standardgraph 1.1.2__tar.gz → 1.2.2__tar.gz

{standardgraph-1.1.2 → standardgraph-1.2.2}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: standardgraph
-Version: 1.1.2
-Summary: 156,000+ education standards (math, science, ELA, social studies, CS, arts) across 298 curriculum systems, cross-referenced via NLP — accessible as a Claude MCP server
+Version: 1.2.2
+Summary: 158,000+ education standards (math, science, ELA, social studies, CS, arts) across 300 curriculum systems, cross-referenced via NLP — accessible as a Claude MCP server
 Author: StandardGraph
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/swoopeagle/standardgraph
@@ -22,9 +22,9 @@ Requires-Dist: numpy>=1.26.0
 
 # standardgraph
 
-**20,000+ math standards across 75+ curriculum systems, accessible as a Claude MCP server.**
+**157,000+ education standards across 298 curriculum systems, accessible as a Claude MCP server.**
 
-Covers the US (CCSS + all 50 states), Canada, Australia, UK, Singapore, Japan, New Zealand, Ireland, Hong Kong, India, Ghana, South Africa, Rwanda, Cambridge International, IB MYP/DP, and more — all cross-referenced to CCSS via NLP semantic similarity.
+Covers seven subjects — Mathematics, Science, ELA, Social Studies, Computer Science, Arts, and World Languages — across the US (CCSS + all 50 states), Canada, Australia, UK, Singapore, Japan, New Zealand, Ireland, Hong Kong, India, Ghana, South Africa, Rwanda, Cambridge International, IB MYP/DP, AP, and more. Standards are cross-referenced to subject hubs (CCSS for math, NGSS for science, etc.) via semantic similarity, with LLM quality scores on the strongest mappings.
 
 ## Install (macOS)
 

{standardgraph-1.1.2 → standardgraph-1.2.2}/README.md

@@ -1,8 +1,8 @@
 # standardgraph
 
-**20,000+ math standards across 75+ curriculum systems, accessible as a Claude MCP server.**
+**157,000+ education standards across 298 curriculum systems, accessible as a Claude MCP server.**
 
-Covers the US (CCSS + all 50 states), Canada, Australia, UK, Singapore, Japan, New Zealand, Ireland, Hong Kong, India, Ghana, South Africa, Rwanda, Cambridge International, IB MYP/DP, and more — all cross-referenced to CCSS via NLP semantic similarity.
+Covers seven subjects — Mathematics, Science, ELA, Social Studies, Computer Science, Arts, and World Languages — across the US (CCSS + all 50 states), Canada, Australia, UK, Singapore, Japan, New Zealand, Ireland, Hong Kong, India, Ghana, South Africa, Rwanda, Cambridge International, IB MYP/DP, AP, and more. Standards are cross-referenced to subject hubs (CCSS for math, NGSS for science, etc.) via semantic similarity, with LLM quality scores on the strongest mappings.
 
 ## Install (macOS)
 

{standardgraph-1.1.2 → standardgraph-1.2.2}/pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "standardgraph"
-version = "1.1.2"
-description = "156,000+ education standards (math, science, ELA, social studies, CS, arts) across 298 curriculum systems, cross-referenced via NLP — accessible as a Claude MCP server"
+version = "1.2.2"
+description = "158,000+ education standards (math, science, ELA, social studies, CS, arts) across 300 curriculum systems, cross-referenced via NLP — accessible as a Claude MCP server"
 readme = "README.md"
 requires-python = ">=3.11"
 license = "MIT"

{standardgraph-1.1.2 → standardgraph-1.2.2}/src/common_core/server.py

@@ -149,6 +149,19 @@ _AP_META       = {"country": "United States", "country_code": "US", "region": "N
 _US_STATE_SUFFIXES = {"-sci", "-ela", "-ss", "-cs"}
 
 
+# Quality scores come from two sources: LLM rubric scoring ("[LLM score N/5]") and
+# deterministic exact-match scoring ("[exact-match N/5]", assigned when source and
+# target standard text are byte-identical). Both are surfaced as quality_score.
+_SCORE_RE = re.compile(r"\[(?:LLM score|exact-match) (\d)/5\]")
+
+
+def _parse_quality(notes: str | None) -> int | None:
+    if not notes:
+        return None
+    m = _SCORE_RE.search(notes)
+    return int(m.group(1)) if m else None
+
+
 def _meta(system: str) -> dict:
     if system in SYSTEM_META:
         return SYSTEM_META[system]
@@ -215,8 +228,19 @@ def _embed_query(text: str) -> np.ndarray:
     return np.array(resp.json()["embeddings"][0], dtype=np.float32)
 
 
-def _cosine_scores(query_vec: np.ndarray, conn: sqlite3.Connection) -> list[tuple[float, str]]:
-    rows = conn.execute("SELECT standard_id, vector, dimensions FROM embeddings").fetchall()
+def _cosine_scores(
+    query_vec: np.ndarray,
+    conn: sqlite3.Connection,
+    system: str | None = None,
+) -> list[tuple[float, str]]:
+    if system:
+        rows = conn.execute(
+            "SELECT e.standard_id, e.vector, e.dimensions FROM embeddings e "
+            "JOIN standards s ON s.id = e.standard_id WHERE s.system = ?",
+            (system,),
+        ).fetchall()
+    else:
+        rows = conn.execute("SELECT standard_id, vector, dimensions FROM embeddings").fetchall()
     if not rows:
         return []
     dim = rows[0]["dimensions"]
@@ -440,14 +464,14 @@ def search_standards(
             "results": results,
         }, indent=2)
     conn = _db()
-    scored = _cosine_scores(query_vec, conn)
+    scored = _cosine_scores(query_vec, conn, system=system)
 
     results = []
     for score, sid in scored:
         if len(results) >= limit:
             break
         row = conn.execute(
-            "SELECT * FROM standards WHERE id=? AND system=?", (sid, system)
+            "SELECT * FROM standards WHERE id=?", (sid,)
         ).fetchone()
         if not row:
             continue
@@ -541,7 +565,7 @@ def get_progression(
             ],
         }, indent=2)
     conn = _db()
-    scored = _cosine_scores(query_vec, conn)
+    scored = _cosine_scores(query_vec, conn, system=system)
 
     # Collect top standards per grade, filtered by grade range
     by_grade: dict[str, list[dict]] = {}
@@ -549,7 +573,7 @@ def get_progression(
         if score < 0.5:
             break
         row = conn.execute(
-            "SELECT * FROM standards WHERE id=? AND system=?", (sid, system)
+            "SELECT * FROM standards WHERE id=?", (sid,)
         ).fetchone()
         if not row:
             continue
@@ -596,6 +620,7 @@ def map_standard(
     from_system: str,
     to_system: str,
     confidence_threshold: float = 0.7,
+    include_flagged: bool = False,
 ) -> str:
     """Find the closest equivalent to a standard in a different curriculum system.
 
@@ -613,8 +638,11 @@ def map_standard(
     from_system: system code of the source standard (e.g. 'tx', 'ca-on', 'sg-moe').
     to_system: target system code (any indexed system).
     confidence_threshold: minimum cosine similarity for primary results (default 0.7).
+    include_flagged: if True, include mappings flagged for review (LLM quality score 1-2).
+      Default False returns only verified-quality mappings.
 
-    Returns matched standards with confidence score, grade alignment, and mapping method.
+    Returns matched standards with confidence score, grade alignment, quality_score (1-5),
+    flagged status, and mapping method.
     """
     sid = _expand_id(standard_id, from_system)
     conn = _db()
@@ -627,40 +655,53 @@ def map_standard(
     src_dict = dict(src)
 
     # ── 1. Precomputed crosswalk above threshold ───────────────────────────────
+    flagged_clause = "" if include_flagged else "AND cm.flagged_for_review = 0"
     mappings = conn.execute(
-        """SELECT cm.*, s.standard_text AS target_text, s.grade AS target_grade,
+        f"""SELECT cm.*, s.standard_text AS target_text, s.grade AS target_grade,
                   s.domain AS target_domain
            FROM crosswalk_mappings cm
            JOIN standards s ON s.id = cm.target_id
            WHERE cm.source_id = ?
              AND cm.target_system = ?
              AND cm.confidence_score >= ?
+             {flagged_clause}
            ORDER BY cm.confidence_score DESC""",
         (sid, to_system, confidence_threshold),
     ).fetchall()
 
     if mappings:
         conn.close()
+        result_list = [
+            {
+                "target_id":            m["target_id"],
+                "target_standard_text": m["target_text"],
+                "relationship":         m["relationship"],
+                "confidence":           m["confidence_score"],
+                "quality_score":        _parse_quality(m["notes"]),
+                "flagged":              bool(m["flagged_for_review"]),
+                "grade_delta":          m["grade_delta"],
+                "grade_alignment":      "exact" if m["grade_delta"] == 0 else (
+                    f"{abs(m['grade_delta'])} year{'s' if abs(m['grade_delta']) > 1 else ''} "
+                    f"{'later' if m['grade_delta'] > 0 else 'earlier'} in target"
+                ),
+                "verified_by_human":    bool(m["verified_by_human"]),
+                "notes":                m["notes"],
+            }
+            for m in mappings
+        ]
+        # Rank by quality then confidence. Unscored rows (quality_score is None) are
+        # of *unknown* quality, not zero quality — treat them as a neutral midpoint so
+        # a high-cosine unscored match isn't buried beneath a mediocre scored one.
+        # (Score-1/2 rows are already filtered out by default via flagged_clause.)
+        result_list.sort(
+            key=lambda x: (x["quality_score"] if x["quality_score"] is not None else 3, x["confidence"]),
+            reverse=True,
+        )
         return json.dumps({
             "source_id":         src_dict["id"],
             "target_curriculum": to_system,
             "mapping_method":    "precomputed_crosswalk",
-            "mappings": [
-                {
-                    "target_id":            m["target_id"],
-                    "target_standard_text": m["target_text"],
-                    "relationship":         m["relationship"],
-                    "confidence":           m["confidence_score"],
-                    "grade_delta":          m["grade_delta"],
-                    "grade_alignment":      "exact" if m["grade_delta"] == 0 else (
-                        f"{abs(m['grade_delta'])} year{'s' if abs(m['grade_delta']) > 1 else ''} "
-                        f"{'later' if m['grade_delta'] > 0 else 'earlier'} in target"
-                    ),
-                    "verified_by_human":    bool(m["verified_by_human"]),
-                    "notes":                m["notes"],
-                }
-                for m in mappings
-            ],
+            "mappings":          result_list,
         }, indent=2)
 
     # ── 2. Best precomputed result below threshold ─────────────────────────────
@@ -748,14 +789,14 @@ def map_standard(
     embedding_error: str | None = None
     try:
         qvec = _embed_query(src_dict["standard_text"])
-        scored = _cosine_scores(qvec, conn)
+        scored = _cosine_scores(qvec, conn, system=to_system)
         for score, candidate_id in scored:
             if len(nearest_by_concept) >= 3:
                 break
             if score < 0.35:
                 break
             row = conn.execute(
-                "SELECT * FROM standards WHERE id=? AND system=?", (candidate_id, to_system)
+                "SELECT * FROM standards WHERE id=?", (candidate_id,)
             ).fetchone()
             if row:
                 nearest_by_concept.append({
@@ -782,6 +823,8 @@ def map_standard(
             "target_id":            best_below["target_id"],
             "target_standard_text": best_below["target_text"],
             "confidence":           round(best_below["confidence_score"], 4),
+            "quality_score":        _parse_quality(best_below["notes"]),
+            "flagged":              bool(best_below["flagged_for_review"]),
             "below_threshold":      True,
             "threshold_used":       confidence_threshold,
         }

{standardgraph-1.1.2 → standardgraph-1.2.2}/src/standardgraph.egg-info/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: standardgraph
-Version: 1.1.2
-Summary: 156,000+ education standards (math, science, ELA, social studies, CS, arts) across 298 curriculum systems, cross-referenced via NLP — accessible as a Claude MCP server
+Version: 1.2.2
+Summary: 158,000+ education standards (math, science, ELA, social studies, CS, arts) across 300 curriculum systems, cross-referenced via NLP — accessible as a Claude MCP server
 Author: StandardGraph
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/swoopeagle/standardgraph
@@ -22,9 +22,9 @@ Requires-Dist: numpy>=1.26.0
 
 # standardgraph
 
-**20,000+ math standards across 75+ curriculum systems, accessible as a Claude MCP server.**
+**157,000+ education standards across 298 curriculum systems, accessible as a Claude MCP server.**
 
-Covers the US (CCSS + all 50 states), Canada, Australia, UK, Singapore, Japan, New Zealand, Ireland, Hong Kong, India, Ghana, South Africa, Rwanda, Cambridge International, IB MYP/DP, and more — all cross-referenced to CCSS via NLP semantic similarity.
+Covers seven subjects — Mathematics, Science, ELA, Social Studies, Computer Science, Arts, and World Languages — across the US (CCSS + all 50 states), Canada, Australia, UK, Singapore, Japan, New Zealand, Ireland, Hong Kong, India, Ghana, South Africa, Rwanda, Cambridge International, IB MYP/DP, AP, and more. Standards are cross-referenced to subject hubs (CCSS for math, NGSS for science, etc.) via semantic similarity, with LLM quality scores on the strongest mappings.
 
 ## Install (macOS)