Fast columnar computation. CPU-native. No GPU required. Same answers as numpy and scipy — faster — plus operations that don't exist in any other tool: sensitivity surfaces, condition discovery, regression discontinuity, and deep conditional chains that compute on history conditioned by the previous step.
Live, on a 4-vCPU AMD virtual machine with no GPU. Pick an operation; the demo runs the engine on a 100,000-row dataset and reports the result and the time.
The dataset has 100,000 rows with four numeric columns
(x, y, z, season).
Pick an operation; results return in milliseconds.
Rate-limited to 10 requests per minute per IP. Numbers vary by load. The live host is a single small virtual machine; production installs run on whatever hardware you already use. The four cyan-bordered buttons run conditional chain mathematics — operations that don't exist in pandas, numpy, scipy, R, or MATLAB.
Six operations that exist because the engine can do thousands of conditional computations per second. Each one is what you'd otherwise spend a week wiring together by hand from scipy primitives — and run overnight when you do.
Sweep a condition threshold across its full range; get back a curve (or a 2D heatmap, two conditions at once) of how a statistic moves. 50 conditional computes on 100,000 rows in ~60ms. Try it: the live demo runs this on the sensitivity-surface button.
Hand it a target statistic and a list of candidate columns; it finds the column AND threshold that maximizes (or minimizes) that statistic. Auto-condition discovery, no hand-tuning. ~80 conditional computes in under 200ms.
Each step computes on history conditioned by the previous step's result. Mean of price → std where price > that mean → mean volume where price > mean+std → ... 10-step chains complete in <100ms. The thing nobody else has, because nobody else is fast enough.
Estimate the jump in an outcome at a threshold of a running variable, across multiple bandwidths, automatically. Causal inference that takes a graduate-econometrics homework afternoon in scipy and runs in under 10 seconds here.
Find the regime variable: which column, at which percentile, causes the target's distribution to change. 60 distributional tests in <5 seconds. Surfaces structure you'd otherwise miss until production exposes it.
Decision-tree-grade split-finding: find the threshold that maximally reduces variance of the target. 40 splits on 100,000 rows in <80ms — orders of magnitude faster than fitting a tree.
Replaces the parts of pandas, numpy and scipy you actually use, on the hardware you actually have.
Mean, standard deviation, variance, min, max, median, quantiles, summary — all the basics, sub-millisecond on 100,000-row tables.
Pearson correlation, covariance, linear regression with prediction, Spearman rank correlation, t-tests and chi-square — same answers as scipy.stats.
Skewness, kurtosis, IQR, coefficient of variation, geometric and harmonic means, trimmed means, mode estimates — the diagnostics you reach for first.
Dominant frequency, spectral entropy, autocorrelation, RMS, peak-to-peak, zero-crossing rate — signal-shape work without spinning up scipy.signal.
Shannon entropy, relative entropy (KL divergence), Euclidean norm, cosine similarity — the comparators you need for ML pipelines.
Trend slope, trend strength, stationarity score — quick reads on whether a series is moving, drifting, or stable.
Every result matches numpy / pandas / scipy reference to floating-point precision. Same answer, faster, on hardware you already own.
Everything in version 1.0. Imported with one line, used without ceremony.
Start free, no credit card. Upgrade only when you need more. Enterprise is a published meter, not a sales call.
Personal projects, course work, exploration. Generous enough to actually finish a project.
Launching soonIndependent researchers, hobbyists, anyone running real numbers on their own time. $190/yr billed annually.
Launching soonA developer running real numbers in real production every day. $590/yr billed annually.
Launching soonA small data team or a backend group standardizing on one engine. $1,990/yr billed annually.
Launching soonProduction fleets, batch pipelines, regulated environments where every random draw needs an audit trail.
Launching soonEvery tier ships the same engine. Same protection. Same operations. The license fee is the only thing that changes — the work happens on your machine.
We are not shipping a download yet. The live demo above runs the same engine
you'll install. When the wheel ships you'll do
pip install validiti-maths,
activate one license per machine, and import.
Reserve your launch slot: contact@validiti.com
Every tier — including Community — ships the same protection envelope.
