Refresh reference outputs and clean repo artifacts

Author: MaxGhenis

.gitignore

@@ -10,6 +10,7 @@ results/**/rerun_*/
 results/reference_outputs.csv
 results/ground_truth.csv
 results/scenarios.csv
+output/
 
 # Paper render scratch outputs. The served paper artifacts live under
 # app/public/paper/ and frozen manuscript inputs live under paper/snapshot/.

README.md

@@ -83,4 +83,3 @@ administrative truth. It also writes `results/local/scenarios.csv`, and the eval
 commands reuse that manifest by default instead of regenerating households from
 the current source dataset. Prediction CSVs also get a `.meta.json` sidecar so
 resumes only happen against the exact same manifest, model set, and program set.
-`policybench ground-truth` remains as a compatibility alias.

RESULTS.md

@@ -14,7 +14,7 @@ policybench analyze --output-dir results/local/analysis
 ```
 
 The first command writes PolicyEngine reference outputs, not administrative
-truth. `policybench ground-truth` remains as a compatibility alias.
+truth.
 
 ## Full runbook
 
@@ -53,7 +53,7 @@ for country in us uk; do
   cp "$RUN_DIR/$country/no_tools_chunked/by_model/"*.csv "$RUN_DIR/$country/by_model/"
 done
 
-python scripts/export_full_run.py --run-dir "$RUN_DIR"
+policybench export-full-run --run-dir "$RUN_DIR"
 ```
 
 For first-pass cost control, run the same commands with `-n 100` in a separate