What is a discourse graph, and why is the paper rendered as one?

A discourse graph is an alternative form of scientific communication. Instead of a single linear document, the argument is composed of typed nodes — Questions, Claims, Evidence, Methods, Sources — connected by typed edges: addresses, supports, opposes, derived from, uses method. Every node is self-contained, addressable, and individually contributable.

The form was developed by Joel Chan, Matthew Akamatsu, and collaborators, and refined inside Roam Research, Protocol Labs, and adjacent research communities. The Q/C/E/S core schema is small enough to remember; this project adds Method (M) as a fifth type for analytical instruments — taxonomies, formulas, frameworks — that Claims invoke but that are not Claims themselves.

What discourse graphs change

The Resilient Data Futures whitepaper was written paper-first and then decomposed into a graph. That origin is unusual — discourse graphs are normally built incrementally, by contributors adding Questions, Claims, Evidence, and counter-evidence over time. We decomposed an existing paper to bootstrap the graph with real content, and to demonstrate what becomes possible once it exists.

Contributions become atomic. A paper bundles a question, methods, claims, and evidence together; none of it gets published until all of it does. To share one new observation, you write the surrounding apparatus — introduction, methods, related work, discussion — even when none of that is new. A discourse graph removes the bundle. One new observation is one Evidence node, with edges to the Claims it supports or opposes. One new assertion is one Claim node, addressing a Question and supporting or opposing other Claims. One new line of inquiry is one Question node. Each attaches to what it bears on, and that's the contribution.

Specialists become authors. A paper demands generalist scaffolding — introduction, methods, related work, framing, discussion — so the people who hold one sharp contribution often can't be authors on their own terms. The data curator who tracked down a hard-to-find Source, the methodologist who formalized a single instrument, the practitioner with one decisive field observation: each typically has to partner with a generalist who will wrap the piece in apparatus, or watch the contribution go uncredited. The graph removes the apparatus requirement. A Method, an Evidence, a Source, a single Claim is itself a complete, citable, credited contribution. Authorship stops being gated on the ability to produce a whole paper, and the population of people who can author scientific work expands to anyone with one good node.

Credit becomes granular. Each node has its own ID and its own PID — citable independently. A Method, a Source, an Evidence, a Claim can be cited (and tracked) on its own merit. The contributor who proposed C-0017 gets credit when C-0017 is invoked, even when the paper that introduced it isn't. Funders, hiring committees, and citation indexes can resolve attribution to the unit of contribution rather than rolling it up into “lead author of paper X.”

Review becomes a linter; validation becomes topological. Peer review of a paper bundles many things at once — gatekeeping, wording, framing, validating the work, signaling trust to the reader. The bundle dissolves at the node level. Reviewing a node is mostly form-checking: does this Evidence cite the Source it claims, is the Claim it points at really a Claim, is the prose self-contained. Most of that is lintable. The substantive work — what's true, what holds up, what matters — doesn't happen in a review pass; it happens in the graph itself, over time. A weak Claim accumulates opposing Evidence. A strong one accumulates supporting Evidence and Claims that build on it. The trust signal is the topology, not a stamp.

Publishing becomes continuous. A paper waits — for a journal slot, a conference deadline, a grant cycle, an annual report. By the time the work appears it is often eighteen months old, and a counter-finding discovered next week has nowhere to land until the next cycle opens. The graph has no cycle. A new Evidence node ships the day it is found; a counter-Claim ships the day it is formulated; a Question that opens up at midnight is addressable by morning. Publishing tracks the rhythm of inquiry instead of the rhythm of institutions.

Narratives become snapshots. A paper captures the state of the argument at the moment it was written, and that is the state it continues to assert long after the evidence has moved. A narrative composed from the graph is dated by construction. Today's telling reflects today's evidence; next year's telling, regenerated against a graph that has accumulated supporting and opposing evidence in the meantime, is a different telling. Nothing is rewritten — the underlying nodes have moved, and the rendering follows. The narrative is a view of the graph at a moment in time, and another view can be composed whenever it is useful.

The original whitepaper, this site, and each composed narrative all derive from the same node files in graph/.

How to read it

• By topology: /graph shows the whole argument at a glance. Nodes are colored by type; edges are colored by relation. Click any node to inspect its bundle — everything one hop away.
• As narratives: /narratives renders the original whitepaper that seeded the graph, section by section, with each citation linked to its Source node. A toggle at the top swaps to other narratives composed directly from the graph for different audiences and framings.
• By node: every node sits at /node/<ID>. Each page shows the prose body, outbound edges, inbound backlinks, and a deep link to open a GitHub issue about that one node.

How to contribute

Discussion happens at node granularity. Open an issue with the node:<ID> label, or open a pull request that adds a counterclaim, counter-evidence, or a new question. The full contribution model lives in CONTRIBUTING.md.

Further reading

• discoursegraphs.com — the canonical Q/C/E framework and its provenance.
• Discourse graphs and the future of science — Protocol Labs' framing of the form's research-infrastructure implications.
• DiscourseGraphs/schemas — the underlying schema repository this work extends.