Strip meta-language, fix consistency bugs, add duration and scatter charts

- Remove "descriptive evidence," "supplementary checks," "The Great
  Drainage," "The Buffer," "ample reserves," "motive-neutral," and
  reading-order directives from all public copy (README, site, publish
  script).  Rename section headings to neutral labels.
- Fix QT2 state split: publish qt2_low_liquidity from the pipeline
  (QT2-subsample median) so the site no longer silently redefines the
  liquidity state client-side.
- Harmonize debt-limit sample notes with actual data: all 68 quarters
  are in-sample; notes now say debt-limit quarters are flagged for
  inspection, not excluded.
- Remove classification_prior from published quarterly panel and manual
  audit (was "review_required" throughout, communicated nothing).
- Add duration decomposition stacked bar chart showing SOMA runoff,
  coupon issuance, bill offset, and buyback offset with a net overlay.
- Add offset/amplify scatter (Fed runoff DV01 vs Treasury mix DV01)
  with debt-limit quarters marked distinctly.

Author: smkwray

README.md

@@ -2,13 +2,13 @@
 
 **[Live site](https://smkwray.github.io/coordwatch/)**
 
-Tracking how Fed runoff, Treasury issuance, the TGA, ON RRP, and private intermediation interact under ample reserves.
+Tracking Fed balance-sheet runoff, Treasury issuance, reserves, and market intermediation.
 
 ## Question
 
-**How do Fed runoff, Treasury debt management, and liquidity conditions combine under an ample-reserves regime?**
+**How do Fed balance-sheet runoff, Treasury issuance, reserves, and market intermediation move together?**
 
-The effective stance of monetary conditions depends jointly on Fed balance-sheet runoff, Treasury issuance composition, TGA management, ON RRP drainage, and private intermediation capacity. CoordWatch follows the balance sheets using free public data.
+CoordWatch follows Fed runoff, Treasury issuance composition, TGA management, ON RRP drainage, dealer balance sheets, and repo spreads using free public data.
 
 ## What it measures
 
@@ -20,8 +20,6 @@ The effective stance of monetary conditions depends jointly on Fed balance-sheet
 6. How does ON RRP drainage change money-market sensitivity?
 7. How do lower-buffer states differ from higher-buffer states?
 
-Descriptive evidence is the core analysis. The site is organized around reserve drainage, buffers, cash mechanics, duration burden, intermediation, sectoral absorption, and realized issuance. Light econometrics appear only as supplementary checks.
-
 ## Data
 
 All data from free public sources. No proprietary data required.
@@ -65,14 +63,13 @@ make verify        Check all artifacts present
 
 ## Public structure
 
-1. Drainage and buffer conditions.
+1. Fed runoff and liquidity buffers.
 2. Debt-ceiling cash mechanics and the DTS cross-check.
-3. Duration burden, dealer absorption, and repo intermediation.
-4. Sectoral absorption and realized auction mix appendices.
-5. Supplementary econometric checks.
+3. Duration burden, dealer balance sheets, and repo spreads.
+4. Treasury holders and realized auction mix appendices.
+5. Regression results.
 
 ## Conventions
 
 - **Positive Fed pressure** = more duration burden on private sector.
 - **Positive buyback offset** = Treasury removes duration from private hands.
-- **Motive-neutral language.** Measures balance-sheet interactions and market effects; does not infer intent.

pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "setuptools.build_meta"
 [project]
 name = "coordwatch"
 version = "0.1.0"
-description = "Measuring Fed-Treasury operational interdependence under ample reserves"
+description = "Fed runoff, Treasury issuance, reserves, and market intermediation"
 readme = "README.md"
 requires-python = ">=3.10"
 dependencies = [

scripts/15_build_publish_artifacts.py

@@ -108,7 +108,6 @@ def _build_manual_input_audit(manual: pd.DataFrame) -> dict:
             "verification_status",
             "debt_limit_flag",
             "clean_sample_flag",
-            "classification_prior",
             "cash_balance_statement_sourced",
             "reviewer_notes",
         ]
@@ -511,7 +510,7 @@ def main() -> None:
     lp_dealer = pd.read_csv(OUTPUTS_TABLES_DIR / "main_lp_dealer.csv")
     lp_repo = pd.read_csv(OUTPUTS_TABLES_DIR / "main_lp_repo.csv")
 
-    # Publish econometric tables (supporting evidence)
+    # Publish econometric tables
     publish_table(reaction, "reaction_function_main")
     publish_table(lp_dealer, "main_lp_dealer")
     publish_table(lp_repo, "main_lp_repo")
@@ -524,7 +523,8 @@ def main() -> None:
     weekly_cols = [
         "week", "soma_treasuries_bn", "reserves_bn", "tga_bn", "on_rrp_bn",
         "dealer_inventory_bn", "repo_spread_bp", "system_liquidity_bn",
-        "low_liquidity", "qt_runoff_dv01", "qt_runoff_source", "qt_runoff_proxy_bn",
+        "low_liquidity", "qt2_low_liquidity",
+        "qt_runoff_dv01", "qt_runoff_source", "qt_runoff_proxy_bn",
         "coupon_dv01_shock", "bill_dv01_offset", "mix_shock_dv01",
         "buyback_offset_dv01", "expected_soma_redemptions_dv01",
         "duration_pressure_dv01", "fed_pressure_dv01",
@@ -553,10 +553,11 @@ def main() -> None:
         write_json(PUBLISH_DIR / "auction_mix_appendix.json", auction_mix_appendix)
         write_json(ROOT / "site" / "data" / "auction_mix_appendix.json", auction_mix_appendix)
 
-    # Quarterly panel: full
-    publish_table(_ts_to_str(quarter), "quarterly_panel")
+    # Quarterly panel: full (drop classification_prior — not meaningful in published data)
+    quarter_pub = quarter.drop(columns=["classification_prior"], errors="ignore")
+    publish_table(_ts_to_str(quarter_pub), "quarterly_panel")
 
-    # Publish descriptive tables (core evidence)
+    # Publish descriptive tables
     for table_name in ["regime_summary", "episode_summary", "quarterly_descriptive", "correlation_matrix"]:
         path = TABLES_DIR / f"{table_name}.csv"
         if path.exists():
@@ -638,7 +639,7 @@ def main() -> None:
             },
             {
                 "title": "Manual review",
-                "detail": "Episode windows and selected refunding inputs use manual review files. Debt-limit periods are treated as confounded windows rather than clean policy comparisons."
+                "detail": "Episode windows and selected refunding inputs use manual review files. Debt-limit quarters are flagged (debt_limit_flag = 1) for separate inspection but remain in the regression sample."
             },
             {
                 "title": "Sample coverage",
@@ -662,7 +663,7 @@ def main() -> None:
             },
             {
                 "title": "Interpretation",
-                "detail": "Balance-sheet identities and descriptive comparisons are primary. Correlations and regressions are supplementary checks, not stand-alone proof of intent."
+                "detail": "The site reports balance-sheet quantities, market variables, and model estimates from the same public data bundle. Regression tables summarize statistical relationships in the published sample."
             }
         ],
         "data_files": data_files,