How to use the stand

You describe an engineering question in plain words; Muse Spark turns it — together with any photo, datasheet or CAD file you give it — into a precise problem; the ZTL stand proves the answer, or proves it wrong, or says honestly that it cannot vouch for it.

In five steps

1. Pick a task from the list on the left. The card under it says what to tell the AI and what to give it. Use this example fills in a ready question.
2. Add files if you have them: a photo of a nameplate or a label, a datasheet PDF, or your CAD export (see below). Drag them onto the box or click it.
3. Ask in the conversation: what must the design do? Muse Spark answers with what it read and what it is checking. If something is missing, it asks.
4. Watch the stand work. The bar shows the stand's own count of designs decided; when there is nothing to count (one fixed design being proved), it shows the phase and the time.
5. Check the values. The table on the right lists every number the answer rests on and where it came from — AI read from your words or files, or yours. Change any value (a range as 0 to 60) and press Recalculate: the stand judges again with your numbers, and the run is added to the list.

What the verdicts mean

Verdict	Meaning
EARNED	The design meets every requirement for every value the tolerances allow — proved, not sampled.
REFUTED	The design fails at a concrete point, shown under Where it fails. When several requirements cannot hold together, the stand names the smallest sets of them that conflict — change one set as a whole.
OPEN	The stand could not decide either way within its limits. It does not guess. (Before showing OPEN, Muse Spark rewrites the problem once — same numbers — in a form the stand can decide better.)

Why it says where each value came from

A proof is only as good as its data. The stand proves the design for the numbers it is given; a digit the model misread on a nameplate is exactly what no proof downstream can catch. So every value is labelled by its source, and the verdict notes how many came from the model's reading. This is the zero-trust idea of ZTL, the logic under the stand: a value that is not ours is not counted as known until someone who can check it does.

Files the stand reads

File	What happens
Photo (JPG, PNG, WebP)	Muse Spark reads the numbers on it — nameplates, labels, tables.
PDF	Its first pages are read as images: datasheets, manuals, catalogue pages.
SPICE netlist (.cir) + BOM (.csv)	Exported by KiCad, Altium, OrCAD and others. The stand imports the circuit itself; the BOM's tolerance column sets each part's spread. You only say what must hold. DC, linear elements (R, V, I, E, G, F, H; L and C at DC).
ANSYS truss deck (.txt)	A plane truss (N, E, MP EX, R, D, F lines). Say the modulus tolerance and the deflection limit.
Brick (.json)	A problem already in the stand's own format — sent as it is.

Not read: mechanical CAD geometry (STEP, DXF, IGES). The stand proves equations and networks, not shapes.

Cost

Each question costs a few cents of Meta Model API usage. The app keeps its own running total and stops itself before a hard cap — shown at the top right.

What is behind it

Muse Spark 1.3 (Meta Model API) reads and models. The proofs come from ztl-stand on ZTL — Zero-Trust Logic, both by Vitaly Reznik.