fix(vectors): clear error for an out-of-range layer in normalize_alpha / dose

[dchaudhari7177]

Problem

Requesting a layer that is not in the vector raised a bare KeyError: 999, which doesn't tell the user which layers exist.

Fix

Guard at the top of normalize_alpha (which dose() also goes through):

ValueError: layer 999 not in steering vector; available layers: [3, 7]

Tests

• normalize_alpha(vec, layer=999, ...) and dose(vec, layer=999, ...) both raise the new ValueError naming the layer and listing available layers (2-layer vector)
• valid-layer behaviour pinned unchanged (norm-5 vector, alpha 2 → by_vector_norm 10)

python -m pytest tests/test_vectors.py passes.

Fixes #20

[bamdadd]

Thanks for this — the change itself is good. The out-of-range layer now raises a clear, typed ValueError that names the bad layer and lists the available ones, dose is covered through normalize_alpha, and the tests pin both. That's exactly what #20 asked for.

The only thing standing between this and a merge is formatting: CI's ruff format --check wants two files reformatted (src/steerbench/vectors.py and tests/test_vectors.py). Nothing wrong with the logic, it's just whitespace/wrapping the formatter is opinionated about. One step should turn CI green:

uv run ruff format .
git commit -am "style: ruff format"
git push

Push that and the checks should go green; I'll merge as soon as they do. Ping me if anything's unclear.

If steerbench is useful to you, a star helps others find it and Watch will keep you posted — glad to have you contributing.

[bamdadd]

Applied the one formatting step for you (ruff format on the two files) so you weren't blocked waiting on it — nothing else was needed. Your fix is solid: the out-of-range layer now raises a clear ValueError that names the bad layer and lists the available ones, dose is covered through normalize_alpha, and the tests pin both paths. Merged, and it closes #20.

Two clean contributions in a row now — really glad to have you here. If steerbench is useful to you, a star helps others find it and Watch keeps you posted on new issues; would love more contributions whenever you're up for it.