How bis-med-it/marco-bis works
MaRCo@BIS is an AI-assisted macroeconometric research framework, combining a human-confirmed research workflow with a Python estimation library and canonical replications. The main journey is a researcher’s question through mode triage, model/specification decisions, data preparation, estimation, diagnostics, and reporting; users can then curate or continue the work. The README documents the workflow and broader capabilities; source-level edges below are limited to relationships visible in excerpts or the source index, so unsampled integrations are not inferred.
Main entry points: loader.py Dataset loading · SKILL.md Mode triage · fetcher.py Fetch and cache · identification.py Shock identification
Components 23
Research workflow
- Knowledge context
skills - Mode triage
skills/orchestrator/SKILL.md - Model or paper entry
skills/model-dictionary/entries - Specification
scripts/lib/specs.py - Confirmation gates
skills/orchestrator/SKILL.md
Data and preparation
- Data sources
scripts/lib/data/fetchers - Fetch and cache
scripts/lib/data/fetcher.py - Dataset registry
scripts/lib/data/registry.py - Dataset loading
scripts/lib/data/loader.py - Transforms
scripts/lib/data/transforms.py - Preprocessing
scripts/lib/preprocess
Econometric analysis
- VAR and BVAR
scripts/lib/var.py - Advanced models
scripts/lib - Shock identification
scripts/lib/identification.py - IRF inference
scripts/lib/inference.py - Forecasting
scripts/lib/forecast.py
Results and curation
- Diagnostics
scripts/lib/diagnostics.py - Sensitivity checks
scripts/lib/sensitivity.py - Reports
scripts/lib/report.py - Canonical replications
scripts/replications - Disposition and curation
skills
Other components
- Researcher
- Claude Code agent
Connections 29
- Researcher → Claude Code agent: asks question
README.md - Claude Code agent → Knowledge context: loads context
README.md - Claude Code agent → Mode triage: triages mode
README.md - Mode triage → Model or paper entry: retrieves or drafts
README.md - Mode triage → Specification: designs specification
README.md - Specification → Confirmation gates: requests confirmation
README.md - Confirmation gates → Researcher: confirms decisions
README.md - Claude Code agent → Data sources: selects data
README.md - Data sources → Fetch and cache: provides series
README.md - Fetch and cache → Dataset registry: registers dataset
scripts/lib/data/fetcher.py - Fetch and cache → Dataset loading: provides cached data
scripts/lib/data/fetcher.py - Dataset loading → Dataset registry: reads registry
scripts/lib/data/loader.py - Dataset loading → Transforms: uses transforms
scripts/lib/data/loader.py - Dataset loading → Preprocessing: prepares data
- Claude Code agent → Dataset loading: loads data
README.md - Claude Code agent → VAR and BVAR: runs estimation
README.md - Claude Code agent → Advanced models: runs estimation
README.md - VAR and BVAR → Shock identification: supplies VAR fit
scripts/lib/identification.py - Shock identification → IRF inference: draws uncertainty
scripts/lib/identification.py - Shock identification → VAR and BVAR: uses VAR utilities
scripts/lib/identification.py - Claude Code agent → Forecasting: forecasts
README.md - Claude Code agent → Diagnostics: checks results
README.md - Claude Code agent → Sensitivity checks: tests sensitivity
README.md - Diagnostics → Reports: summarizes findings
scripts/lib/report.py - Claude Code agent → Reports: generates report
README.md - Model or paper entry → Canonical replications: selects replication
README.md - Claude Code agent → Canonical replications: runs script
README.md - Reports → Researcher: presents results
README.md - Researcher → Disposition and curation: chooses disposition
README.md