ER Commons

er-commons is a small, reproducible Python workspace for environmental-review data work. The first planned capability is er_bench, a CEQA-oriented benchmark and evaluation harness. That benchmark is deliberately the first component, not the project boundary: later work may add data preparation or other reusable environmental-review workflows when a scoped task justifies them.

The repository favors maintained open-source tools and thin, explicit glue code. Before new custom code is added, a task should establish that an existing package, format, or command-line tool does not cleanly solve the need.

Current milestone

The benchmark has 49 questions: 19 development and 30 test. Presented model results cover Luna and Haiku, with three answers per test question and sealed figures and tables. Development and synthetic judge diagnostics remain separate.

Task 14 is active for a small, structured review of saved answers with the user. Task 15 will turn the results and confirmed examples into a short methods/results report and remains provisional.

Use the docs index for accepted inputs and status, the queue for next actions, and the reporting maintainer guide for charts, tables and future model comparisons. Generated reports live outside Git.

Quick start

make bootstrap
make about
make paths
make check

Project commands require ER_COMMONS_DATA_ROOT in a local .env file. This checkout is configured for /Volumes/x10pro/er_commons; use .env.example to create the file on another machine. make loads the value and validates it before commands that use project settings.

Layout

er_commons/                         # Git repository: code, docs, configs, tests
  src/er_commons/                   # Small package-backed command-line glue
  pipelines/                        # Versioned pipeline specifications and notes
  benchmarks/er_bench/              # Versioned benchmark contract and schemas
  configs/                          # Small, reviewable configuration files
  docs/                             # Product, architecture, planning, decisions
  tasks/                            # Narrow agent-sized implementation contracts

/Volumes/x10pro/er_commons/         # Not tracked by Git: data and run artifacts
  datasets/ceqa/                    # CEQA source files and normalized derivatives
  pipelines/                        # Pipeline manifests and generated run outputs
  benchmarks/er_bench/              # Benchmark inputs, snapshots, and run results

See docs/index.md for routing and current status. Agents should start with AGENTS.md.