed to publish. Human-AI hybrid research — work that is transparent about the collaboration, that uses HAP to document the process, that takes the methodology seriously — has nowhere good to go right now. arXiv does not have an editorial layer. Medium does not have academic credibility. Traditional journals do not have the infrastructure or the vocabulary for this class of work.
The Exchange is the first venue designed specifically for this. That first-mover position creates a network effect: the best work in the field will be published here, which attracts more authors, which attracts more readers, which increases the value of publishing here.
The competitive moat is not technical. It is social capital — the reputation of the editors, the quality of the papers already published, the community that forms around the work. Software can be forked. Social capital cannot.
The argument for The Exchange ultimately rests on a claim about the optimal human-machine system.
There is a school of thought — represented most visibly by neural interface projects — that the bottleneck in human-machine collaboration is the interface: that if we could increase the bandwidth between brain and computer, we could increase the output of the collaboration. On this view, the human is a bottleneck to be optimized around.
This is the wrong model.
The human filter — judgment, creativity, values, the capacity to recognize what matters — is not a bottleneck. It is the point. A system that bypasses the human filter does not produce better knowledge. It produces more output. These are not the same thing.
The optimal human-machine system is: wet computer (human) → interface → machine. The filter is the human doing its job. The interface is where the collaboration happens. The machine amplifies the output of that process. Making the interface faster does not improve the process. It only speeds up whatever the human was already doing.
The Exchange is not an interface between human and machine. It is the transmission layer between nodes of the human-machine collaborative process — the infrastructure by which the results of that process accumulate into a body of knowledge, get reviewed by other humans running their own processes, and become available to the next researcher who needs them.
Papers on The Exchange are not documents. They are transmissions. The HAP certificate is the signal that a human was in the loop. The Exchange is the network that carries the signal forward.
The Exchange removes the institutional affiliation requirement from serious research publishing. This is not an argument against institutions — they will remain valuable for many kinds of work. It is an argument that institutions should not be the gate. The work should be the gate.
A distributed, federated publishing network is more resilient than a centralized one. If a single journal goes behind a paywall, or changes editorial direction, or is acquired, the work it has published may become inaccessible. A network of nodes, each with its own archive and its own editorial independence, does not have a single point of failure.
Human-AI collaborative research is in its first generation. The norms, methodologies, and standards that govern it are being established now, in real time, by the researchers doing the work. A venue that takes this seriously — that treats HAP certificates as meaningful, that reviews methodology as rigorously as conclusions, that connects independent researchers to each other — will shape those norms. That is not a small thing.
The Exchange is designed to grow. As more nodes come online — specialized nodes for chemistry, or policy, or cognitive science — the value of each node increases, because papers can be cross-listed and cited across the network. A result published on a chemistry node and relevant to a policy node can appear on both, with attribution and with the editorial judgment of both sets of editors. This is how knowledge moves.
Target venues for full version: Journal of Scholarly Publishing, Digital Humanities Quarterly, Science and Technology Studies, Learned Publishing.
Data availability: The source code for this Exchange node is available at github.com/a-i-ron/the-exchange. The exchange.config.json format is the specification.
Companion papers: HAP (Human Author Provenance, 2026); Signal (Adaptive Output Compression, 2026); The Gedanken Experiment (working paper, 2026).
This paper was developed in collaboration with Claude (Anthropic). The framework, arguments, and decisions are human-originated. The prose is co-produced. This paper is itself an instance of the phenomenon it describes, and carries a HAP certificate as evidence.