F13LD.synth
Inverse design. Set a target on the property pad, and a model trained on the vault returns candidate recipes that should meet it.
What it does.
F13LD.synth runs design in reverse: instead of choosing parameters and measuring the result, you set target properties and the tool proposes recipes likely to match, using a model trained on the validated designs in the vault.
It is the inverse counterpart to sweep's forward search.
Where it fits.
Synth reads the vault as its training corpus and emits new recipes. A candidate can go straight to mesh, or re-enter the vault through ingest once it has been validated.
Worked example: a target.
Set a target
On the property pad, set the targets — stiffness, anisotropy, relative density.
Generate candidates
The trained model emits candidate recipes aimed at the target.
Inspect
Review the candidates and their predicted properties, and pick one.
Verify
A predicted property is not a measured one. Confirm by homogenizing in lab, or by meshing and checking.
Use it
Open the recipe in mesh to print, or send it through ingest to add to the corpus.
Predicted properties come from the model, not a simulation. Always verify a candidate before relying on its numbers.
Output.
Candidate recipes tagged with predicted properties. The predictions are model estimates trained on the vault; verify them in lab before relying on them.