Verify your install#

Nine checks. Each is one command with an unambiguous answer, and each tells you something different. If they all pass, the tool, your science stack and your workspace are sound.

1. The command exists and reports its version#

$ matmelt --version
matmelt 0.2.0

If this fails, the launcher is not on your PATH — see Installation.

2. Every command is present#

$ matmelt --help

You should see init, doctor, run, status, retry, results, monitor, cif and update.

3. The scaffolded config is valid as written#

$ mkdir /tmp/matmelt-check && cd /tmp/matmelt-check
$ matmelt init

init refuses to overwrite an existing config, so run it in an empty directory. It scaffolds and then immediately diagnoses; a scaffold that could not be loaded would fail here.

4. Backends are reachable#

In the doctor output, the Backend reachability table must show healthy for every backend. For the scaffolded local backend this means its work directory is writable. A remote backend being unreachable is reported here, not discovered halfway through a run.

5. Every stage’s dependencies are importable#

The Stage readiness table must show ok for each stage you intend to run. MISSING prints the exact install command underneath. Remember that the probe runs in that stage’s interpreter, on the machine that stage will use.

6. `doctor` exits 0#

$ matmelt doctor
$ echo $?
0

Exit code 0 means every check passed. Exit 1 means at least one problem was reported; exit 2 means the config itself could not be read.

7. The enumeration is the size you expect#

doctor prints the seed enumeration before any work is possible:

Seed enumeration: 2 element(s) -> 3 combination(s) (size 1: 2, size 2: 1).

This is the blast radius of your element set. It grows combinatorially — check it before starting a run you did not intend, especially after widening elements or combination_sizes.

8. A real job runs end to end#

The strongest check is a complete run, because it exercises the whole path: job submission, the state database, output collection and the stage contracts. Your first run is that check — if it produced a record.json, everything below the CLI works.

9. Read-only commands leave the workspace untouched#

$ matmelt status
$ matmelt results
$ matmelt monitor --once

All three open the state database read-only and cannot modify a run. monitor --once renders a single dashboard frame and exits, which is a safe way to confirm it works without holding a live view open.

If something fails#

Symptom	Where to look
`matmelt: command not found`	Installation — the `PATH` note
A stage says `MISSING`	The tool and the science stack — run the printed command
A backend is unhealthy	Check that a machine can do the work
The run starts but materials go to `FAILED`	Diagnose a failed material
A model download is refused or fails	The tool and the science stack — the MACE model section