fix(synth): rho2024cluster was fabricated — real paper is Rho et al. 2025 ClusterSC

Systematic orphan-audit sprint (156 entries against primary sources)
found one confirmed bib metadata fabrication.

Investigation
-------------
paper.bib claimed:
  @Article{rho2024cluster,
    title={Cluster Synthetic Control Methods},
    author={Rho, Saeyoung},
    journal={arXiv preprint},    # <- NO arXiv id, NO DOI
    year={2024}
  }

Evidence that this paper doesn't exist as described:
  1. No arXiv ID in the bib entry (red flag — every other synth/ arxiv
     reference has one).
  2. Semantic Scholar search "Rho synthetic control cluster" returns
     zero solo-author Rho 2024 papers.
  3. The ONLY real Rho paper on cluster synthetic controls is
     Rho, Tang, Bergam, Cummings, Misra (2025) "ClusterSC: Advancing
     Synthetic Control with Donor Selection", arXiv:2503.21629,
     submitted March 2025 — different year, different title, four
     additional co-authors.

Per §10 "zero fabricated citations" rule, the bib entry is replaced
with the verified real paper.

Changes
-------
paper.bib:
  rho2024cluster → rho2025clustersc
    authors: Rho (solo) → Rho, Tang, Bergam, Cummings, Misra
    title  : "Cluster Synthetic Control Methods" →
             "ClusterSC: Advancing Synthetic Control with Donor Selection"
    year   : 2024 → 2025
    arXiv  : (none) → 2503.21629
    DOI    : (none) → 10.48550/arXiv.2503.21629

src/statspai/synth/__init__.py:24
  prose attribution "(Rho 2024)" → "(Rho et al. 2025, arXiv:2503.21629)
  [@rho2025clustersc]"

src/statspai/synth/scm.py:22, 110
  analogous prose updates; the docstring Reference site gets the
  [@rho2025clustersc] annotation.

PKG-INFO carries a stale "Rho 2024" line too but it's a generated
artifact; left to be regenerated on next release.

Larger audit summary
--------------------
Of 156 total paper.bib orphans this sprint systematically checked:
  - 141 verified CLEAN against primary sources (Phase 1-3)
  -  14 HIGH-RISK 2020+ preprints pending (API rate limits, retry
        in follow-up with longer delays)
  -   1 FLAGGED + FIXED (this commit)

Full status saved to memory:
  ~/.claude/projects/.../memory/project_orphan_audit_status.md

Verification
------------
  coverage:        68.3% -> 68.5%       (336 -> 337 cited keys)
  orphans:         156   -> 155
  citation audit:  316   -> 319 OK     (3 new arXiv:2503.21629
                                         mentions added to prose)

  Gate 0  pytest 81 tests                                  pass
  Gate 1  audit_bib_duplicates --strict   0/0/0            pass
  Gate 2  audit_bib_coverage  --strict-dangling  0 dangling pass
  Gate 3  audit_citations     --strict    319/0/0          pass

Refs verified via arxiv.org primary source (per CLAUDE.md §10).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: brycewang-stanford

paper.bib

@@ -4048,11 +4048,12 @@ @article{qian2025hal
   year={2025}
 }
 
-@article{rho2024cluster,
-  title={Cluster Synthetic Control Methods},
-  author={Rho, Saeyoung},
-  journal={arXiv preprint},
-  year={2024}
+@article{rho2025clustersc,
+  title={{ClusterSC}: Advancing Synthetic Control with Donor Selection},
+  author={Rho, Saeyoung and Tang, Andrew and Bergam, Noah and Cummings, Rachel and Misra, Vishal},
+  journal={arXiv preprint arXiv:2503.21629},
+  year={2025},
+  doi={10.48550/arXiv.2503.21629}
 }
 
 @article{richardson2013single,

src/statspai/synth/__init__.py

@@ -21,7 +21,7 @@
 - **bsts / causal_impact** — Bayesian Structural Time Series (Brodersen et al. 2015)
 - **penscm / abadie_lhour** — Penalized SCM with pairwise discrepancy (Abadie & L'Hour 2021)
 - **fdid / forward_did** — Forward DID with optimal donor selection (Li 2024)
-- **cluster** — Cluster SCM with donor grouping (Rho 2024)
+- **cluster** — Cluster SCM with donor grouping (Rho et al. 2025, arXiv:2503.21629) [@rho2025clustersc]
 - **sparse / lasso** — Sparse SCM with L1 penalties (Amjad, Shah & Shen 2018)
 - **kernel / kernel_ridge** — Kernel-based nonlinear SCM
 

src/statspai/synth/scm.py

@@ -19,7 +19,7 @@
 * **bsts / causal_impact** — Bayesian Structural Time Series (Brodersen et al. 2015)
 * **penscm / abadie_lhour** — Penalized SCM (Abadie & L'Hour 2021)
 * **fdid / forward_did** — Forward DID (Li 2024)
-* **cluster** — Cluster SCM (Rho 2024)
+* **cluster** — Cluster SCM (Rho et al. 2025, arXiv:2503.21629)
 * **sparse / lasso** — Sparse SCM (Amjad, Shah & Shen 2018)
 * **kernel** — Kernel-based nonlinear SCM
 * **kernel_ridge** — Kernel ridge regression SCM
@@ -107,7 +107,7 @@ def synth(
           pairwise discrepancy (Abadie & L'Hour 2021).
         * ``'fdid'`` / ``'forward_did'`` — Forward DID
           (Li 2024).
-        * ``'cluster'`` — Cluster SCM (Rho 2024).
+        * ``'cluster'`` — Cluster SCM (Rho et al. 2025, arXiv:2503.21629). [@rho2025clustersc]
         * ``'sparse'`` / ``'lasso'`` — Sparse SCM
           (Amjad, Shah & Shen 2018).
         * ``'kernel'`` — Kernel-based nonlinear SCM.