The Uniqcret Research Suite Is Live — Six Research Apps, One Sign-In

Most of what I write on this blog is about method — calibration, validation, sample size, the unglamorous decisions that quietly determine whether a result means anything. For the past several months I have been turning some of that into software instead of prose. Today the whole set is live.

The Uniqcret Research Suite is six research applications you open in a browser and use on your own study data. They live at webapps.uniqcret.com under /research/, and they ask you to be signed in with a Uniqcret account.

What just went live

The site already had a shelf of small interactive visualisers — one idea each, open a tab, move a slider, close it. The suite is a different animal: full applications that hold a project, carry you from data in to draft out, and are built for the study you are actually running rather than a toy example.

Each app is served at its own address, and each sits behind a signed-in Uniqcret membership. Which membership includes them is described on the site itself — I have deliberately kept those details out of this post so there is only one place that has to stay correct.

The six apps

Stat Prep Toolkit

The bench you reach for before the analysis. It holds 33 in-browser calculators for the recovery work that eats an afternoon of any systematic review: getting a missing standard deviation out of a standard error, a confidence interval, a t or p value, a median and IQR, or a change-from-baseline; pooling subgroups into a single arm; converting between effect measures. There is also a figure digitiser for reading numbers straight off a published plot when the paper never reported them. Every calculator shows the formula it used, so you can check the arithmetic instead of trusting it. Open Stat Prep Toolkit.

Randomizer

Reproducible allocation for a clinical trial. It generates simple or permuted-block sequences, handles stratification, and works from a stated seed so the same sequence can be regenerated and defended later. You get two exports — a full audit list and a concealed one for the people who must not see what comes next — and a paste-ready methods paragraph describing exactly what the app did. Open Randomizer.

Causal DAG Builder

Name your exposure and outcome, connect them with arrows, read what to adjust for. The adjustment-set advice comes from the DAGitty engine, but it is reported in plain clinical language rather than graph theory, which is the part I wanted most. Share links carry the whole diagram, so a co-author opens your assumptions and can argue with them, instead of squinting at a screenshot. Open Causal DAG Builder.

Sample Size Studio

Eight sample-size benches — single group, two means, two proportions, non-inferiority, equivalence, repeated measures, survival, and prediction-model sizing that calls the real R pmsampsize package. Each bench returns the required n together with a manuscript-ready methods statement. If you do not yet know which design you are in, there is an AI consult bench: describe the study in prose, answer a short guided walk of questions, and it proposes a bench with the inputs already staged — which you then check and adjust yourself, because the proposal is a starting point, not a verdict. Open Sample Size Studio.

Missing Data Advisor

A six-step wizard for the question everyone postpones. It profiles your dataset, shows you where the gaps actually are, gives plain-language advice about what the missingness pattern implies and what it does not license you to assume, and then runs genuine R checks on your data — see the next section for how that works. It finishes with a take-home pack you can drop into a supplement. Open Missing Data Advisor.

Meta-Analysis Studio

Upload, analyse, review, draft. It starts from your question rather than from a model name — two groups on a number, two groups on a yes/no outcome, a single proportion, or effect sizes you have already computed elsewhere. Pooled estimates use REML with a Hartung-Knapp adjustment by default. Forest plots render in four journal layouts (Default, JAMA, RevMan 5, BMJ). Heterogeneity, publication-bias advisories and GRADE certainty each get their own reviewable step, and the export gives you manuscript-ready Methods and Results alongside a reproducible R script. Open Meta-Analysis Studio.

Your data stays in your browser

Two of the apps — Missing Data Advisor and Meta-Analysis Studio — do their statistics in real R, compiled to WebAssembly and running inside your browser tab (the webR project). This is not a server somewhere that runs R on your behalf. It is R itself, on your machine, in the tab.

That has three consequences worth knowing before you click:

• Your data is never uploaded. There is no upload step to opt out of, because there is nowhere for the file to go. The spreadsheet you drop in is read by code already running on your computer.
• The first run downloads an engine. The R runtime is tens of megabytes — Missing Data Advisor asks for consent first and quotes roughly 48 MB. After that it is cached, so every later run starts immediately, including offline. On a slow connection the first time is a genuine wait, and I would rather say so than let it surprise you.
• The numbers are R's numbers. Not a JavaScript reimplementation that agrees with R most of the time.

One habit worth building: across the whole suite, projects and analyses persist in your browser's own storage, not on a server. Your work survives closing the tab, but it does not follow you to another laptop, and clearing site data removes it. Export anything you want to keep.

The AI application harness

The chat at chat.uniqcret.com has a mode called the Application harness — the 🧰 domain in the composer. Select it and the assistant can drive a suite app embedded beside the conversation: open it, stage and validate your inputs, deep-link a particular state so you can take over by hand, and emit a reproducible R script for what it set up. For Missing Data Advisor it goes end to end — the in-browser R analysis runs inside the chat and reports back.

The honest scope: not every app runs in-chat. Meta-Analysis Studio deliberately stages and validates only. Starting its R engine is a download of around 100 MB, and in that surface there is no consent step in front of it, so I would rather you press that button yourself than have a chat message spend your bandwidth on your behalf. When you want the analysis actually run, open the app.

How to get in

Sign in at webapps.uniqcret.com with your Uniqcret account and open the researcher section — the six suite apps sit there alongside the smaller visualisers. The suite is included with the Pro (฿300/month) and Max (฿1,000/month) plans. Or go straight to one:

• Stat Prep Toolkit
• Randomizer
• Causal DAG Builder
• Sample Size Studio
• Missing Data Advisor
• Meta-Analysis Studio

Any current desktop browser will do. The two webR apps want a little memory and a little patience on that first load; after it, they are the fastest way I know to run R without installing R.

A few honest notes

The Thai interface in some of the suite apps still carries an upstream flag saying the translation has not been reviewed by a human. I would rather ship it flagged than pretend it is finished. If you meet a term that is wrong — and in Thai statistical vocabulary there is plenty of room to be wrong — please tell me, and I will fix it.

None of these apps is an oracle. Every one of them shows its formula, its method text, or its R code, and that is on purpose: a number you cannot interrogate is not evidence, it is decoration. If a result surprises you, the app should let you find out why. If it does not, that is a bug I want to hear about.

Go and break something. I will be reading.