Add PUF + source impute modules, fix AGI ceiling (issue #530)

Makefile

@@ -1,4 +1,4 @@
-.PHONY: all format test install download upload docker documentation data publish-local-area clean build paper clean-paper presentations database database-refresh promote-database promote-dataset
+.PHONY: all format test install download upload docker documentation data calibrate publish-local-area clean build paper clean-paper presentations database database-refresh promote-database promote-dataset
 
 HF_CLONE_DIR ?= $(HOME)/huggingface/policyengine-us-data
 
@@ -97,6 +97,10 @@ data: download
 	python policyengine_us_data/datasets/cps/small_enhanced_cps.py
 	python policyengine_us_data/datasets/cps/local_area_calibration/create_stratified_cps.py
 
+calibrate: data
+	python -m policyengine_us_data.calibration.unified_calibration \
+		--puf-dataset policyengine_us_data/storage/puf_2024.h5
+
 publish-local-area:
 	python policyengine_us_data/datasets/cps/local_area_calibration/publish_local_area.py
 

changelog_entry.yaml

@@ -0,0 +1,12 @@
+- bump: minor
+  changes:
+    added:
+    - PUF clone + QRF imputation module (puf_impute.py) with state_fips predictor and stratified subsample preserving top 0.5% by AGI
+    - ACS re-imputation module (source_impute.py) with state predictor; SIPP/SCF imputation without state (surveys lack state identifiers)
+    - PUF and source impute integration into unified calibration pipeline (--puf-dataset, --skip-puf, --skip-source-impute flags)
+    - 21 new tests for puf_impute and source_impute modules
+    - DC_STATEHOOD=1 environment variable set in storage/__init__.py to ensure DC is included in state-based processing
+    changed:
+    - Refactored extended_cps.py to delegate to puf_impute.puf_clone_dataset() (443 -> 75 lines)
+    - PUF QRF training uses stratified subsample (20K target) instead of random subsample(10_000), force-including high-income tail
+    - unified_calibration.py pipeline now supports optional source imputation and PUF cloning steps