Inference
Preview a manifest before you commit a cell.
Send up to 40 SKUs and get an estimated success table per family before a cell is committed. The preview runs on H100 GPUs inside Tenon's training infrastructure, not at the edge. The cell runtime does not change: it stays on Jetson AGX Thor and runs offline.
What it returns
An estimated success rate for each SKU family in the manifest you send, in the same per-family format as the gate report.
Limits
| SKUs per preview | Up to 40 |
|---|---|
| Where it runs | H100 in Tenon's training infrastructure |
| What it is not | Not the deployment gate. A policy still has to clear ≥88% held-out success, ≤3.5 s cycle and ≤45 N peak force before it ships. |
| Edge | Unchanged: Jetson AGX Thor, offline, 30 Hz |
Records
Each preview is tied to the manifest revision and container digest it ran on, like any training generation. See reproducibility.