What does your regression actually support?

Paste a data table and a formula. Your browser fits ordinary least squares exactly as statsmodels does - coefficients, HC3 robust errors, diagnostics, collinearity, influence - free, nothing uploaded. A paid run then interprets the fit or writes the statsmodels script that reproduces and fixes it.

Each example has a saved model run, so you can see the whole page for free.

Analysis

Data (CSV, TSV or semicolon-separated, header row first; up to 100,000 rows; $, %, thousands separators and dash cells are cleaned)

Drop a .csv / .tsv / .txt or

Formula (patsy, as in smf.ols; empty = first numeric column on the rest)

Terms: x, C(group), np.log(x), I(x ** 2), a:b, Q("column name"); - 1 drops the intercept; np.log(y) on the left. Text columns are read as categories.

Notes (optional: what the columns mean, units, how the data were collected, what you want to conclude)

Load set .json

OLS fit (free, in your browser, no AI)

Coefficients

Classical standard errors and t as statsmodels' summary(); HC3 robust errors with z, as statsmodels reports cov_type="HC3". Evidence: clear = p < 0.05 both ways and no collinearity flag; fragile = only one way, or inflated; none = neither.

Assumption checks

Residuals against fitted values

A fan or a curve here is what the Breusch-Pagan test and a missing term look like. The most influential rows are ringed.

Most influential rows

Ref	Line	y	Fitted	Leverage	Studentized	Cook's D

Flags

statsmodels-style summary

Free: no account, no run.

Anything to ask (optional)

Paste your data to price the run.

Raw model reply

The reply did not parse as the structured result, so it is shown as it arrived.

Your recent runs

What this does, and what it does not

The fit follows statsmodels' smf.ols(formula, data).fit(): patsy's treatment coding (the first level in sorted order is the reference) and term order, rows with a missing value in a used column dropped, least squares by QR, then the summary() statistics, HC3 robust errors, variance_inflation_factor, the robust Breusch-Pagan test and OLSInfluence. It was checked against statsmodels 0.14.6 on 1,000 random models - one to four numeric predictors, categoricals, logs, squares, interactions, missing cells, no-intercept fits - with identical term names and degrees of freedom and every statistic agreeing to within 2e-5 relative (6e-8 when the design is well conditioned).

Spreadsheet formatting is cleaned before the fit - currency signs, thousands separators, percent signs, accounting negatives, decimal commas in a semicolon table, and dash or #DIV/0! cells read as missing - and every cleaning is listed under the table. The free data.csv holds the table exactly as fitted, and the free reproduce .py refits it in statsmodels and checks every coefficient. Pin as baseline, change the formula, and the page compares the two fits, with the nested-model F-test when one contains the other.

The paid run reads only what the browser computed and your notes. It is told never to compute a new number, and the page checks every number it writes. A script is a starting point you run yourself on your own copy of the data. Derived from the agent skill @k-dense-ai/statsmodels (k-dense-ai/scientific-agent-skills, K-Dense Inc.). The example data are illustrative, not measured.