Product

Selector — screening and ranking, with the evidence visible

Selector takes constraints and an objective, screens candidates against the constraints, and ranks the survivors. What it adds to a classical selection tool is a column: every value shows where it came from and how far it can be defended, and any index or margin it computes carries the evidence level of its weakest input rather than the confidence of the arithmetic.

Start free · What data sits behind it


What it does

Screening is a gate, not a score. A candidate clears a constraint or it does not. Constraints are recorded as a property, an operator, a value, a unit and a condition — if a requirement cannot be written that way, it goes into the record as a qualitative requirement instead of being quietly converted into a number.

Ranking uses material indices. E/ρ, E^(1/2)/ρ, σ^(2/3)/ρ and the rest, with the index and its assumed free variable stated rather than implied. Why the exponents differ.

Derived quantities inherit. An index computed from a specified minimum and a predicted density is a predicted index. Selector labels it as one. This is the weakest-link rule, and it is enforced rather than suggested.

Candidates that do not survive are kept. With the reason. A shortlist without its rejections is a conclusion, not an analysis, and the first question a reviewer asks is what you ruled out.

Ties are reported as ties. Where two candidates are separated by less than the width their inputs carry, Selector says they are not separated instead of presenting an order the data cannot support.


What it is not

It is not Granta. Granta Selector is a mature environment with a curated engineering dataset, decades of work behind it and a training requirement to match. If you have it and it does what you need, keep it. We are faster in the phase before you know what you are looking for, we cite open computational sources, and we are honest about coverage.

It is not complete. Our data layer is strong on inorganic crystalline materials and absent on anything formulated. A selection problem in polymers, composites or coatings is one we cannot serve today. The coverage picture, written before you pay rather than after.

It does not decide. It shortens the list and shows the working. The choice, and the professional responsibility attached to it, stay with you.


Where it sits relative to the agent

The research agent is for the open-ended phase: a question in plain language, an unbounded candidate space, a report. Selector is for the phase after that, when you know the constraints and want them applied consistently.

Most real work moves between the two. Both produce the same artefact — a record you can put into design documentation.


What this page does not cover

  • Coverage by material class. Uneven, and documented separately.
  • Property charts. Ashby-style log–log charts are on the roadmap, not shipped.
  • Custom or in-house datasets. Not supported. If your organisation has its own approved materials list, Selector cannot read it yet.
  • Multi-objective optimisation. One objective at a time; a trade-off surface is a judgement, and collapsing it into a weighted score hides that judgement inside a number.

Published
Last verified

Evidence levels on this page follow the TRACE E-ladder. Values are reproduced from the sources named beside them; where a source could not be resolved, the page says so instead of filling the gap.

Check the citations on a question you already know

The free plan gives you three complete reports and the whole knowledge base, with no card. If the sources do not hold up, you will know in the first week.