ACADEMIC EXPERIMENT // BUILD 0.4.3

Open-Science Statistical Computing. Verified to 10⁻⁶ Precision.

An experimental, 100% Safe Rust alternative for statistical data analysis. Clean-room mathematical engine built on Apache Arrow and Polars with native Model Context Protocol (MCP) integration for AI-guided empirical research.

Academic Notice: StatCraft is an independent open-science software project. It is not affiliated with or certified by IBM SPSS Statistics. Do not use for clinical diagnostic decisions or legal compliance without independent statistical replication.
20 / 20
Audit Blocks Verified
Independent validation against SciPy Golden Oracle (Δ < 10⁻⁶)
0 Panics
100% Safe Rust Architecture
No unwrap, no unsafe, typed Result<T, E> error propagation
Apache Arrow
Columnar Data Engine
Zero-copy memory layouts, Polars vectorization, SPSS .sav codec
MCP & IPC
Autonomous Agent Bridge
Real-time JSON-RPC socket 18492 for Antigravity, Claude, and local LLMs

02MATHEMATICAL AUDIT

20 OF 20 VERIFIEDPRECISION TOLERANCE: Δ ≤ 10⁻⁶

The 20-Block SciPy Golden Oracle Matrix

Every statistical procedure in StatCraft is verified against reference Python/SciPy implementations and classical academic benchmarks. Clean-room implementations without proprietary IBM code.

IBM Manual Reference
pp. 368–370, 446–448, 221–226
Rust Algo Filescrates/statcraft_algo/src/
Key Formulas & Estimators
Spicer Moments

Recursive moments weighting via Spicer (1972)

Xˉ=∑wiXiW,s2=1W−1∑wi(Xi−Xˉ)2\bar{X} = \frac{\sum w_i X_i}{W}, \quad s^2 = \frac{1}{W-1}\sum w_i (X_i - \bar{X})^2
Bliss Skewness & Kurtosis

Unbiased sample skewness and Fisher excess kurtosis with SE

g1=W(W−1)(W−2)⋅∑wi(Xi−Xˉ)3s3g_1 = \frac{W}{(W-1)(W-2)} \cdot \frac{\sum w_i (X_i - \bar{X})^3}{s^3}
Yates Continuity & Fisher Exact

2x2 table continuity correction & exact hypergeometric test

χYates2=W(∣O11O22−O12O21∣−0.5W)2R1R2C1C2\chi^2_{Yates} = \frac{W (|O_{11}O_{22} - O_{12}O_{21}| - 0.5W)^2}{R_1 R_2 C_1 C_2}
Audit Notes:HAVERAGE canonical quantile interpolation r = p*(N+1) verified.
GitHub

03AGENT-FIRST PROTOCOL

LOCAL IPC SOCKET: 127.0.0.1:18492MODEL CONTEXT PROTOCOL (MCP 1.0)

Model Context Protocol (MCP) & Local IPC Socket 18492

StatCraft is built from the ground up for the agentic era. Any AI assistant—Antigravity, Claude, or local LLMs—can autonomously inspect variables, execute commands, and write peer-reviewed interpretations directly into the Output Viewer.

Zero-Hallucination Metadata

Agents query the exact VariableDictionary directly from RAM. No guessing variable names, types, or missing value codes.

CommandBus Execution

All procedures execute through typed commands with full Undo/Redo tracking and SPSS syntax generation.

Direct Output Injection

AI writes structured APA-style tables and academic interpretations directly into the desktop Output Viewer.

statcraft-mcp-bridge
{
  "name": "statcraft_execute_syntax",
  "arguments": {
    "syntax": "T-TEST GROUPS=treatment(1, 2) /VARIABLES=cholesterol /CRITERIA=CI(0.95)."
  }
}
Socket 18492 Ready (Async Tokio)L5 / statcraft_bridge

04ARCHITECTURAL INTEGRITY

Built on 100% Safe Rust & Apache Arrow

Statistical software must never crash mid-calculation or produce silent NaN errors. StatCraft enforces strict memory safety, strict layered dependencies, and a zero-panic contract.

Zero-Panic Contract

All non-test code strictly forbids unwrap(), expect(), panic!(), and unsafe. Every potential error (singular matrix, division by zero, insufficient sample size) is propagated as typed Result<T, StatError>.

Clean-Room Implementation

Developed without a single line of proprietary IBM SPSS code or assets. Algorithms are derived strictly from published academic papers, ISO standards, and GNU PSPP open-source specifications.

Arrow Columnar Memory

Datasets live in contiguous Apache Arrow arrays. Numerical transformations and filtering execute via SIMD-accelerated Polars engines with zero GC pauses.

Vector-First Design System

Native desktop GUI built in egui/eframe. Custom DialogKit with strict Design Guard checks, sub-millisecond render cycles, and 0 Unicode tofu glyphs.

Layered Hierarchy (L0 → L6)

Enforced at compile-time by cargo xtask layers. No cyclic dependencies, no UI logic leaking into math.

L6statcraft_app
Desktop GUI (egui / eframe / egui_dock / DialogKit)
L5statcraft_syntax
CommandBus, .sps parser & command history
L4statcraft_reporting
Pivot Tables model, export (PDF, DOCX, LaTeX)
L3statcraft_charts
Vector charts (egui_plot, brushing & linking)
L2statcraft_algo
Numerical tests, linear algebra (statrs, faer)
L1statcraft_core
Apache Arrow / Polars DataSet & VariableDictionary
L0statcraft_io
Codecs: .stc Parquet, SPSS .sav, CSV, Excel
Gate: cargo xtask layersPass (Zero violations)

05INVITATION TO COLLABORATE

An Open Invitation to Researchers & Developers

We do not claim StatCraft is a ready-made commercial replacement. It is an open-science computational experiment. We need your domain expertise, your datasets, and your scrutiny to make scientific computing transparent and reproducible.

For Statisticians & Researchers

Challenge our numbers. Run your empirical datasets through StatCraft and compare results against SPSS, R, or Stata.

  • Test edge cases: ill-conditioned matrices, collinear predictors, tiny frequencies, and weighted survey strata.
  • Privacy First: never upload raw patient or proprietary data. Submit anonymized mock cases or summary correlation matrices.
  • Found a delta > 10⁻⁶? File an issue directly from the site without needing a GitHub account.

For Rust & Systems Engineers

Help us architect high-performance, safe computational foundations using modern systems programming.

  • Optimize linear algebra routines using faer and statrs without dropping to unsafe.
  • Contribute new statistical procedures with SciPy oracle regression tests.
  • Expand the Model Context Protocol (MCP) toolset for autonomous AI research pipelines.
Explore GitHub Issues & RFCs
«Open science cannot exist on proprietary black-box software. Reproducibility demands open code, open algorithms, and relentless peer verification.»
StatCraft Open-Science Project // MIT OR Apache-2.0