Statistics you can defend in front of a reviewer.
Upload a dataset from any field. Celeus suggests the right statistical method and tells you why, runs it deterministically in R, and interprets the result.
| Method | Published | Celeus | Match |
|---|---|---|---|
| Mixed model, F testPinheiro & Bates (2000)Published: 22.36 | 22.36 | 22.3556 | Matches the published value to its printed precision |
| Bland-Altman agreement, biasBland & Altman (1986)Published: -2.1 | -2.1 | -2.11765 | Matches the published value to its printed precision |
| Equivalence test (TOST), tLakens (2017)Published: 3.14 | 3.14 | 3.13917 | Matches the published value to its printed precision |
Why you can trust the numbers
The exact R script in every package
Each run downloads with its plan, R script, seed, session info, results and audit log, runnable in plain R.
Seeded and pinned
Every analysis runs with a fixed seed on pinned package versions, so the same run gives the same numbers.
Data rows never reach the AI
The AI works from column summaries, dataset names, aggregate results and text you write. The statistics are computed in R.
What reaches the AI
Student's sleep data, run through Celeus
The data come from the 1908 paper that introduced the t-test. Each value in the table is the engine's own output, and each one matches an independent R computation from the validation run.
Paired t-test
| Mean difference (95% CI) | -1.58[-2.459890, -0.700114] | matches the independent R computation |
|---|---|---|
| t (df) | -4.06213 (9) | matches the independent R computation |
| p | 0.00283289 | matches the independent R computation |
| Cohen's dz (95% CI) | -1.28456[-2.118020, -0.414628] | matches the independent R computation |
A paired t-test found a difference between the two conditions, t(9) = -4.06, p = .003, mean difference = -1.58, 95% CI [-2.46, -0.70], dz = -1.28, 95% CI [-2.12, -0.41].
A tick means the value matches the independent R computation. See all checks
What reaches the AI, and what never does
Every path from your data to the AI model passes a guard that refuses data tables. Every upload is scanned for identifiers, and each column the scan flags is excluded from the AI by default until you decide what happens to it. A column the scan does not flag is summarized for the AI like any other column, so exclude any column you know holds identifiers.
Reaches the AI
- Column names, types and variable labels stored in your file
- Summary statistics, including each numeric column's minimum and maximum
- Category names, with small counts withheld
- Aggregate results of your analyses
- Text you write: chat messages, research questions, notes and codebook descriptions
- Dataset names (by default, the name of the file you uploaded)
Never reaches the AI
- Your uploaded file
- Data rows or data tables
The statistics are computed in R, never by the AI. An AI model writes a plain-language interpretation of each run (on by default, and you can switch it off), drafts and edits your text, answers your questions in chat and, when you ask, suggests a test.
Text you write is sent as you wrote it, so keep identifiers out of it. That matters most for patient records, and just as much for survey responses, student records or anything else collected under a confidentiality promise.
How Celeus handles your dataHow it works
Upload
CSV, TSV or Excel on every plan; SPSS, Stata and SAS on paid plans. Or add one of the public sample datasets to try it first.
Method chosen and explained
Celeus suggests a method for your question and your data, grades the alternatives it considered, and checks the assumptions. If you pick something else, that choice is recorded.
Deterministic run
Seeded, in R, on pinned package versions. Run it again on the same engine and you get the same numbers.
Interpretation and flags
A plain-language reading of the result, written by an AI model from the aggregate results (on by default), with failed assumptions and other problems flagged where you will see them.
Reproducible package
The plan, the exact R script, the seed, session info, the results and the audit log, in one download.
See a run from upload to package
Transcript
Celeus: from a dataset to a reproducible result.
We drop in a synthetic dataset: forty rows, five columns.
Celeus scans every column first. Two of them, patient name and MRN, look like direct identifiers.
The review recommends replacing both columns with coded stand-ins. We apply the recommendation as proposed.
That creates a new, de-identified version. The original upload is kept, and a fresh scan finds no identifiers.
Next, Celeus ranks suitable analyses from the de-identified profile, without calling an AI model. To compare the outcome between the two arms, it recommends Welch's t-test.
The test, outcome and group come pre-filled. We add a research question, which is saved with the results.
The analysis runs in R, the statistical programming language. The wait is shortened in this video.
The result: the treatment group's mean is 6.7 points higher, with a 95 percent confidence interval of 4.25 to 9.15, and p below 0.001.
Below it is a plain-language summary written without AI, and a warning: with forty rows, only large effects can be detected reliably.
An AI interpretation is added on top, clearly labelled. The AI works from summaries and aggregate results, never your data rows, and the computed numbers remain the ground truth.
Here is the exact R code that reproduces this result, with its seed and the engine version.
We check the results and confirm them right here, next to the download. The confirmation is saved with the run, and the reproducible package unlocks: the data, the script and the pinned package versions, together.
40 statistical methods
Every plan gets every method. The validation report lists the cases compared.
Comparing groups
- Welch's t-test
- Student's t-test
- Paired t-test
- Mann-Whitney U test
- Wilcoxon signed-rank test
- One-way ANOVA
- Welch's ANOVA
- Kruskal-Wallis test
- Repeated-measures ANOVA (Greenhouse-Geisser)
- ANCOVA (analysis of covariance)
- Two-way ANOVA (factorial)
Association and correlation
- Chi-squared test of association
- Fisher's exact test
- Pearson correlation
- Spearman correlation
- McNemar's test
Regression
- Linear regression
- Logistic regression
- Firth penalised logistic regression
- Poisson regression
- Negative-binomial regression
- Ordinal logistic regression
- Multinomial logistic regression
- Factors associated with outcome (continuous outcome)
- Factors associated with outcome (binary outcome)
Mixed-effects models
- Linear mixed-effects model (random intercept)
- Mixed-effects logistic regression (random intercept)
- Mixed-effects Poisson regression (random intercept)
Survival and censored data
- Peto-Peto generalized Wilcoxon test
- Censored regression (Tobit-style)
- Kaplan-Meier (log-rank test)
- Cox proportional-hazards regression
- Weibull accelerated failure time regression
- Factors associated with outcome (time-to-event outcome)
Diagnostic accuracy and agreement
- ROC curve / AUC
- Diagnostic accuracy (2x2)
- Bland-Altman agreement
- Compare two ROC curves (DeLong)
Descriptive
- Descriptive table (Table 1)
- Descriptive summary (mean, SD, SE, quartiles)
Pricing
$19 a month per researcher on Pro, or $19 a month per seat on Team (minimum 3 seats).
- Free$0
- Pro$19
- Team$19
- EnterpriseCustom pricing
Common questions
How do I know the numbers are right?
Every statistical method is checked on reference datasets against an independent computation in R, and a subset of results is also checked against values printed in published papers and package documentation. The validation report lists each case and each compared number. Every run also ships its exact R script, so you can re-run it yourself.
Read the validation reportCan I see the code behind my result?
Yes. Every run comes with a reproducible package: the exact R script, the random seed, the package versions and the session information, which you can run in plain R. The source code of Celeus itself is not public.
What does the AI see?
The AI model receives column names, types and variable labels stored in your file; summary statistics, including each numeric column's minimum and maximum; category names, with small counts withheld; aggregate results of your analyses; text you write: chat messages, research questions, notes and codebook descriptions; and dataset names (by default, the name of the file you uploaded). It never receives your uploaded file, data rows or data tables. The statistics themselves are computed in R, not by the AI.
How your data is handledHow is Celeus different from a desktop statistics package?
Unlike a desktop package, where you pick the test yourself, Celeus suggests a method when you ask for one, checks the method's assumptions, and runs it deterministically in R with a fixed seed and pinned package versions. Every run produces a reproducible package, and the AI can draft Methods and Results text from the run's numbers for you to edit.
How much does it cost?
There is a free plan with 10 analysis runs a month. Pro costs $19 a month or $190 a year for one researcher, and Team is $19 a month or $190 a year per seat, with a minimum of 3 seats.
See plans and limitsHow long do you keep my data, and can I delete it?
You can delete datasets and runs individually at any time, and deleting your account is self-service. How long reproducible packages and backup copies are kept is set out in section 4 of the Privacy Policy.
Read the Privacy PolicySee it work on your own data.
The free plan includes 10 analysis runs a month and all 40 statistical methods.