Research Lab

ℹ Illustrative scenario — a composite archetype, not a real named adopter.

Meet Dr. Maya Chen

Maya is a principal investigator running a computational biology lab with 8 researchers, 3 AI agents, and a growing mountain of experimental data. She publishes 10+ papers a year and her lab generates protocols, datasets, analysis pipelines, and collaboration notes faster than anyone can organize them.

The Challenge

Maya’s lab knowledge lives everywhere: Slack threads, Google Docs, Jupyter notebooks, a shared drive with 2,000 files in folders named things like “data_final_v3_REAL.” New postdocs spend their first month just figuring out where things are. AI agents asked to help with analysis produce generic output because they have no context about the lab’s methods, conventions, or active projects.

How aDNA Helps

Maya adopts aDNA as the lab’s knowledge architecture. She creates a bare-triad vault and extends the ontology with domain-specific types:

Knowledge layer (what/): The context library gets subtopics on the lab’s core methods — protein folding, binding assays, computational docking. Each file is 150-300 lines, table-driven, quality-scored. When an agent starts a new analysis, it loads the relevant context and produces lab-specific output instead of textbook answers.

Operations layer (how/): Experiments become missions with objectives. A 6-month project tracking protein binder candidates is a campaign with phased execution. Session tracking creates an audit trail — who did what, when, and what happened.

People layer (who/): Lab governance defines agent roles and human oversight policies. Coordination notes handle handoffs between researchers working different shifts.

What Their Vault Looks Like

chen_lab.aDNA/
├── what/
│   ├── context/          # 8 topics: protein_folding, binding, docking, ...
│   ├── protocols/        # [EXT] 25 experimental protocols
│   ├── datasets/         # Dataset metadata + lineage tracking
│   ├── lattices/         # Analysis pipelines (lattice YAML)
│   └── decisions/        # Method choices with rationale
├── how/
│   ├── campaigns/        # Multi-month research projects
│   ├── missions/         # Individual experiments
│   ├── sessions/         # Audit trail
│   └── templates/        # Protocol templates, report templates
└── who/
    ├── governance/       # Lab policies, agent oversight
    ├── team/             # 8 researchers + 3 agents
    └── coordination/     # Shift handoff notes

Outcome

New postdocs orient in a day instead of a month — they read CLAUDE.md, browse the context library, and understand the lab’s methods. AI agents produce lab-specific analysis instead of generic output. The protocol library eliminates “how did we do this last time?” conversations. Federation lets Maya share validated analysis lattices with collaborating labs without emailing YAML files.

Typical Ontology Extensions

Entity	Triad	Purpose
`protocol`	what/	Experimental protocols (wet lab + computational)
`dataset`	what/	Dataset metadata, lineage, and FAIR annotations
`experiment`	how/	Individual experiment records with methods and results
`collaboration`	who/	Cross-lab coordination and data sharing agreements

Where to go next

A reading path, in order. Each tutorial is self-contained.

Write a Context File — 30 min · Intermediate — curate a quality-rubric-scored knowledge file your agents load before working on a specific domain or dataset.
Extend the Ontology — 25 min · Intermediate — add domain-specific entity types: experiment, protocol, finding, hypothesis — whatever your research taxonomy demands.
Design a Mission — 25 min · Intermediate — decompose a multi-week analysis arc into claimable objectives. Each objective fits in a single agent session.
Build a Lattice — 30 min · Advanced — compose a research workflow as a validated .lattice.yaml graph of modules. Machine-executable, human-readable.

Governance Files — how CLAUDE.md orients new lab members
FAIR Metadata — how aDNA satisfies Findable, Accessible, Interoperable, Reusable at the knowledge-architecture level
The Convergence Model — load exactly the context a task needs, not the entire dataset catalog
Federation Readiness — the six-point checklist that gates cross-lab sharing
Mission — the unit that decomposes a multi-week arc into claimable objectives