The right material exists. Finding it is the problem.

A UK defence manufacturer is bringing component production back onshore. One of the materials is a cellulose nitrate grade it has bought from the same mainland European supplier for decades. It is specified by that supplier’s own designation rather than by properties, because nobody ever needed to describe it any other way.

To source it in the UK, someone first has to work out what the grade actually is: nitrogen content, viscosity grade, damping solvent, plasticiser compatibility. Then find who in the country makes something equivalent. Supplier catalogues are organised by supplier, so there is no way to ask that question directly.

1. 01Paste the incumbent datasheet. The grade comes back described by what it is rather than by whose catalogue it sits in.
2. 02Set the binding constraints, nitrogen content and viscosity grade, and rank the field by how closely each grade matches. Filtering by region is in development, so for now you narrow the result to UK suppliers yourself.
3. 03The result is equivalent grades ranked by property match, each with the supplier that published it. Where a supplier publishes nothing on a critical property, the gap is named rather than guessed.

A shortlist of grades that hold the specification, with the supplier for each, and a clear list of what still needs testing before qualification. The same query answers supply-risk and dual-sourcing questions, not only reshoring.

You are specifying a grade for a moulded part that has to survive under-bonnet temperatures under load. Tensile and melt flow are on the datasheet. Heat deflection at 1.8 MPa is not, and neither is the shrinkage window that determines whether the part holds tolerance.

So the options are to email the supplier and wait two weeks per grade, or to buy samples and measure it yourself. That only works once the field is narrowed, which was the problem in the first place.

1. 01Paste the incumbent datasheet. The grade comes back described by its properties, its chemistry and its end use, rather than by the catalogue it sits in.
2. 02Set the floor on the property that actually governs the part, whether or not it was published. Predicted values are filterable and marked as predicted.
3. 03Read the shortlist by criteria met, then by similarity. Each candidate states which numbers are measured, which are model-generated, and which still need a lab.

A ranked candidate set with the unknowns named, so the lab list is short and every entry on it is there for a reason.

Commercial has a market. Compliance has a blocker. The usual sequence is that R&D reformulates for performance, compliance reviews the result, something fails, and R&D starts again. Each loop is weeks, and the number of loops is unpredictable because the compliance constraint was never inside the search.

Localisation compounds it: an ingredient that is compliant and available in Europe but unobtainable in the target region has not solved anything.

1. 01Restricted-substance lists for the target market are applied as a hard filter before ranking, so compliance review checks a shortlist that already clears, instead of finding the blocker on the third loop.
2. 02Properties the supplier never published are predicted, so a candidate is not dropped just for leaving a blank on its datasheet.
3. 03Because ingredient lists overlap across a portfolio, the search targets substitutions that resolve several products at once.

A shortlist where every candidate already clears the restricted-substance list for the market you are entering. Matterialis narrows the field; your regulatory function still owns the determination. Filtering by regional availability is in development.

Amphico makes PFAS-free waterproof membranes. Replacing the polyolefin compound in our own product with a bio-based alternative is constrained on three axes at once: it has to hold membrane performance, it has to be bio-based, and it has to be sourceable at volume by a company our size.

The useful candidates are not in the technical textile supply chain. They are in packaging and biomedical. Those suppliers have no reason to call us, and a membrane engineer has no reason to read their datasheets.

1. 01The index is organised by what materials are and do, not by which industry's suppliers sell them, so a packaging polymer surfaces for a membrane problem when it fits.
2. 02Candidates are compared on how they behave, which is what lets a grade sold into packaging be read against a membrane requirement at all.
3. 03Cross-sector candidates carry what the transfer involves: different regulatory status in your application, different grades and purities, different supply structure.

We built Matterialis because we needed it. This work is in progress and we will publish what we find, including what does not work.

An ingredient becomes unavailable, uneconomic, or unacceptable, and the replacement has to hold texture, stability and shelf life without changing the label more than it has to. Screening, before a single physical trial, takes weeks. It is reading a few hundred datasheets by hand.

1. 01Screening moves to the index. Property filtering, including on values that were never reported, narrows the candidate set without downloading anything.
2. 02The output separates properties we can predict from properties genuinely unknown. That second list is the lab list.
3. 03Candidates that enter trial have already cleared the property and compliance constraints, so the trials test real uncertainty.

A shortlist in days rather than weeks, and lab time spent on the questions data cannot settle. It does not replace the trial; it decides which trials are worth running.

Excipient selection runs through one person, because only one person can actually do it. The procedure is documentable: build a matrix of target properties, filter a few hundred grades to six or eight, generate data on what nobody reported, then design the trials.

What is not documentable is which grades enter the matrix at all. That rests on inferences made without writing them down, and reconstructed after the decision, if at all. So a junior scientist learns the procedure and builds a defensible-looking matrix from the wrong candidate set.

1. 01The inference layer becomes explicit: a predicted property is marked as predicted, so it can be inspected and disagreed with.
2. 02Constraints that used to be applied silently (qualification complexity, acceptable property trade-offs, prior familiarity with a supplier) become parameters you can review and reuse.
3. 03Every shortlist exports as a rationale: constraints applied, candidates rejected and why, values relied on and how confident, properties still to measure.

A shortlist a junior scientist can defend in review, and a specialist who audits the reasoning rather than redoing the work. The judgement stays with your expert; it stops being the throughput limit.