docs(citations): correct upstream attributions for 5 estimators

Verified each reference via arXiv / NBER primary sources. Changes are
metadata-only — algorithm paths, signatures, and numerics unchanged.

* bcf_longitudinal (arXiv:2508.08418): "Alessi, Zorzetto et al." ->
  Prevot, Haring, Nichols, Holmes & Ganjgahi (2025)
* surrogate_index (NBER WP 26463): "Pollmann, Taubinsky" ->
  Athey, Chetty, Imbens & Kang (2019)
* proximal_surrogate_index + bridge(surrogate_pci): Imbens-Kallus-Mao
  arXiv:2601.17712 (2026) -> Imbens, Kallus, Mao & Wang,
  JRSS-B 87(2) 2025, arXiv:2202.07234 (fix arXiv ID + add Wang)
* bayes_dml (arXiv:2508.12688): "Chernozhukov et al." ->
  DiTraglia & Liu (2025); DML framework attribution kept
* causal_bandit (NeurIPS 2015): "Bareinboim & Pearl" ->
  Bareinboim, Forney & Pearl

pytest on touched modules (bcf_longitudinal / surrogate / bayes_dml /
causal_rl_core / bridge / bridge_full): 43 passed / 0 failed.

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

Author: brycewang-stanford

docs/guides/bridging_theorems.md

@@ -25,14 +25,14 @@ disagree beyond sampling error.
 
 ## The six bridges
 
-| `kind`            | Theorem                                          | Reference                                 |
-|-------------------|--------------------------------------------------|-------------------------------------------|
-| `did_sc`          | DiD ≡ Synthetic Control                          | Shi & Athey, arXiv:2503.11375 (2025)       |
-| `ewm_cate`        | EWM ≡ CATE-max policy                            | Ferman et al., arXiv:2510.26723 (2025)     |
-| `cb_ipw`          | Covariate Balancing ≡ IPW ≡ DR                   | Zhao & Percival, arXiv:2310.18563 v6 (2025)|
-| `kink_rdd`        | Bunching ≡ Kink RDD first-order expansion        | Lu, Wang, Xie, arXiv:2404.09117 (2025)     |
-| `dr_calib`        | Doubly-robust ≡ outcome + Riesz joint calibration | Zhang et al., arXiv:2411.02771 (2025)     |
-| `surrogate_pci`   | Long-term Surrogate Index ≡ PCI                  | Kallus & Mao, arXiv:2601.17712 (2026)      |
+| `kind`          | Theorem                                           | Reference                                                       |
+|-----------------|---------------------------------------------------|-----------------------------------------------------------------|
+| `did_sc`        | DiD ≡ Synthetic Control                           | Shi & Athey, arXiv:2503.11375 (2025)                            |
+| `ewm_cate`      | EWM ≡ CATE-max policy                             | Ferman et al., arXiv:2510.26723 (2025)                          |
+| `cb_ipw`        | Covariate Balancing ≡ IPW ≡ DR                    | Zhao & Percival, arXiv:2310.18563 v6 (2025)                     |
+| `kink_rdd`      | Bunching ≡ Kink RDD first-order expansion         | Lu, Wang, Xie, arXiv:2404.09117 (2025)                          |
+| `dr_calib`      | Doubly-robust ≡ outcome + Riesz joint calibration | Zhang et al., arXiv:2411.02771 (2025)                           |
+| `surrogate_pci` | Long-term Surrogate Index ≡ PCI                   | Imbens, Kallus, Mao & Wang, JRSS-B 87(2) 2025; arXiv:2202.07234 |
 
 Each bridge is importable at top level as `sp.bridge(kind="..." ...)` or
 via the per-module function in `statspai.bridge.*`.
@@ -186,13 +186,14 @@ explicit.
 
 ---
 
-## 6. Long-term Surrogate Index ≡ Proximal Causal Inference — Kallus-Mao (2026)
+## 6. Long-term Surrogate Index ≡ Proximal Causal Inference — Imbens-Kallus-Mao-Wang (2025)
 
-Surrogate indices (Athey-Chetty-Imbens, 2020) use short-term
-measurements as proxies for long-term outcomes. Kallus-Mao show that
-under a completeness condition, the surrogate-index estimand is
-*identical* to a proximal-causal-inference (PCI) estimand using the
-same short-term variables as proxies for an unobserved confounder.
+Surrogate indices (Athey-Chetty-Imbens-Kang, NBER WP 26463, 2019) use
+short-term measurements as proxies for long-term outcomes. Imbens,
+Kallus, Mao & Wang show that under a completeness condition, the
+surrogate-index estimand is *identical* to a proximal-causal-inference
+(PCI) estimand using the same short-term variables as proxies for an
+unobserved confounder.
 
 ```python
 r = sp.bridge(

docs/guides/proximal_family.md

@@ -155,12 +155,13 @@ single-control versions.
 
 ---
 
-## Long-term surrogate + PCI — `sp.proximal_surrogate_index` (Kallus-Mao, 2026)
+## Long-term surrogate + PCI — `sp.proximal_surrogate_index` (Imbens-Kallus-Mao-Wang, 2025)
 
 You ran a randomised experiment for 3 months but care about the 2-year
-outcome.  The classical solution (Athey-Chetty-Imbens 2020 surrogate index)
-needs the surrogates to fully mediate the long-term effect — a strong
-assumption.  Kallus-Mao show that combining the surrogate index with a PCI
+outcome.  The classical solution (Athey-Chetty-Imbens-Kang, NBER WP 26463,
+2019 surrogate index) needs the surrogates to fully mediate the long-term
+effect — a strong assumption.  Imbens, Kallus, Mao & Wang show that
+combining the surrogate index with a PCI
 layer on the **observational** data lets you drop the full-mediation
 requirement: short-term surrogates play the role of `W`, observational
 proxies play the role of `Z`, and the two together identify the long-term
@@ -181,7 +182,9 @@ r = sp.proximal_surrogate_index(
 This is also available as the `surrogate_pci` bridge in `sp.bridge` — see
 [Bridging theorems](bridging_theorems.md).
 
-Citation: Kallus & Mao (2026), arXiv:2601.17712.
+Citation: Imbens, Kallus, Mao & Wang (2025). "Long-term Causal Inference
+Under Persistent Confounding via Data Combination." *Journal of the Royal
+Statistical Society Series B* 87(2), 362-388. arXiv:2202.07234.
 
 ---
 

src/statspai/bayes/dml.py

@@ -1,4 +1,4 @@
-"""Bayesian Double Machine Learning (arXiv:2508.12688, 2025).
+"""Bayesian Double Machine Learning (DiTraglia & Liu, arXiv:2508.12688, 2025).
 
 Wraps :func:`sp.dml` cross-fitted residuals in a Normal-Normal conjugate
 posterior, turning frequentist DML into a fully Bayesian analogue with
@@ -120,7 +120,10 @@ def bayes_dml(
 
     References
     ----------
-    Chernozhukov et al. (arXiv:2508.12688, 2025).
+    DiTraglia, F. J., & Liu, L. (2025). "Bayesian Double Machine
+    Learning for Causal Inference." arXiv:2508.12688.
+    (The underlying orthogonal-moments DML construction is due to
+    Chernozhukov et al. 2018, Econometrics Journal.)
     """
     if mode not in ("conjugate", "full"):
         raise ValueError(f"mode must be 'conjugate' or 'full'; got {mode!r}.")

src/statspai/bcf/longitudinal.py

@@ -1,8 +1,8 @@
 """
 BCFLong — hierarchical Bayesian Causal Forest for longitudinal data.
 
-Alessi, Zorzetto et al. (arXiv:2508.08418, 2025) extend BCF to
-longitudinal/panel data with a two-level hierarchy:
+Prevot, Häring, Nichols, Holmes & Ganjgahi (arXiv:2508.08418, 2025)
+extend BCF to longitudinal/panel data with a two-level hierarchy:
 
 * Level 1 (observation): ``Y_{it} = mu_t(X_{it}) + tau_t(X_{it}) D_{it} + u_i + e_{it}``
 * Level 2 (subject)    : ``u_i ~ Normal(0, sigma_u^2)`` (random intercept)
@@ -15,7 +15,10 @@
 
 References
 ----------
-Alessi, Zorzetto et al. (arXiv:2508.08418, 2025).
+Prevot, E., Häring, D. A., Nichols, T. E., Holmes, C. C., & Ganjgahi, H.
+(2025). "A hierarchical modelling approach for Bayesian Causal Forests
+on longitudinal data: A Case Study in Multiple Sclerosis Clinical
+Trials." arXiv:2508.08418.
 """
 
 from __future__ import annotations
@@ -178,7 +181,7 @@ def bcf_longitudinal(
 
     References
     ----------
-    Alessi, Zorzetto et al. (arXiv:2508.08418, 2025).
+    Prevot, Häring, Nichols, Holmes & Ganjgahi (arXiv:2508.08418, 2025).
     Hahn, Murray, Carvalho (2020), Bayesian Analysis.
     """
     if not isinstance(data, pd.DataFrame):
@@ -354,6 +357,6 @@ def _single_pass(df_in: pd.DataFrame, seed: int) -> tuple:
             "n_trees_mu": n_trees_mu,
             "n_trees_tau": n_trees_tau,
             "n_bootstrap_effective": len(boot_point),
-            "reference": "Alessi, Zorzetto et al. (arXiv:2508.08418, 2025)",
+            "reference": "Prevot, Häring, Nichols, Holmes & Ganjgahi (arXiv:2508.08418, 2025)",
         },
     )

src/statspai/bridge/__init__.py

@@ -8,14 +8,14 @@
 the estimate is consistent.
 
 Six bridges shipped (per arXiv 2503.11375 / 2510.26723 / 2310.18563 v6 /
-2404.09117 / 2411.02771 / 2601.17712, all 2025-2026):
+2404.09117 / 2411.02771 / 2202.07234, 2022-2025):
 
 * ``did_sc``        — DiD ≡ Synthetic Control (Shi-Athey 2025)
 * ``ewm_cate``      — EWM ≡ CATE → policy (Ferman et al. 2025)
 * ``cb_ipw``        — Covariate Balancing ≡ IPW × DR (Zhao-Percival 2025)
 * ``kink_rdd``      — Kink-Bunching ≡ RDD (Lu-Wang-Xie 2025)
 * ``dr_calib``      — Doubly Robust via Calibration (Zhang 2025)
-* ``surrogate_pci`` — Long-term Surrogate ≡ PCI (Kallus-Mao 2026)
+* ``surrogate_pci`` — Long-term Surrogate ≡ PCI (Imbens-Kallus-Mao-Wang 2025, JRSS-B)
 
 The unified entry point is ``sp.bridge(data, kind=..., **kwargs)``,
 returning a :class:`BridgeResult` reporting the two path estimates,

src/statspai/bridge/surrogate_pci.py

@@ -1,6 +1,6 @@
 """
-Bridge: Long-term Surrogate ≡ Proximal Causal Inference (Kallus-Mao
-2026, arXiv 2601.17712).
+Bridge: Long-term Surrogate ≡ Proximal Causal Inference (Imbens, Kallus,
+Mao & Wang 2025, JRSS-B 87(2) 362-388; arXiv 2202.07234).
 
 Both paths target the long-term ATT under unobserved confounding via
 short-term proxy variables. Surrogate Index assumes the proxy fully
@@ -85,7 +85,7 @@ def _surrogate(Yi, Di, Si, Xi):
         att_surr = np.nan
 
     # ---------- Path B: PCI bridge (two-model counterfactual) ---------- #
-    # Kallus-Mao (2026) show that under proxy completeness the bridge
+    # Imbens-Kallus-Mao-Wang (2025) show that under proxy completeness the bridge
     # function b(W, X) = E[Y | W, X, D = 0] identifies the ATT via
     #     ATT = E[b(W, X) | D = 1] - E[Y | D = 0].
     # We go one step further and fit a symmetric bridge on BOTH arms —
@@ -159,5 +159,5 @@ def _pci(Yi, Di, Si, Xi):
         se_dr=se_dr,
         n_obs=n,
         detail={"n_short_term": len(short_term)},
-        reference="Kallus-Mao (2026), arXiv 2601.17712",
+        reference="Imbens, Kallus, Mao & Wang (2025), JRSS-B 87(2); arXiv 2202.07234",
     )

src/statspai/causal_rl/core.py

@@ -3,7 +3,7 @@
 
 Three building blocks on top of :mod:`statspai.causal_rl`:
 
-- :func:`causal_bandit` — Bareinboim-Pearl contextual causal bandit.
+- :func:`causal_bandit` — Bareinboim-Forney-Pearl contextual causal bandit.
 - :func:`counterfactual_policy_optimization` — Oberst-Sontag 2019 style
   counterfactual policy evaluation via SCM noise inversion (Gaussian
   linear special case).
@@ -124,7 +124,7 @@ def causal_bandit(
     n_samples: int = 500,
     rng_seed: int = 0,
 ) -> CausalBanditResult:
-    """Bareinboim-Pearl contextual causal bandit.
+    """Bareinboim-Forney-Pearl contextual causal bandit.
 
     Given a callable ``reward_fn(arm, context)`` that samples the
     potential outcome of an arm under the current context, Monte Carlo

src/statspai/registry.py

@@ -2652,7 +2652,7 @@ def _build_registry():
         name="surrogate_index",
         category="surrogate",
         description=(
-            "Athey-Chetty-Imbens surrogate-index estimator for the "
+            "Athey-Chetty-Imbens-Kang surrogate-index estimator for the "
             "long-term ATE: combines an experimental sample (treatment + "
             "short-term surrogate) with an observational sample "
             "(surrogate + long-term outcome) to extrapolate the effect on "
@@ -2677,7 +2677,7 @@ def _build_registry():
         ),
         tags=["surrogate", "long_term", "causal", "ate"],
         reference=(
-            "Athey, Chetty, Imbens, Pollmann, Taubinsky (2019). NBER WP 26463."
+            "Athey, Chetty, Imbens & Kang (2019). NBER WP 26463."
         ),
     ))
 
@@ -2799,7 +2799,7 @@ def _build_registry():
         name="causal_bandit",
         category="causal_rl",
         description=(
-            "Bareinboim-Pearl contextual causal bandit: pick the optimal "
+            "Bareinboim-Forney-Pearl contextual causal bandit: pick the optimal "
             "arm by Monte-Carlo estimation of E[Y(a) | context]."
         ),
         params=[
@@ -2810,7 +2810,7 @@ def _build_registry():
         ],
         returns="CausalBanditResult",
         tags=["causal_rl", "bandit", "pearl"],
-        reference="Bareinboim & Pearl (NIPS 2015).",
+        reference="Bareinboim, Forney & Pearl (NeurIPS 2015). 'Bandits with Unobserved Confounders: A Causal Approach.'",
     ))
     register(FunctionSpec(
         name="counterfactual_policy_optimization",
@@ -3017,7 +3017,7 @@ def _build_registry():
         name="bayes_dml",
         category="bayes",
         description=(
-            "Bayesian Double Machine Learning (Chernozhukov et al. 2025): "
+            "Bayesian Double Machine Learning (DiTraglia & Liu 2025): "
             "Normal-Normal conjugate update on a DML point estimate, with "
             "optional full PyMC MCMC over the orthogonal moment equation."
         ),
@@ -3039,7 +3039,7 @@ def _build_registry():
             "covariates=['x1','x2'])"
         ),
         tags=["bayes", "dml", "double_ml", "posterior"],
-        reference="Chernozhukov et al. (arXiv:2508.12688, 2025).",
+        reference="DiTraglia & Liu (arXiv:2508.12688, 2025). DML framework: Chernozhukov et al. (2018).",
         pre_conditions=[
             "prior_sd is weakly informative relative to the expected effect scale",
             "for mode='full': pymc installed (sp.bayes extra)",
@@ -3443,7 +3443,7 @@ def _build_registry():
             "unit='id', time='t', covariates=['x1','x2'])"
         ),
         tags=["bcf", "longitudinal", "panel", "hte"],
-        reference="Alessi, Zorzetto et al. (arXiv:2508.08418, 2025).",
+        reference="Prevot, Häring, Nichols, Holmes & Ganjgahi (arXiv:2508.08418, 2025).",
     ))
 
     # -- Time-series causal discovery extensions --------------------- #
@@ -3632,7 +3632,7 @@ def _build_registry():
         description=(
             "Proximal surrogate-index estimator: long-term ATE when an "
             "unobserved U confounds S→Y, using a proxy W and 2SLS-style "
-            "bridge-function identification (Imbens-Kallus-Mao 2026)."
+            "bridge-function identification (Imbens-Kallus-Mao-Wang 2025, JRSS-B)."
         ),
         params=[
             ParamSpec("experimental", "DataFrame", True),
@@ -3650,7 +3650,7 @@ def _build_registry():
             "surrogates=['s'], proxies=['w'], long_term_outcome='Y')"
         ),
         tags=["surrogate", "long_term", "proximal", "unobserved_confounding"],
-        reference="Imbens, Kallus, Mao (arXiv:2601.17712, 2026).",
+        reference="Imbens, Kallus, Mao & Wang (2025). JRSS-B 87(2), 362-388. arXiv:2202.07234.",
     ))
 
     # ------------------------------------------------------------------
@@ -3924,7 +3924,7 @@ def _build_registry():
             "Unified dispatcher for six causal-inference bridging theorems "
             "(2025-2026): DiD≡SC (Shi-Athey), EWM≡CATE (Ferman), "
             "IPW≡DR≡CB (Zhao-Percival), Bunching≡RDD (Lu-Wang-Xie), "
-            "DR-via-Calibration (Zhang), Long-term-surrogate≡PCI (Kallus-Mao). "
+            "DR-via-Calibration (Zhang), Long-term-surrogate≡PCI (Imbens-Kallus-Mao-Wang). "
             "Reports both path estimates + doubly-robust recommendation."
         ),
         params=[
@@ -3940,7 +3940,7 @@ def _build_registry():
         reference=(
             "Shi-Athey (2503.11375); Ferman et al. (2510.26723); "
             "Zhao-Percival (2310.18563); Lu-Wang-Xie (2404.09117); "
-            "Zhang et al. (2411.02771); Kallus-Mao (2601.17712)."
+            "Zhang et al. (2411.02771); Imbens-Kallus-Mao-Wang (2202.07234, JRSS-B 2025)."
         ),
     ))
 

src/statspai/surrogate/__init__.py

@@ -9,24 +9,25 @@
 
 Estimators
 ----------
-- :func:`surrogate_index` — Athey, Chetty, Imbens, Pollmann & Taubinsky (2019).
+- :func:`surrogate_index` — Athey, Chetty, Imbens & Kang (2019).
   Classical single-wave surrogate index.
 - :func:`long_term_from_short` — Ghassami, Yang, Shpitser, Tchetgen Tchetgen
   (arXiv:2311.08527, 2024). Long-term effect of long-term treatments from
   short-term experiments.
-- :func:`proximal_surrogate_index` — Imbens, Kallus, Mao (arXiv:2601.17712,
-  2026). Proximal identification when unobserved confounders link surrogate
-  and long-term outcome.
+- :func:`proximal_surrogate_index` — Imbens, Kallus, Mao & Wang (2025, JRSS-B
+  87(2); arXiv:2202.07234). Proximal identification when unobserved
+  confounders link surrogate and long-term outcome.
 
 References
 ----------
-Athey, S., Chetty, R., Imbens, G., Pollmann, M., & Taubinsky, D. (2019).
+Athey, S., Chetty, R., Imbens, G. W., & Kang, H. (2019).
 "The Surrogate Index: Combining Short-Term Proxies to Estimate Long-Term
-Treatment Effects More Rapidly and Precisely." NBER WP 26463.
+Treatment Effects More Rapidly and Precisely." NBER Working Paper 26463.
 
-Imbens, G., Kallus, N., Mao, X. (2026).
-"The Proximal Surrogate Index: Long-Term Treatment Effects under
-Unobserved Confounding." arXiv:2601.17712.
+Imbens, G., Kallus, N., Mao, X., & Wang, Y. (2025).
+"Long-term Causal Inference Under Persistent Confounding via Data
+Combination." Journal of the Royal Statistical Society Series B,
+87(2), 362-388. arXiv:2202.07234.
 """
 
 from .index import (