eduevidence 5.2.0 → 6.2.0

package/scripts/did_regression.py

@@ -19,6 +19,9 @@ import json
 import math
 import sys
 from pathlib import Path
+from typing import Any, Dict, List
+
+
 def _two_tailed_p_from_z(z: float) -> float:
     """Standard normal two-tailed p-value."""
     return 2.0 * (1.0 - 0.5 * (1.0 + math.erf(abs(z) / math.sqrt(2.0))))
@@ -39,51 +42,40 @@ def _null_inference() -> Dict[str, Any]:
 def _solve_linear_system(A: List[List[float]], b: List[float]) -> List[float]:
     """Gaussian elimination with partial pivoting for small OLS systems (p <= 10)."""
     n = len(b)
-    # Augmented matrix
     M = [A[i][:] + [b[i]] for i in range(n)]
-
     for i in range(n):
-        # Pivot
         max_row = max(range(i, n), key=lambda r: abs(M[r][i]))
         if abs(M[max_row][i]) < 1e-12:
             raise ValueError("Singular matrix in OLS estimation")
         M[i], M[max_row] = M[max_row], M[i]
-
         pivot = M[i][i]
         for j in range(i, n + 1):
             M[i][j] /= pivot
-
         for r in range(n):
             if r != i:
                 factor = M[r][i]
                 for c in range(i, n + 1):
                     M[r][c] -= factor * M[i][c]
-
     return [M[i][n] for i in range(n)]
 
 
 def _matrix_inverse(A: List[List[float]]) -> List[List[float]]:
     """Inverts an n x n matrix using Gauss-Jordan elimination."""
     n = len(A)
-    # Augment with identity
     M = [A[i][:] + [1.0 if i == j else 0.0 for j in range(n)] for i in range(n)]
-
     for i in range(n):
         max_row = max(range(i, n), key=lambda r: abs(M[r][i]))
         if abs(M[max_row][i]) < 1e-12:
             raise ValueError("Singular matrix in inversion")
         M[i], M[max_row] = M[max_row], M[i]
-
         pivot = M[i][i]
         for j in range(2 * n):
             M[i][j] /= pivot
-
         for r in range(n):
             if r != i:
                 factor = M[r][i]
                 for c in range(2 * n):
                     M[r][c] -= factor * M[i][c]
-
     return [[M[i][n + j] for j in range(n)] for i in range(n)]
 
 
@@ -104,19 +96,28 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
         return {"status": "error", "error_code": "ERR_INSUFFICIENT_ROWS",
                 "message": "Insufficient data rows (minimum 4 required)"}
 
-    # Normalize column names (+ cluster candidates; exact names only, never guessed)
     field_map = {}
     cluster_columns: List[str] = []
     for col in rows[0].keys():
         cl = col.strip().lower()
         if cl in ("cluster_id", "class_id", "school_id", "group_id") or cl.endswith("_cluster"):
             cluster_columns.append(col)
+        # Order matters. An outcome column is often named "post_test_score",
+        # which also contains "post": matching the period rule first stole the
+        # outcome column and the run failed with ERR_MISSING_COLUMNS. Outcome
+        # and treatment are the more specific patterns, so they are tested
+        # before the period keyword.
         if "treat" in cl or cl in ("group", "condition", "is_treatment"):
             field_map["treat"] = col
-        elif "post" in cl or "after" in cl or "period" in cl or "time" in cl or "pre_post" in cl:
+        elif cl in ("post", "posttest", "pre_post", "period", "time_period"):
             field_map["post"] = col
-        elif "score" in cl or "outcome" in cl or "grade" in cl or "result" in cl or "performance" in cl or cl == "y":
+        elif ("score" in cl or "outcome" in cl or "grade" in cl or "result" in cl
+              or "performance" in cl or cl == "y"):
             field_map["outcome"] = col
+        elif "post" in cl or "after" in cl or "period" in cl or "time" in cl:
+            # Generic period/phase column, only after the specific patterns
+            # above have had their chance.
+            field_map.setdefault("post", col)
 
     if "treat" not in field_map or "post" not in field_map or "outcome" not in field_map:
         return {
@@ -124,13 +125,10 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
             "message": f"CSV missing required columns (need treat/post/outcome). Found: {list(rows[0].keys())}"
         }
 
-    # Parse numeric arrays
     y_vals: List[float] = []
     treat_vals: List[float] = []
     post_vals: List[float] = []
     treat_post_vals: List[float] = []
-
-    # Cells for 2x2 table
     cell_y = {(0, 0): [], (0, 1): [], (1, 0): [], (1, 1): []}
 
     for r in rows:
@@ -138,7 +136,6 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
             t = 1.0 if float(r[field_map["treat"]]) > 0.5 else 0.0
             p = 1.0 if float(r[field_map["post"]]) > 0.5 else 0.0
             y = float(r[field_map["outcome"]])
-            
             treat_vals.append(t)
             post_vals.append(p)
             treat_post_vals.append(t * p)
@@ -152,7 +149,6 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
         return {"status": "error", "error_code": "ERR_PARSE",
                 "message": "Failed to parse sufficient numeric rows"}
 
-    # Cell means
     means = {}
     stds = {}
     for k, v in cell_y.items():
@@ -165,7 +161,6 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
             means[k] = 0.0
             stds[k] = 1.0
 
-    # --- Estimability gates (fail closed; never fabricate inference) ---
     if len(set(treat_vals)) < 2:
         return {
             "status": "error", "error_code": "ERR_NO_TREAT_VARIATION",
@@ -205,15 +200,12 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
     y_t_pre = means[(1, 0)]
     y_t_post = means[(1, 1)]
 
-    # Simple 2x2 delta
-    delta_simple = (y_t_post - y_t_pre) - (y_c_post - y_c_pre)
+    # Keep the simple 2x2 contrast as a diagnostic equivalence check.
+    _delta_simple = (y_t_post - y_t_pre) - (y_c_post - y_c_pre)
 
-    # OLS Estimation: Y = X * beta + e, X = [1, Treat, Post, Treat*Post]
-    # Build X^T X (4x4) and X^T Y (4x1)
     X = [[1.0, treat_vals[i], post_vals[i], treat_post_vals[i]] for i in range(n)]
     XtX = [[0.0] * 4 for _ in range(4)]
     XtY = [0.0] * 4
-
     for i in range(n):
         row = X[i]
         yi = y_vals[i]
@@ -232,7 +224,6 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
             **_null_inference(),
         }
 
-    # Residual sum of squares & Standard Error
     rss = 0.0
     for i in range(n):
         y_hat = beta[0] + beta[1] * treat_vals[i] + beta[2] * post_vals[i] + beta[3] * treat_post_vals[i]
@@ -241,7 +232,6 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
     df_resid = n - 4
     sigma2 = rss / df_resid
     r_squared = max(0.0, 1.0 - (rss / tss)) if tss > 0 else 0.0
-
     if rss <= 0:
         return {
             "status": "error", "error_code": "ERR_ZERO_RESIDUAL",
@@ -252,16 +242,12 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
     t_stat = beta[3] / se_delta if se_delta > 0 else 0.0
     p_val = _two_tailed_p_from_z(t_stat)
 
-    # Standardized Effect Size: Hedges' g
     s_pooled_pre = math.sqrt((stds[(0, 0)] ** 2 + stds[(1, 0)] ** 2) / 2.0) if stds[(0, 0)] and stds[(1, 0)] else 1.0
     hedges_j = 1.0 - (3.0 / (4.0 * df_resid - 1.0)) if df_resid > 2 else 1.0
     hedges_g = round(hedges_j * (beta[3] / s_pooled_pre), 3) if s_pooled_pre > 0 else 0.0
 
-    # Baseline Equivalence
     baseline_diff = y_t_pre - y_c_pre
     baseline_g = baseline_diff / s_pooled_pre if s_pooled_pre > 0 else 0.0
-    # QED/DID can never meet WWC 5.0 standards WITHOUT reservations; and without
-    # covariate-adjustment fields a 0.05<|g|<=0.25 baseline is not passable.
     if abs(baseline_g) <= 0.05:
         wwc_rating = "Meets Standards With Reservations"
     elif abs(baseline_g) <= 0.25:
@@ -273,7 +259,6 @@ def run_did_analysis(csv_path: str) -> Dict[str, Any]:
         round(beta[3] - 1.96 * se_delta, 3),
         round(beta[3] + 1.96 * se_delta, 3)
     ]
-
     cluster_note = (
         f"cluster column(s) detected ({', '.join(cluster_columns)}) but cluster-robust "
         "inference is not implemented in this build; p-value is not cluster-robust"

package/scripts/enrich_projects_human_and_lieflat.py

@@ -7,7 +7,7 @@ import json
 from pathlib import Path
 
 ROOT = Path(__file__).resolve().parent.parent
-EXAMPLES_DIR = ROOT / "examples"
+EXAMPLES_DIR = ROOT / "tests" / "fixtures" / "legacy-examples"
 
 
 def enrich_math_project():

package/scripts/evidence_score.py

@@ -36,8 +36,11 @@ from evidence_semantics import claim_relation, decision_relation
 DIMENSIONS = ["D1_study_design", "D2_sample_quality", "D3_measurement_validity",
               "D4_temporal_strength", "D5_directness"]
 
-#: Version of the deterministic confidence policy (bump on any formula change).
-CONFIDENCE_POLICY_VERSION = "2026-08-12.v2"
+#: Version of the deterministic confidence policy. Imported from the engine so
+#: there is one authority: evidence_score previously hard-coded ".v2" while
+#: engine/versions.py declared ".v3", so verdicts recorded a policy version the
+#: engine did not recognise and nothing compared the two.
+from engine.versions import CONFIDENCE_POLICY_VERSION  # noqa: F401  (re-export)
 
 
 def quality_score(dimensions: dict[str, int]) -> float:

package/scripts/generate_metrics.py

@@ -10,10 +10,11 @@ Metrics:
 - engine_version          from engine/versions.py (the version authority)
 - test_functions          grep 'def test_' across tests/
 - test_files              number of collected test modules in tests/
-- schema_count            schemas/*.json at root + v2/ + v3/ + v4/
+- schema_count            schemas/*.json recursively
 - reference_doc_count     references/*.md
 - gold_annotation_count   benchmarks/annotations/gold-Q*.json
-- example_packs           examples/*/ directories shipping result.json
+- example_packs           real examples/*/ directories shipping result.json
+                          (compatibility symlink aliases are excluded)
 
 Usage:
     python3 scripts/generate_metrics.py            # regenerate docs/metrics.json
@@ -56,7 +57,7 @@ def collect() -> dict:
 
     example_packs = sorted(
         p.name for p in (REPO_ROOT / "examples").iterdir()
-        if p.is_dir() and (p / "result.json").exists()
+        if not p.is_symlink() and p.is_dir() and (p / "result.json").exists()
     ) if (REPO_ROOT / "examples").is_dir() else []
 
     return {

package/scripts/generate_new_projects.py

@@ -12,7 +12,7 @@ import json
 from pathlib import Path
 
 ROOT = Path(__file__).resolve().parent.parent
-EXAMPLES_DIR = ROOT / "examples"
+EXAMPLES_DIR = ROOT / "tests" / "fixtures" / "legacy-examples"
 
 
 def build_math_project():
@@ -194,7 +194,7 @@ def build_math_project():
         "execution": {
             "complexity": "L",
             "mode": "agent_mcp_enhanced",
-            "agents": ["education-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
+            "agents": ["research-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
         },
         "research_frame": {
             "question": question_en,
@@ -266,7 +266,7 @@ def build_math_project():
         "execution": {
             "complexity": "L",
             "mode": "agent_mcp_enhanced",
-            "agents": ["education-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
+            "agents": ["research-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
         },
         "research_frame": {
             "question": question_zh,
@@ -512,7 +512,7 @@ def build_writing_project():
         "execution": {
             "complexity": "L",
             "mode": "agent_mcp_enhanced",
-            "agents": ["education-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
+            "agents": ["research-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
         },
         "research_frame": {
             "question": question_en,
@@ -584,7 +584,7 @@ def build_writing_project():
         "execution": {
             "complexity": "L",
             "mode": "agent_mcp_enhanced",
-            "agents": ["education-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
+            "agents": ["research-planner", "evidence-retriever", "evidence-analyst", "skeptic", "method-reviewer", "evidence-judge", "intervention-designer", "evaluation-designer"]
         },
         "research_frame": {
             "question": question_zh,