microplex-us

US-specific survey adapters, calibration targets, pipelines, and PolicyEngine integration built on top of the generic microplex engine.

Docs

• Docs index
• Architecture
• Source semantics
• Imputation conditioning contract
• Benchmarking
• Methodology ledger
• PolicyEngine oracle compatibility path
• PE construction parity
• Superseding policyengine-us-data

Diagnostics dashboard

The static dashboard in dashboard/ loads the full PE-native per-target diagnostic JSON written by:

microplex-us-pe-native-target-diagnostics \
  --from-dataset /path/to/enhanced_cps_2024.h5 \
  --to-dataset /path/to/policyengine_us.h5 \
  --policyengine-targets-db /path/to/policy_data.db \
  --output-path artifacts/pe_native_target_diagnostics_current.json

The dashboard uses the exported Cosilico design tokens from @cosilico/config/theme.css; run python scripts/sync_cosilico_theme.py --check to verify the local browser-readable token copy is still synced. When a PolicyEngine target DB is available, the JSON annotates PE-native legacy labels with structured target IDs and flags legacy-only gaps.

Current focus

microplex-us is being built as a library-first US runtime with policyengine-us as the shared measurement operator and policyengine-us-data as the incumbent comparator, not as the thing we are trying to clone wholesale:

• canonical source and target metadata
• PE-US-compatible export
• full-target benchmarking against the active targets DB
• run registry and DuckDB index for frontier analysis

The architecture is still evolving, so the docs are deliberately technical and operational rather than paper-like.

Method-level decomposable-family bakeoffs now live in the sibling eval repo: /Users/maxghenis/CosilicoAI/microplex-evals. microplex-us should keep the runtime helpers and pipeline-adjacent diagnostics, not the long-lived eval orchestration and artifact curation.