PLC tank controller in Structured Text

The agent writes a tank fill/drain controller in IEC 61131-3 Structured Text, compiles it with the open MATIEC compiler, links it into a 10 ms scan-cycle harness and runs it against a virtual plant. The judge's rules are numeric: the level never leaves its safe band, the pump never runs dry even when a level switch lies, an emergency stop de-energises every actuator within one scan and latches, and the plant recovers afterwards.

Results per scenario, the scenario-by-mutant matrix, real per-scan plant traces, the mutant kills and the final verdict. Made from real judge output, no illustrative data.

Measured

8 / 8

plant scenarios passed
all

88.02 cm

highest level in any scenario
≤ 94 cm design cap

Byte-identical

verdict JSON across two runs with the same seed
determinism

6/6

known-bad artifacts rejected
self-test, ≥ 90 % required

reproduced Independently reproduced from the README alone in a fresh sandbox. Numbers from the demo's README and receipt.

Ask your agent

With the sandbox plugin or any agent that can use a sandbox, paste:

In a sandbox, build the IEC 61131-3 Structured Text tank controller for `cookbook/plc-st` of https://github.com/agent-next/sandbox from scratch. Read `SPEC.md` first, then write your own `tank_ctrl.st` (program + configuration) using the pinned MATIEC toolchain from `judge/setup.sh`. Iterate with `bash judge/selftest.sh` and `bash judge/grade.sh your_tank_ctrl.st --json /tmp/v.json` until the judge passes every gate and your controller is not one of the six known-bad single-edit mutants. Show me the final verdict JSON and the key numbers (max/min level, kill rate). Do not modify anything under `judge/`.

Or run it yourself

Creates a sandbox, installs the pinned tools, runs the judge's self-test, grades the known-good example, prints the verdict and kills the sandbox.

pip install "git+https://github.com/agent-next/sandbox#subdirectory=cli"
export SBX_API_KEY=<your key>  # from the console
git clone --depth 1 https://github.com/agent-next/sandbox && cd sandbox
python3 cookbook/plc-st/run.py

No key? Run the judge on any Debian 12 box

# from cookbook/plc-st/
bash judge/setup.sh
bash judge/selftest.sh
bash judge/grade.sh examples/good/tank_ctrl.st --json verdict.json

Results

Highest and lowest tank level per scenario for the known-good controller, with the design cap, high switch and dry band. Full size

Judge self-test: which scenario catches each of the six mutants. Full size

Why a sandbox

The agent installs a pinned compiler, compiles its own code and iterates against the judge in an isolated, disposable machine; nothing it does touches your machine.

Measured wall time

Step	Wall time
`judge/setup.sh (fresh install of pinned MATIEC)`	~39 s
`judge/selftest.sh (good + 6 mutants)`	~33 s
`judge/grade.sh examples/good/tank_ctrl.st`	~5 s

Limits, plainly

An educational interlock exercise, not a certified safety system: no SIL rating, no redundant architecture, no formal proof.
The plant model is analytic and deterministic; faults are scripted stuck or false switch states and E-stop pulses.

Source

Demo folder · SPEC · RECEIPT · REPRODUCE — pinned at ba328e6

Run one yourself

Sign in with GitHub for a free key with 10 sandbox-hours. Every demo above finishes in minutes and kills its sandbox when it is done.

Get a free key Quickstart