1 min
The 30% problem: why formulations fail at scale-up
Why lab-perfect recipes break in the plant — and how process digital twins with quantified risk help.
scale-upResources
Chemistry, process, and AI — written for the people who challenge the model.
Long-form notes on physics-informed methods, formulation, uncertainty, regulation, and scale-up. Real numbers. No gating for the first read.
All writing
The library
11 articles
1 minFirst principles. Then ML. Always uncertainty.
The Lavoisier prediction contract — science baseline, ML residual, and confidence — so design reviews can challenge the answer instead of accepting a black box.
physics-informed
1 minProcess digital twins before you commit capital
Turn a formula into a manufacturable process with thermodynamics and quantified scale-up risk — before pilot spend and plant modifications.
digital-twin
1 minPhysics-informed vs black-box: a side-by-side on the same problem
Same mixture viscosity task, two modelling philosophies — what each gets right and where each fails.
physics-informed
1 minPlant agents that reason about chemistry — not just setpoints
Process Control Agents keep operations on-spec by reasoning about constraints, incomplete data, and chemical risk — with humans still in the loop.
process-control
1 minUK REACH after Brexit: what actually changed for formulators
A practical overview of UK REACH divergence that formulators need when screening candidate materials.
REACH
1 minWhy black-box AI fails chemical design reviews
Design reviews demand mechanism and uncertainty. Black-box models can fit history and still lose the room when someone asks 'why'.
trust
1 minCompressing coatings R&D: fewer failed experiments
How inverse formulation and uncertainty-aware ranking shrink trial-and-error loops in coatings and specialty formulations.
coatings
1 minWhat a confidence interval means in a formulation prediction
Uncertainty quantification for chemists — confidence, extrapolation flags, and when not to trust a number.
uncertainty
1 minFrom molecule to plant: one chemical data backbone
Why formulation, process design, and plant control need a shared CAS-keyed backbone — not three disconnected AI tools.
platform
1 minWhy mixture properties aren't averages
The physics of non-ideal mixing — with a viscosity worked example chemists can check.
mixtures
Stay close
New notes when we publish.
Formulation, process, and plant AI — without the marketing fog.