BERVO - Biological and Environmental Research Variable Ontology

(Formerly the EcoSIM Ontology)

Browse

https://bioportal.bioontology.org/ontologies/BERVO (OLS coming soon)

Editing BERVO

The source of truth for BERVO is the repository on this branch.

Editors should update:

• src/ontology/bervo-src.csv for the ROBOT template that defines BERVO terms
• src/ontology/bervo-edit.owl for the ODK edit file that imports the generated component

All terms are preceded by the BERVO: prefix.

The Google Sheet is still available as a collaboration artifact:

https://docs.google.com/spreadsheets/d/1mS8VVtr-m24vZ7nQUtUbQrN8r-UBy3AwRzTfQsmwVL8/edit?usp=sharing

However, it is no longer the authoritative source for builds or pull requests. Changes should be proposed in this repository, ideally through a GitHub issue and pull request.

To rebuild the generated source component from the tracked CSV, run:

cd src/ontology
make components/bervo-src.owl

To prepare a CSV for uploading back into Google Sheets, run:

cd src/ontology
make export-google-sheet

Working on BERVO with an AI agent

This repository is set up for AI coding agents (Claude Code, Codex, Copilot, Goose).

• AGENTS.md is the canonical instruction file — the source-of-truth rules, the template contract, ID allocation policy, and the pitfalls that matter. CLAUDE.md, .github/copilot-instructions.md, and .goosehints are symlinks to it.
• .claude/skills/ holds task-specific guidance: bervo-terms for curation, bervo-build for builds and QC, bervo-pr-review for reviewing changes.
• .claude/commands/ provides /add-term, /qc, and /pm.
• A PostToolUse hook validates bervo-src.csv automatically after any edit.
• Mentioning @claude on an issue or PR triggers a response, and PRs touching ontology content or agent instructions get an automated review. Both need ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN in the repository secrets, and are skipped without either.
• The agents post as github-actions[bot] by default. To give them their own identity, create a GitHub App, install it on this repository, and add its credentials as the AI4C_AGENT_APP_ID and AI4C_AGENT_PRIVATE_KEY secrets — the workflows pick them up automatically and fall back to the default token when they are absent. See docs/agent-identity.md.

Commands

Common tasks are exposed through just (just --list for all):

just validate                # structural checks on bervo-src.csv
just find "soil carbon"      # search IDs, labels, definitions, synonyms, EcoSIM names
just next-id 0               # next free variable ID
just show BERVO:0000001      # inspect one term
just stats                   # term counts by ID block and category
just test                    # run the test suite
just build                   # rebuild the OWL component from the CSV

just validate runs src/scripts/validate_bervo_src.py, which checks the invariants ROBOT does not: ragged rows, duplicate IDs and labels, malformed IDs, and references to terms that do not exist.

Methods

See also this slide deck

Seeding of initial parameter list

chatgpt ADA was used to create a program to iterate through the bervo fortran codebase and generate an obo format file of all parameter codes plus their names.

IDs of the form BERVO:<CODE> were created

Note: in future these may be translated to numeric IDs but for now the codes are convenient

Generation of definitions

The OAK generate-definitions command was used to generate definitions for all terms

Generation of grouping classes

Each parameter was organized into a grouping class.

We used Claude due to the large context window. A csv of all CODE-label pairs were uploaded to Claude, Claude then suggested groupings for these. These were examined in text format, we then asked Claude to convert to OBO format.

Inferring linkages to other concepts

We curated a handful of OBO stanzas where we linked each parameter to other concepts.

This was loaded into a curategpt database, to serve as in-context examples.