StandardGraph — K-12 Curriculum Standards Graph

A unified, embedding-indexed SQLite database of 146,525 K-12 curriculum standards across 256 systems and 5 subjects, with semantic crosswalk mappings linking every standard to its corresponding hub framework.

Built for AI tutoring systems, curriculum alignment tools, and LLM context retrieval. Ships as a Model Context Protocol (MCP) server for use directly in Claude, Cursor, and other MCP-compatible tools.

---

What's inside

Subject	Hub	Systems	Standards
Mathematics	CCSS (343)	82	22,543
ELA / Literacy	CCSS ELA (504)	52	46,062
Science	NGSS (208)	59	34,011
Social Studies	C3 Framework (232)	51	42,711
Computer Science	CSTA (92)	12	1,198
Total	5 hubs	256	146,525

Additional tables:

• embeddings — 768-dim nomic-embed-text vectors for all 146,525 standards
• crosswalk_mappings — 97,949 semantic mappings linking every non-hub standard to its nearest hub equivalent (cosine ≥ 0.70)

---

Coverage

Mathematics

All 50 US states + DC, plus:

• International: Singapore MOE, Japan MEXT, Hong Kong EDB, New Zealand, Australia (ACARA + Victoria), India NCERT (grades 1-12), Quebec MEES, Scotland CfE, Ireland NCCA, Ghana NaCCA, South Africa CAPS, Rwanda REB
• Other US: AERO (DoD Dependents Schools), DoDEA, Cambridge IGCSE, IB MYP/DP, AP Calculus AB/BC, AP Statistics, AP Precalculus, UK AQA / National Curriculum
• Hub: CCSS (343 standards, exact count)

Science

All 50 US states + DC, AP Biology, AP Chemistry, AP Environmental Science, AP Physics 1/2/C

• Hub: NGSS (208 standards)

ELA / Literacy

All 50 US states + DC

• Hub: CCSS ELA (504 standards)

Social Studies

All 50 US states + DC, including California, Illinois, and Massachusetts full H-SS frameworks

• Hub: C3 Framework (232 indicators)

Computer Science

CSTA K-12 hub + 11 US state standards (CT, IA, ID, KY, MA, MD, NH, UT, WI, WV, and more)

• Hub: CSTA 2017 (92 standards)

---

Database schema

-- Core standards table
CREATE TABLE standards (
    id                TEXT PRIMARY KEY,   -- e.g. "CCSS.MATH.6.RP.A.3", "sg-moe.P5.NS.1"
    system            TEXT NOT NULL,      -- e.g. "ccss", "sg-moe", "ngss"
    subject           TEXT NOT NULL,      -- "mathematics" | "science" | "ela" | "social-studies" | "cs"
    grade             TEXT NOT NULL,      -- "6", "K", "HS", "9-12"
    grade_band        TEXT,              -- "6-8" where applicable
    domain            TEXT NOT NULL,      -- strand or domain label
    cluster           TEXT,              -- CCSS-specific grouping; NULL for other systems
    standard_text     TEXT NOT NULL,      -- verbatim official text
    last_verified_date TEXT NOT NULL,
    source_url        TEXT NOT NULL,
    created_at        TEXT NOT NULL,
    updated_at        TEXT NOT NULL
);

-- 768-dim nomic-embed-text vectors (float32, little-endian BLOBs)
CREATE TABLE embeddings (
    standard_id  TEXT PRIMARY KEY REFERENCES standards(id),
    model        TEXT NOT NULL,          -- "nomic-embed-text"
    vector       BLOB NOT NULL,          -- float32 little-endian
    dimensions   INTEGER NOT NULL        -- 768
);

-- Semantic crosswalk mappings (non-hub → hub)
CREATE TABLE crosswalk_mappings (
    id               INTEGER PRIMARY KEY,
    source_id        TEXT NOT NULL REFERENCES standards(id),
    source_system    TEXT NOT NULL,
    target_id        TEXT NOT NULL REFERENCES standards(id),
    target_system    TEXT NOT NULL,
    confidence_score REAL NOT NULL,      -- cosine similarity [0.70, 1.00]
    method           TEXT NOT NULL       -- "nlp-cosine"
);

---

Quick start

import sqlite3
import numpy as np

conn = sqlite3.connect("common_core.db")

# Look up a standard by ID
row = conn.execute(
    "SELECT grade, domain, standard_text FROM standards WHERE id=?",
    ("CCSS.MATH.6.RP.A.3",)
).fetchone()
print(row)
# ('6', 'Ratios and Proportional Relationships',
#  'Use ratio and rate reasoning to solve real-world and mathematical problems...')

# Find the CCSS equivalent of a Singapore standard
sg_id = "sg-moe.P5.NS.1"
crosswalk = conn.execute(
    "SELECT target_id, confidence_score FROM crosswalk_mappings WHERE source_id=?",
    (sg_id,)
).fetchone()
# ('CCSS.MATH.5.NBT.A.1', 0.87)

# List all systems
systems = conn.execute(
    "SELECT system, subject, COUNT(*) FROM standards GROUP BY system, subject ORDER BY subject, system"
).fetchall()

Semantic search

Embed queries with nomic-embed-text via Ollama, then compare against stored vectors:

import httpx, numpy as np, sqlite3

def embed(text: str) -> np.ndarray:
    resp = httpx.post("http://localhost:11434/api/embed",
                      json={"model": "nomic-embed-text", "input": [text]})
    return np.array(resp.json()["embeddings"][0], dtype=np.float32)

def search(query: str, system: str, k: int = 5, conn=conn):
    rows = conn.execute(
        "SELECT s.id, e.vector FROM standards s JOIN embeddings e ON e.standard_id=s.id WHERE s.system=?",
        (system,)
    ).fetchall()
    ids  = [r[0] for r in rows]
    vecs = np.array([np.frombuffer(r[1], dtype=np.float32) for r in rows])
    vecs /= np.linalg.norm(vecs, axis=1, keepdims=True) + 1e-9
    q    = embed(query)
    q   /= np.linalg.norm(q) + 1e-9
    sims = vecs @ q
    top  = np.argsort(sims)[::-1][:k]
    return [(ids[i], float(sims[i])) for i in top]

search("adding fractions with unlike denominators", "ccss")
# [('CCSS.MATH.5.NF.A.2', 0.91), ('CCSS.MATH.5.NF.A.1', 0.88), ...]

---

MCP server

StandardGraph ships as a Model Context Protocol server for direct use inside Claude, Cursor, and other MCP-compatible AI tools.

git clone https://github.com/swoopeagle/standardgraph
cd standardgraph
uv run python -m common_core.server

Tools: search_standards, get_standard, find_crosswalk, list_systems, get_system_overview.

See the GitHub repo for the one-line installer for Claude Desktop.

---

Evaluation

Run the automated eval suite against the DB:

git clone https://github.com/swoopeagle/standardgraph
cd standardgraph
uv run python scripts/eval/run_all.py

Current results (June 2026):

Check	Result
Coverage (15 required IDs + hub counts)	✅ Pass
Crosswalk quality (confidence distribution + routing)	✅ Pass
Search quality — Recall@5 on 15 golden queries	✅ 93% (threshold: 70%)
Duplicate detection	✅ Pass (WARNs are intentional cross-grade text sharing in source standards)

---

Sources

System	Source
CCSS Math & ELA, NGSS, CSTA, C3	Common Standards Project API (achieve.org)
US state standards (50 states + DC, 5 subjects)	Common Standards Project API
Singapore MOE	Ministry of Education Singapore (official syllabi PDFs)
Japan MEXT	Ministry of Education, Culture, Sports, Science and Technology
Hong Kong EDB	Education Bureau Hong Kong
Australia ACARA	Australian Curriculum, Assessment and Reporting Authority
New Zealand	Ministry of Education New Zealand
Scotland CfE	Education Scotland (Curriculum for Excellence)
Ireland NCCA	National Council for Curriculum and Assessment
India NCERT	National Council of Educational Research and Training (grades 1-12)
Quebec MEES	Ministère de l'Éducation et de l'Enseignement supérieur
Ghana NaCCA	National Council for Curriculum and Assessment
South Africa CAPS	Department of Basic Education
Rwanda REB	Rwanda Education Board
AP Courses	College Board (Course and Exam Descriptions, PDFs)
Cambridge IGCSE	Cambridge Assessment International Education
IB MYP/DP	International Baccalaureate Organization
New Hampshire CS	NH Department of Education CS Standards (2018)
Wisconsin CS	Wisconsin DPI CS Standards (December 2025)

International and AP standards were extracted from official PDF documents using Gemma 4 31B (via Ollama).

---

License

Embeddings and crosswalk mappings: original work, released under CC BY 4.0.

Standards text: reproduced from official public government and organizational documents. Each standard's source_url field links to the authoritative source. This compilation is released under CC BY 4.0; users should verify compliance with the originating body's terms for their use case.