Markdown | JSON with evidence | Recipe contract
# Use positive state costs for this bounded LQR example
This adapter requires positive state costs and checks the discrete linear closed-loop eigenvalues. This is a linear model check, not nonlinear or hardware stabilization.
Runtime: `pythonrobotics`. Recipe: `pythonrobotics-pendulum-lqr`. Source revision: `cdd0cc888802b584c2d654ca85e3c5460973487d`.
[Find applicable prepared recipes](https://api.superfastrobotics.com/v1/search?q=pendulum+lqr&runtime=pythonrobotics&kind=preparation_recipe&limit=3). Follow the returned pinned `inspect_url` for the exact interface, evidence and prior findings. [Preparation and execution](https://hub.superfastrobotics.com/executable-skills.md) remain separate local operations. Source files retain their original license and authors.
## Reproduction input
Mode: `input`. Reproduction recipe: `pythonrobotics-pendulum-lqr`.
```json
{
"state": [
0,
0,
0.1,
0
],
"state_cost_diagonal": [
0.0,
1.0,
10.0,
1.0
],
"force_cost": 1.0
}
```
## Valid adapter input
```json
{
"state": [
0,
0,
0.1,
0
],
"state_cost_diagonal": [
1.0,
1.0,
10.0,
1.0
],
"force_cost": 1.0
}
```
Use the `tools/verify_problem_cases.py` harness at the published Hub commit with the checksum-verified source cache. `input` cases require rejection of the first input and success for the valid input. `upstream_interpolation` separately demonstrates the original source function's wrong point; `handoff` tests contract compatibility only. Neither mode claims physical execution.
The [source and receipts](https://github.com/Superfast-Robotics/robot-skill-hub/tree/main/evidence/integration/agent-discovery-v1) distinguish reproduced failures from successful bounded checks. A useful new observation can be saved locally with the scoped Hub CLI; public sharing requires the project's existing permission.
Authored reproduction; not autonomous agent discovery, general accuracy or hardware verification.