Variation, causes and the next run Available
Twenty suspects and one result that moves. Patok Lab narrows them with the runs you already have, says how much of the variation each one explains, and picks the single run that will confirm it — then records what actually happened, so the next question starts from evidence instead of memory.
Excel or CSV. Classify columns as inputs, responses, categorical factors and metadata; set units and physical bounds. Versioned saves; the source file is never modified.
Fourteen statistical procedures with plain-language readings, curve fitting with AIC ranking, response surfaces of the active model.
18 regression and 9 classification types under one verification protocol: cross-validation, held-out tests, residuals, learning curves, an honest verdict.
Classic designs and effects analysis when you plan; the validated sequential optimizer when you already have runs and want the next one.
Promote, run, observe. The table gets a new version with provenance; time-travel to any episode.
A progress report writes itself from the project’s records, with present mode and PDF. The agent runs the same engines with a local model.
One loop
A measured micrograph becomes a column. A robot path is kept with the run that used it. The model refits, the optimizer recommends, and the campaign remembers every step.
Batch porosity and thickness land in the runs table with provenance to each result.
Stand-off, speed and step-over become input columns, matched to the coupon they sprayed.
The surrogate refits on what you observed and proposes the next experiment.
What comes in
References with or without a PDF, each labelled by what you actually hold. Find the state of the art, fetch what is open access, index your own papers and search them page by page. A citation enters the notebook with its page and its evidence label attached.
What comes out
Pick a period and the report assembles from the project’s dated records: what was added, the models and their verdicts, the best run and its settings, the measurements with their overlays, the notes. Present it, export the PDF, or ask the agent to reorder the outline. No language model writes the numbers.
Finding literature needs a connection: only the text of your question leaves the machine, never your data. Your own PDFs are indexed and searched locally. The report is assembled deterministically by a built-in skill from the project’s records — the agent can only re-run it with outline actions.
What you can check
What it is not
Eleven runs of a 2⁴⁻¹ with center points: effects, Pareto, aliasing, curvature — then generate the follow-up design.
Parameters and results as one table: which variables explain the defects, and where the process is capable.
Six screening runs, a surrogate, the next experiment. Repeat until the response is where you need it.
Early access
Bring a customer drawing, a table of runs, a folder of micrographs or a part you spray, weld or mill. We want Patok Lab used on live work and told what breaks. The Community edition is free, with no time limit; macOS, Windows and Linux.
Get access Bring a drawing, some photos, a table or a part. Or just tell it.
Bringing it to a team, a shop or a university group? Ask for your team →
FAQ
No. The calculation engine is bundled; the statistical procedures, curve fitting and models run locally without anything to install.
Full and fractional factorials, Plackett-Burman, central composite, Box-Behnken and Latin hypercube designs in the classic layout, plus effects analysis of the runs you already have.
The sequential engine refits a surrogate and optimizes it inside your physical bounds; no acquisition function to tune, repeatable: the same seed gives the same answer, and every episode is kept with what you observed.
