Access, handoffs, duplicated work, incompatible systems, and unclear approvals.
Session theme · Independent laboratories, coordinated learning
Coordinate research.
Keep scientific control.
How can independent biotech labs work together without centralizing their data or surrendering scientific control?
LRN is exploring a coordination layer for sharing requests, capabilities, status, approved evidence, and provenance—while sensitive data, methods, judgment, custody, and authority remain local.
Sensitive data, scientific judgment, methods, custody, and institutional authority.
Requests, capabilities, status, approved aggregates, evidence, and provenance.
Coordination, preparation, validated computation, and progress tracking.
One narrow tauopathy workflow, named owners, success measures, and a 30–60 day step.
Why this network
Science is becoming executable.
Agents can already search literature, use databases, write analyses, coordinate experiments, and interact with laboratory systems. The central challenge is no longer more information. It is knowing what changed, why it matters, what evidence supports it, and what a research team can inspect or try.
LRN is being designed as a shared intelligence and coordination layer for scientific work that must remain inspectable.
It joins publication, evidence, people, capabilities, and executable artifacts without pretending that an automated recommendation is a scientific approval.
LocalData, methods, scientific judgment, laboratory custody, and release authority.
SharedRequests, capability descriptions, status, approved results, evidence, and provenance.
AutomatedPreparation, routing, deterministic execution, progress tracking, and evidence assembly.
HumanPurpose, interpretation, exceptions, scientific validation, and consequential release.
One source · Many experiences
Everything begins with a verified Signal.
A Signal is not an article. It is a structured record of an event, its evidence, its participants, its artifacts, and what can be investigated next. Editorial formats are views over that same record.
SourcesPapers · repositories · datasets · programs · talks
Verified SignalsExplicit claim separated from editorial inference
Living research graphPeople · labs · projects · artifacts · relationships
Useful interfacesNewspaper · audio · visual · feeds · agents · experiments
What one Signal records
Proposed record contractWhat happened
Why it matters
Explicit evidence versus inference
People and laboratories
Projects and collaborators
Papers and preprints
Repositories and commits
Datasets and benchmarks
Funding programs and calls
Architecture and practices revealed
Licensing and openness
What can be reproduced locally
Science Agent Daily
A small daily edition about meaningful change.
Not a summary of the internet. A quiet repository commit that exposes a validation architecture may matter more than a major announcement. Each edition should be short enough to finish and deep enough to act on.
Five things worth knowing. Three things you can try. One thing your team should discuss.
5Meaningful changes
3Inspect or try
1Team question
Reported computation is not executed computation.
What happened
A scientific-agent result remains a claim until its code, inputs, environment, and outputs are inspectable.
Why we care
Comparing final answers is weaker than comparing reproducible artifacts and their provenance.
Try it
Run two systems against the same evidence and compare manifests, executed code, outputs, and failure records.
Editorial interfaces
The same evidence, experienced differently.
These are product directions, not claims of present availability. Each format must preserve links back to evidence and make uncertainty visible.
The Daily Newspaper
What happened?Why care?What can we try?
A deliberately small edition that favors consequential technical changes over announcement volume.
Editorial prototypeInteractive Podcast
Show evidenceCompare projectsGo deeper
A default listening path that can pause, answer a question, inspect an artifact, and return to the episode.
Experience previewVisual Briefing
Inspect figureOpen repositoryCheck hardware fit
A responsive research briefing where the screen follows the paper, project, architecture, or executable artifact under discussion.
Experience previewCoffee Bites
Thirty secondsOne ideaOne action
Compact research packets that state what changed, why it matters, the reusable idea, and a concrete way to inspect it.
Format prototypeAgents are participants too
Structured intelligence should not require scraping articles.
The same Signal that supports human reading can support feeds and research-agent queries. Machine access must preserve source boundaries, uncertainty, licensing, and the difference between a provider claim and independent validation.
Ask the research graph directly.
Conventional RSS and JSON feeds can sit beside richer interfaces for authorized research agents. Access to metadata never implies access to restricted data.
Show biomedical-agent changes from the last seven days with public code and something runnable on a 4090.
What changed this week that affects an autonomous-laboratory architecture?
Find projects related to a named system, expand verified collaborators, and identify a reproducible next step.
News becomes executable context
A story can earn a Research Capsule.
When licensing, inputs, and infrastructure permit, a publication can point to an inspectable execution package rather than ending at a link. The labels below describe intended evidence states, not currently available downloads.
signal/ ├── signal.json ├── README.md ├── sources.json ├── run.sh ├── environment.yml ├── inputs/ ├── expected/ ├── evaluate.py └── provenance.json
Evidence states
SipUnderstand the development and its evidence.
TryA documented path appears runnable locally.
ReproducedOur infrastructure successfully executed the recorded path.
ExtendA concrete experiment or research direction follows.
WatchImportant, but not yet actionable.
InferredEditorial interpretation rather than an explicit source claim.
The mental model
Organize the field around the scientific loop.
The emerging AI scientist is not one model. It is a system of models, scientific tools, deterministic execution, evidence, evaluation, provenance, infrastructure, and human judgment.
01Find
02Understand
03Hypothesize
04Design
05Execute
06Measure
07Verify
08Remember
09Repeat
Scientific generation is becoming cheap. Reliable scientific validation is not.
Strong modelsSpecialized scientific toolsStructured evidenceReproducible executionExplicit evaluationPersistent provenanceHuman scientific judgmentOwner-controlled release
Coming into focus
What this public preview is pointing toward.
No launch dates, provider relationships, or operational access are implied here. Each direction must earn publication through source review, governance, and working evidence.
The Newspaper
A concise daily map of meaningful changes in AI-enabled scientific work.
Morning Coffee
An interruptible research discussion connected to the underlying evidence.
The Briefing
A visual edition that follows papers, people, architectures, repositories, and results.
Research Graph
A persistent map of people, labs, projects, papers, datasets, benchmarks, and programs.
Agent Access
Structured, permission-aware access to the same verified intelligence.
Try It
A path from a story to an artifact that can be inspected, reproduced, or extended.
Know what matters. Inspect the evidence. Participate without surrendering control.
Science is moving faster. The Learning Research Network is exploring how people and their agents can understand it, challenge it, reproduce it, and build on it together.