Platform
Know what scale will do before the furnace does.
Every ScaleNexus engagement runs on our own software platform: computational screening, thermodynamic simulation and lab-to-plant property prediction, with the uncertainty stated on every number.
01 · Four gates
Every candidate earns its way to the plant.
A chemistry advances only when it clears the gate in front of it. Cheap checks run first, expensive ones last — so lab time and pilot heats are spent on the few candidates that deserve them.
01Empirical screen
Stability, in milliseconds.
Electronic-structure indices computed from the chemistry flag phase-instability risk before anything is simulated. Borderline candidates are flagged, not silently passed.
02Thermodynamic simulation
What the alloy wants to become.
Equilibrium and solidification simulation predict phase constitution, segregation and freezing range — reported as a band between the two, so the answer carries its own spread.
03Lab verification
Prediction meets measurement.
Measured phase fractions are compared against the predicted band. Every discrepancy is logged with the versions that produced it, and methods that disagree are reported side by side, never averaged away.
04Acceptance at scale
Judged on the pessimistic case.
Predicted plant-scale properties are checked against your targets at the 10th percentile, not the median — so a pass means the part meets spec in the realistic worst case.
02 · Lab to plant
What changes when the ingot gets bigger.
A coupon cools in seconds; a production ingot cools over hours. Coarser microstructure, more segregation, different strength. The platform predicts that shift — and how sure it is.
- Two independent estimates, fusedAn empirical transfer learned from matched lab-and-plant data, and physics-based models of cooling rate, microstructure scale and grain size. Where they agree the interval tightens; where they disagree the result is flagged and widened.
- Distributions, not point valuesEvery cross-scale prediction is a median with a 10th–90th percentile interval. With little data the interval is wide — and says so — instead of pretending to precision.
- Process windows and costDiffusion models size the homogenization schedule under the incipient-melting ceiling, segregation-defect risk is scored at production section sizes, and cost per kilogram is built from raw materials and processing.
- Reproducible decisionsEvery result pins the data, thresholds and models it depended on. A decision made today can be reproduced — and defended — a year from now.
03 · Evidence
Built on data you can trace.
The platform harvests open technical literature and public materials datasets, and uses AI-assisted extraction to turn tables and text into structured records — every value carrying its own source and location in that source.
AI reads and explains; it never decides. Rankings and gate decisions are deterministic arithmetic over the data, and any narrative the platform writes is checked number by number against the report it describes.
Per-value provenance, not per-document
Human review queue for low-confidence extractions
Physical sanity checks flag values — never silently drop them
04 · What you receive
A decision report, not a black box.
Each candidate gets a decision report: composition in both unit systems, every gate with its inputs, intervals shown as intervals, and every flag — set or unset — so “not flagged” is never confused with “not checked.”
Values that still depend on constants awaiting calibration are marked as such. You see exactly how much of an answer is measurement, how much is model, and how much is assumption.
The platform is how we work, not a subscription. Its reports are delivered as part of every engagement.
See what your material does at scale.
Send us the chemistry and your targets. We'll tell you which gate it clears — and which one it doesn't — before you commit to a pilot heat.