Skip to content
Conyso.
Conyso Labs · Founded 2026

Free Lean Six Sigma statistical software.

The free Lean Six Sigma statistical workbench. Minitab-grade analysis, deterministic, validated against NIST, no AI black box. Use it free in your browser, or self-host to keep your data in-house.

Free in your browser · Deterministic · Self-host from $49 · No LLM
Compare against▾
105 analyzes across 14 families
15 calculators · 8 platform tools
NIST StRD-validated to ≥10 significant digits
440+ automated tests, 367 statistical · 69 API · 6 monitor
Positioning

More than a stats package.

Bench began as a free alternative to Minitab. It is now a full Lean Six Sigma platform, 105 analyzes across 14 families, a DMAIC copilot, DFSS and Monte-Carlo, Voice-of-Customer, flow and Agile analytics, governance, and a published NIST validation.

Use it free in your browser. When you need it in production with your data staying in-house, on-prem, air-gapped, audit-ready, self-host it. That on-premise data control is the reason regulated teams (pharma, medical device, defense, finance) reach for the self-hosted tier.

And no AI black box. Every number is computed by validated statistical code, not generated by a model, reproducible from a documented algorithm, and defensible in an audit. Built by Conyso Labs.

See the full Bench vs. Minitab comparison →  ·  Try the free calculators →


What it does that a desktop tool cannot

The things a desktop tool structurally can't do.

DMAIC Copilot

An engine that reads your results and recommends the next step, with tollgates and a one-page A3 export. Rule-based, not generative: the same inputs always produce the same guidance.

DFSS & Monte-Carlo

Tolerance simulation, delivery forecasting, and sensitivity analysis. Minitab gates this behind Workspace / Engage; in Bench it ships in the base build.

Voice of Customer

Survey and Likert analysis with Cronbach's α, plus automatic comment Pareto, free-text categorised and ranked without an LLM.

Flow & Agile

Cycle-time analytics and a Monte-Carlo “when will it be done” forecast, the services-and-software side of Lean Six Sigma, where desktop tools are weakest.

Conyso Originals

Analyzes you won't find anywhere else, like the Variance Budget, which attributes total variation to its sources in a single view.

Shareable, embeddable results

Every analysis is a live URL you can link or embed, not a .mpx file mailed around. Results travel to where the decision is made.

Live data connector

Point Bench at a published Sheet or CSV URL and refresh in one click. The analysis stays current as the data moves.

Governance

Lock a verified result and keep an append-only audit trail, who ran what, against which data, when. Recorded, not remembered.

Numerical validation

Bench publishes its reproduction of the NIST StRD reference datasets, agreeing to ≥10 significant digits. The trust wedge no closed tool publishes.

Bench Monitor

A separate, self-contained continuous-SPC watchdog, its own deployable, that watches a live process and alerts on out-of-control signals.


Every number traces to a documented algorithm. No model. No guesswork. No black box.


What's in the box

A full Black Belt toolkit.

Hypothesis tests

28 tests, one-sample, two-sample, k-sample, paired (including McNemar for binary paired), and the non-parametric equivalents (Mann-Whitney, Wilcoxon, Kruskal-Wallis, Friedman, Mood's median, sign test, runs, Grubbs, Dixon Q, Fisher's exact). Post-hoc comparisons include Tukey HSD, Fisher LSD, Games-Howell, Dunnett, and Hsu MCB. Pre-flight assumption checks and a Test Chooser wizard whose every recommendation drops you into the right form, no broken paths.

Control charts

13 chart types, Xbar/R, Xbar/S, I-MR, MA, p, np, c, u, CUSUM, EWMA, Hotelling T2, MEWMA, Laney p′ (overdispersion-adjusted), Z-MR (short-run). Western Electric and Nelson rules applied automatically. The multivariate and overdispersion-adjusted specialties Minitab markets as paid features ship as standard. Annotate any point. Save any chart as a recipe.

Capability analysis

Cp / Cpk / Pp / Ppk / Cpm / Z-bench, computed properly, σwithin from subgroup R̄/d₂ (AIAG) or moving-range MR̄/1.128 for individuals, not collapsed to overall sigma. Attribute capability for binomial and Poisson processes. Full Gauge R&R in all three designs, crossed, nested (parts nested in operators for destructive testing), expanded (extra variance sources via factor_cols), all three implemented and dispatched. The Capability Sixpack on a single screen. Predictive Cpk for what-if scenarios.

Regression and DOE

The full regression ladder, each reachable from the Method dropdown, OLS, GLM, logistic and ordinal, robust, quantile, ridge / lasso / elastic-net, PLS, spline, and random forest. DOE covers factorial, fractional, response-surface, and D- and I-optimal designs, with power and sample-size. Multi-response Derringer-Suich desirability optimization.

Multivariate, time series & reliability

PCA, LDA, clustering, MANOVA, and factor analysis. ARIMA and auto-ARIMA with decomposition and ACF / PACF. GEE and GLMM for correlated and hierarchical data, plus Bayesian methods. Reliability and survival analysis across nine distributions including Weibull, with right-censoring throughout.

Sample size & power

Eleven case-types covering one/two-sample t, paired t, one/two-proportion, one-variance, one-way ANOVA, correlation, chi-square goodness-of-fit, equivalence, and Poisson rate.

DMAIC project layer

Every workspace can hold projects with five tracked phases (Define / Measure / Analyze / Improve / Control), default checklists per phase, and pinned analyzes. A Process Health Score per project blends Cpk (40%), control-chart violations (30%), MSA acceptability (15%), and trend stability (15%) into one number.

Seventeen LSS deliverable templates

Charter, SIPOC, A3, FMEA (auto-RPN), AIAG-style Control Plan, 8D, Voice-of-Customer/CTQ, Pugh matrix, Capability Report, MSA Report, Tollgate Deck, Closure Report, Kaizen, 5S, RACI, Stakeholder, and an IQ/OQ/PQ Validation Kit Stub, each renderable to HTML, Markdown, Word, or PowerPoint (one slide per section).

The UX layer other tools don't have

A Test Chooser wizard that picks the right test from three or four questions, and routes every recommendation to a real form, McNemar through Hsu MCB. A plain-English query bar that fills the form. Pre-flight traffic lights for assumption violations. A "what this means" interpreter that gives every result a plain-English paragraph.

Rule-based action plans, "Cpk = 0.6 → run Gauge R&R first; takes a day, saves you a week." An eight-step anchored-spotlight guided tour on first launch, a real anchored tour with a bronze pulse-ring around the actual target element, not a popup; replayable from the header.

Every result is a cross-link

Below every result card: Learn more · Method provenance · Run again. The Methods page has Run / Open affordances per row. Guides have in-text runnable links. One navigate() router; consistent data-nav-* attributes across the SPA. The whole tool is densely interconnected, click a method anywhere, you are in the form.

DMAIC phase recommendations

Each project phase surfaces a pill-button row of recommended analyzes, Pareto for Define, MSA + Baseline Cpk for Measure, ANOVA + Hsu MCB + Regression for Analyze, DOE + Multi-response desirability for Improve, I-MR + Tolerance interval for Control.

The community decides the roadmap

Bench ships with an in-app feedback board. Anyone can submit a feature request or a bug; everyone can upvote or downvote. Top-voted items feed the next development cycle. The roadmap is not closed.

Learn, in the workbench

In-app Learning Paths and sample datasets, free for a whole cohort, alongside fourteen in-depth guides, ~25 FAQs, and a per-method reference where every row drops you into the analyzer. Bench ships with 440+ automated tests (367 statistical · 69 API · 6 monitor), CI-runnable and citable, and publishes its NIST StRD reproduction to ≥10 significant digits.


In the workbench

A Capability run, end-to-end.

Cycle time on a 248-sample run. Bench computes Cp, Cpk, Pp, Ppk against your spec limits, returns the σwithin from subgroups (not the overall sigma shortcut), plots the distribution against LSL / USL, and writes the interpretation in plain English, including the predicted defect rate in DPMO and the off-centring in process-sigma units. Pin, save, copy, CSV, delete. Every analysis carries the same affordances.

tolerance · LSL–USL −1σ +1σ LSL USL μ
Capability: the process spread against the spec limits (LSL/USL).
Bench — Capability (Cpk) view showing N=248, Cpk=0.87, distribution vs LSL/USL, and a plain-English interpretation.
Capability (Cpk) · Bench

Bench and Lens

The math, and the picture.

Bench is the math. Lens is the picture. There is no model anywhere inside either of them. Bench is the deterministic math layer, it stands alone and free, and it is the dataset Lens visualises.

Because Bench is deterministic and verifiable, every number traceable to a documented algorithm and the whole engine reachable over an API, anything built on top of it can call the statistics rather than guess at them. That is a property of the engine, not a feature inside it: Bench does the math, the same way every time, whatever is asking.

Lens, now live →  ·  Open Lens ↗


Pricing and licence

Free in the browser. Yours to keep, self-hosted.

Run the full workbench in your browser at no cost. Self-host when your data has to stay in-house: $49 for an individual, $499 for a company, one-time and perpetual, the same model across every Conyso Labs tool.

What it costs

$499 once against Minitab at about $1,700 per seat, every year. Self-hosting is a single docker compose up, SQLite and filesystem storage, no Postgres, no Redis, no S3.

What the licence gives you

A perpetual licence that never expires and never phones home. Run the whole engine on your own infrastructure, offline and air-gapped, so your data never leaves your network. No per-seat fees.

The browser tier is free for everyone, so cost is never the barrier to using Bench. The self-hosted tier is paid because it does one thing: keep your data inside your own walls, where regulated teams need it. We could have made it free and lived on ads, harvested data or grant cycles. A fair one-time price funds the engine's upkeep and means Bench answers to the people who run it rather than to whoever pays to steer it.

Questions about self-hosting or volume licensing? Email krishna@aetris.co. Full pricing →  ·  How the licence works →


Free, no signup

Quick calculators from the same engine.

Need one number, not the whole workbench? These run the same deterministic math in your browser, free, with nothing to install.

Part of the Conyso ecosystem. Conyso. ↗


FAQ

Questions, answered.

Is Bench a free alternative to Minitab?

Yes. Bench runs 105 analyses across 14 families, capability studies, control charts, hypothesis tests, gauge R&R, regression and designed experiments, free in your browser and validated against the NIST Statistical Reference Datasets. Minitab has decades of validated tooling, training and formal support; where you want the core Lean Six Sigma statistics without a per-seat licence, Bench is a real substitute.

Do I need to install anything or sign up?

No. Bench runs in the browser with no signup. Self-host it for a one-time licence when your data needs to stay inside your network.

Is my data uploaded anywhere?

No. Every calculation runs locally in your tab; nothing is sent to a server, and self-hosting keeps it entirely in-house.

Are the results validated?

Yes. Bench's statistics are checked against the NIST Statistical Reference Datasets, the same reference the commercial packages use.

Does Bench use AI to compute the statistics?

No. Every number is produced by deterministic statistical code from a documented algorithm, reproducible and defensible in an audit.