Research portal · being built

Rigorous statistics, without writing code.

Door is being built as a research portal for scholars, PhD candidates and MBA students: open your CSV file, choose an engine such as PLS-SEM with bootstrap or DEA, and read the results in your browser. The engines are written in Arwad, an R-syntax language that compiles to native code. The first are built and verified, but the portal does not run them yet; this page says exactly what exists today and what comes next.

Not open yet: the engines are built and verified in Arwad first

Where things stand

Nothing runs here yet, and that is on purpose.

Every engine must first reproduce its reference R package on 50 datasets (Arwad, Phase 2). Only then is it built for WebAssembly and opened in the browser (Phase 3). Until an engine passes, this page lists it as planned.

  1. Built today

    The Arwad compiler

    R functions compiled to native code, with results bit-identical to GNU R across a 201-function test corpus on Linux, macOS and Windows.

  2. In progress

    The statistical engines

    Built so far in Arwad: PLS-SEM with bootstrap, and regression and GLM, each reproducing its reference R package bit for bit. The rest follow (Phase 2).

  3. Then

    A browser build

    A WebAssembly build, so the computation runs on your own device (Phase 3). That is when the portal opens.

Arwad, Phase 2

The engines

Each engine is written in Arwad and checked against its reference R package on at least 50 datasets before it reaches the portal.

  • Built in Arwad · not yet in the portal

    PLS-SEM

    Partial least squares structural equation modelling: the path algorithm with bootstrap. The engine reproduces seminr bit for bit: estimates, loadings, weights and bootstrap draws.

    Reference: seminr

  • Planned · Phase 2

    Data envelopment analysis

    CCR and BCC models, with input or output orientation.

    Reference: Benchmarking

  • Built in Arwad · not yet in the portal

    Regression and GLM

    Least squares and generalised linear models (logistic, Poisson, Gaussian), with bootstrap. The engines reproduce lm and glm.fit bit for bit.

    Reference: stats::lm, stats::glm

  • Planned · Phase 2

    Panel fixed and random effects

    With the Hausman test to choose between them.

    Reference: plm

  • Planned · Phase 2

    Meta-analysis

    Fixed-effect and random-effects (REML) models.

    Reference: metafor

  • Planned · Phase 2

    Time series: ARIMA and GARCH

    The ARIMA likelihood and the GARCH(1,1) family.

    Reference: stats::arima, rugarch

Source: docs/ROADMAP.md in the Arwad repository.

Planned

How the portal will work

  1. Open your data

    Choose a CSV file on your computer. The plan is for the engines to run inside your browser, so the file is processed on your own device.

  2. Choose an engine and its options

    For example PLS-SEM with 10,000 bootstrap resamples and bias-corrected percentile intervals, or DEA with input or output orientation.

  3. Read the results

    Tables of estimates with their confidence intervals, from engines that reproduce their reference R packages.

Tied to Arwad

Roadmap

The portal opens at the end of Arwad’s Phase 3. No dates are promised; every phase ends on a measured exit criterion.

  1. Done, 24 October 2026

    Phase 0 · Foundations

    A hand-written Goose bootstrap kernel, called from R, ran 1.59× faster than R’s fastest loop, allocated about 46× less memory in R, and matched R to about 15 significant digits.

  2. Done, 27 October 2026

    Phase 1 · Compiler

    aw_compile() compiles loop-heavy R functions: 11–33× faster than R’s byte code on one laptop, bit-identical to GNU R across a 201-function corpus, with no memory errors under AddressSanitizer.

  3. In progress

    Phase 2 · Statistical engines

    Three engines are built (regression, GLM and PLS-SEM), each matching its reference package on at least 50 datasets; DEA, panel models, meta-analysis and ARIMA/GARCH follow.

  4. Planned

    Phase 3 · WebAssembly and the browser

    Exit criterion: a 5,000-replicate bootstrap in the browser, and a methods paper submitted. That is when Door’s portal opens.

Details and figures: langwi.com, and the benchmark reports at github.com/bashar-al-agha/arwad.

Stay informed

Follow the build

There is no sign-up form on this page, and no account to create.