docs(citations): 6 verified backfills across bcf/causal_rl/did/synth/transport/conformal

Final sweep of remaining backfillable orphans. All 6 arXiv IDs
independently verified against arxiv.org primary source — author
lists + titles in paper.bib match the paper abstract pages exactly.
No fabrications in this batch.

Verified backfills (single clean site per key)
----------------------------------------------
  cunha2025unifying       arXiv 2512.18135  src/causal_rl/benchmarks.py:2
  han2025synthetic        arXiv 2511.14133  src/synth/survival.py:1
  peng2024local           arXiv 2404.04794  src/did/overlap_did.py:204
  prevot2025hierarchical  arXiv 2508.08418  src/bcf/longitudinal.py:184
  schroder2024conformal   arXiv 2407.03094  docs/guides/conformal_family.md:298
  yang2025integrating     arXiv 2511.19735  src/transport/evidence_synthesis.py:140

For overlap_did and evidence_synthesis the Reference line was just a
bare arXiv id / partial attribution — this commit also rounds out
the line with the full verified author list before appending
[@bibkey], so the rendered docstring is complete ("Peng, Li, Wu & Li
(arXiv:2404.04794, 2024)" instead of just "arXiv:2404.04794 (2024)").

Verification
------------
  coverage:        67.1% -> 68.3%       (330 -> 336 cited keys)
  orphans:         162   -> 156         (6 keys fully consumed)
  suggester queue: 6     -> 0           (no more trivially-safe
                                         backfill candidates remain;
                                         the 156 residual orphans have
                                         no arxiv/DOI occurrence in
                                         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    316/0/0          pass
  import test (5 touched modules)                          pass

Sprint cumulative (batches 1..4)
--------------------------------
  coverage:  64.0%  -> 68.3%             (+4.3pp)
  bibkeys:   315    -> 336                (+21 active citations)
  orphans:   177    -> 156                (-21)
  bib fabrications eliminated:  2         (shi2025bidirectional,
                                           yang2025fortified)
  bib attributions corrected:   1         (vashishtha2024causal →
                                           jiralerspong2024efficient)

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

docs/guides/conformal_family.md

@@ -295,7 +295,7 @@ print(r.dose_curves)       # full curve per test row across the grid
 fertilizer application, time-in-app. Returns a point estimate + band
 at each `(test_row, dose)` pair in the grid.
 
-Citation: Schröder et al. (arXiv:2407.03094, 2024).
+Citation: Schröder et al. (arXiv:2407.03094, 2024). [@schroder2024conformal]
 
 ---
 

src/statspai/bcf/longitudinal.py

@@ -181,7 +181,7 @@ def bcf_longitudinal(
 
     References
     ----------
-    Prevot, Häring, Nichols, Holmes & Ganjgahi (arXiv:2508.08418, 2025).
+    Prevot, Häring, Nichols, Holmes & Ganjgahi (arXiv:2508.08418, 2025). [@prevot2025hierarchical]
     Hahn, Murray, Carvalho (2020), Bayesian Analysis.
     """
     if not isinstance(data, pd.DataFrame):

src/statspai/causal_rl/benchmarks.py

@@ -1,5 +1,5 @@
 """
-Causal RL Benchmark Suite (Cunha, Liu, French & Mian 2025, arXiv 2512.18135).
+Causal RL Benchmark Suite (Cunha, Liu, French & Mian 2025, arXiv 2512.18135). [@cunha2025unifying]
 
 Five canonical benchmarks for evaluating causal RL algorithms under
 unobserved confounding:

src/statspai/did/overlap_did.py

@@ -201,7 +201,7 @@ def dl_propensity_score(
 
     References
     ----------
-    arXiv:2404.04794 (2024).
+    Peng, Li, Wu & Li (arXiv:2404.04794, 2024). [@peng2024local]
     """
     from sklearn.neural_network import MLPClassifier
     X = data[list(covariates)].to_numpy(dtype=float)

src/statspai/synth/survival.py

@@ -1,4 +1,4 @@
-r"""Synthetic Survival Control (Han & Shah 2025, arXiv:2511.14133).
+r"""Synthetic Survival Control (Han & Shah 2025, arXiv:2511.14133). [@han2025synthetic]
 
 Estimates the **survival difference** caused by treatment for a single treated
 unit (or a small set thereof) by constructing a synthetic control on the

src/statspai/transport/evidence_synthesis.py

@@ -137,7 +137,7 @@ def synthesise_evidence(
 
     References
     ----------
-    Dahabreh et al. (2020); arXiv:2511.19735 (2025).
+    Dahabreh et al. (2020); Yang, Gamalo & Fu (arXiv:2511.19735, 2025). [@yang2025integrating]
     """
     if rct_se <= 0 or rwd_se <= 0:
         raise ValueError("SEs must be > 0.")