release: v1.5.1 — PyPI readme hotfix + unshipped-import cleanup

Patch release. No code / behaviour changes vs. v1.5.0.

- Fix PyPI project-page broken image: README's interactive-plot-editor
  screenshot was referenced by relative path (./image-1.png), which
  GitHub resolves but PyPI cannot, so the PyPI v1.5.0 page showed a
  broken-image icon. Both README.md and README_CN.md now use the
  absolute raw.githubusercontent.com URL so both renderers work.

- Remove premature v1.6 imports from src/statspai/__init__.py that
  were introduced by commit 8155ada but whose implementations had not
  yet landed: llm_dag_constrained / llm_dag_validate /
  LLMConstrainedDAGResult / DAGValidationResult, paper / PaperDraft,
  and mr_lap / mr_clust / grapple / mr_cml. These broke `import
  statspai` on any fresh clone of main between v1.5.0 and v1.5.1; the
  shipped PyPI v1.5.0 wheel was unaffected. They will return in v1.6
  alongside their implementations.

- README visual-weight fix: v1.5.0 section now carries a "v1.5
  Highlights" table paralleling the v1.4 Highlights table so the
  current release is the README's visual subject.

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

Author: brycewang-stanford

CHANGELOG.md

@@ -2,6 +2,60 @@
 
 All notable changes to StatsPAI will be documented in this file.
 
+## [1.5.1] — 2026-04-21 — PyPI readme hotfix + unshipped-import cleanup
+
+Patch release.  No new features, no behaviour changes to any shipped
+estimator, no schema changes.  Every public signature present in v1.5.0
+is byte-for-byte identical in v1.5.1.
+
+### Fixed
+
+- **PyPI project-page broken image.** The README's `interactive plot
+  editor` screenshot was referenced via a relative path
+  (`./image-1.png`), which GitHub resolves correctly but PyPI cannot.
+  The PyPI v1.5.0 page therefore showed a broken-image icon.  Both
+  `README.md` and `README_CN.md` now use the absolute
+  `https://raw.githubusercontent.com/brycewang-stanford/StatsPAI/main/image-1.png`
+  URL so both renderers work.
+- **`import statspai` was broken on `main` between v1.5.0 and this
+  release.** Commit `8155ada` ("feat(agent): unified exception
+  taxonomy + per-function agent cards") pre-wired four in-progress
+  v1.6 symbols into the top-level package —
+  `llm_dag_constrained`, `llm_dag_validate`,
+  `LLMConstrainedDAGResult`, `DAGValidationResult` (closed-loop LLM-DAG),
+  `paper`, `PaperDraft` (data → publication-draft pipeline), and the
+  four MR-frontier estimators `mr_lap`, `mr_clust`, `grapple`,
+  `mr_cml` — before their implementations landed in the corresponding
+  sub-packages.  Any fresh `git clone` + `pip install -e .` would
+  raise `ImportError: cannot import name 'llm_dag_constrained' from
+  'statspai.causal_llm'`.  These premature imports are removed in
+  v1.5.1; they will be re-introduced in v1.6 alongside their
+  implementations.  **This bug did not affect the PyPI v1.5.0 wheel,
+  which pre-dates `8155ada` and has a clean import graph.**
+
+### Changed — README visual ordering
+
+- **v1.5 Highlights table.** v1.5.0's "NEW" section was a short
+  paragraph plus bullets, which was visually outweighed by the full
+  `v1.4 Highlights` table immediately below it.  v1.5.0's bullets are
+  now promoted into a parallel `v1.5 Highlights` table (family guides,
+  unified dispatchers, two ⚠️ correctness fixes, one ⚠️ breaking
+  change, registry coverage) so the current release is the README's
+  visual subject instead of being crowded out by the previous-version
+  section.  `README_CN.md` updated in lockstep.
+
+### Upgrade notes
+
+If you installed v1.5.0 from PyPI: there is no functional reason to
+upgrade to v1.5.1.  The PyPI v1.5.0 wheel is not affected by the
+`import statspai` bug described above, which only existed on the
+`main` branch between v1.5.0 and v1.5.1.
+
+If you installed from source off `main`: upgrading (or pulling v1.5.1)
+fixes the `ImportError`.
+
+---
+
 ## [1.5.0] — 2026-04-21 — Interference / Conformal / Mendelian family consolidation
 
 Minor release.  Three concurrent improvements to the interference,

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "StatsPAI"
-version = "1.5.0"
+version = "1.5.1"
 description = "The Agent-Native Causal Inference & Econometrics Toolkit for Python"
 readme = "README.md"
 license = {text = "MIT"}

src/statspai/__init__.py

@@ -22,7 +22,7 @@
 >>> sp.outreg2(result, filename="results.xlsx")
 """
 
-__version__ = "1.5.0"
+__version__ = "1.5.1"
 __author__ = "Biaoyue Wang"
 __email__ = "brycew6m@stanford.edu"
 
@@ -281,9 +281,6 @@
     llm_unobserved_confounders, UnobservedConfounderProposal,
     llm_sensitivity_priors, SensitivityPriorProposal,
     causal_mas, CausalMASResult,
-    # P1-A: closed-loop LLM-DAG (constrained discovery + CI validation)
-    llm_dag_constrained, llm_dag_validate,
-    LLMConstrainedDAGResult, DAGValidationResult,
 )
 
 # === Causal RL (Causal-DQN, benchmarks, offline-safe) ===
@@ -396,7 +393,7 @@
 from . import datasets
 
 # === End-to-end workflow orchestrator ===
-from .workflow import causal, CausalWorkflow, paper, PaperDraft
+from .workflow import causal, CausalWorkflow
 
 # === LLM agent tool-definition surface ===
 from . import agent
@@ -500,9 +497,6 @@
     ModeBasedResult, FStatisticResult,
     mr_multivariable, mr_mediation, mr_bma,
     MVMRResult, MediationMRResult, MRBMAResult,
-    # v1.6 frontier: MR-Lap / MR-Clust / GRAPPLE / MR-cML
-    mr_lap, mr_clust, grapple, mr_cml,
-    MRLapResult, MRClustResult, GrappleResult, MRcMLResult,
     # v1.5 unified dispatcher (replaces the `sp.mr` module alias)
     mr, mr_available_methods,
 )
@@ -1326,8 +1320,6 @@
     "llm_dag_propose", "LLMDAGProposal",
     "llm_unobserved_confounders", "UnobservedConfounderProposal",
     "llm_sensitivity_priors", "SensitivityPriorProposal",
-    "llm_dag_constrained", "llm_dag_validate",
-    "LLMConstrainedDAGResult", "DAGValidationResult",
     # Causal RL
     "causal_dqn", "CausalDQNResult",
     "causal_rl_benchmark", "BanditBenchmarkResult",
@@ -1343,9 +1335,6 @@
     # Bunching frontier
     "general_bunching", "GeneralBunchingResult",
     "kink_unified", "KinkUnifiedResult",
-    # v1.6 MR frontier: sample-overlap / clusters / profile-LL / cML
-    "mr_lap", "mr_clust", "grapple", "mr_cml",
-    "MRLapResult", "MRClustResult", "GrappleResult", "MRcMLResult",
     # v1.5 unified family dispatchers
     "mr", "mr_available_methods",
     "conformal", "conformal_available_kinds",