[kevinjqliu]

Rationale for this change

Adds an AGENTS.md testing guideline to prefer a for loop in a single test over @pytest.mark.parametrize. Each parametrized case is a separate pytest item with its own collection, setup, and reporting overhead, which adds up across the suite.

Measured locally with the project's tests/conftest.py (pytest -m unit -q, trivial bucket-transform assertion as the test body):

Cases	parametrize	for loop
1,000	~0.30s	~0.01s
10,000	~2.95s	~0.01s

Roughly 0.3ms of overhead per case. About 2,258 of the 4,022 collected unit tests (56%) are parametrized cases.

Are these changes tested?

Documentation-only change. prek hooks (markdownlint, codespell, end-of-file, trailing whitespace) passed on commit.

Are there any user-facing changes?

No.

[kevinjqliu]

I noticed that a lot of new PRs tend to use @pytest.mark.parametrize when a for loop can suffice (my own generated PRs included 😄 )

Hopefully this nudge all the agents to do the right thing

[Copilot]

🟢 Approval recommended

The documentation-only change matches the stated intent and has no unresolved review findings.

No open issues 🎉

What changed in this PR

This documentation-only PR updates contributor testing guidance to reduce pytest overhead without changing runtime behavior.

Changes:

• Prefer loops within a single test over parametrization.
• Require input values in assertion messages to identify failing cases.
• Reserve parametrization for cases needing different fixtures.

File	Description
AGENTS.md	Adds guidance on loops, assertion messages, and when to use parametrization.

_{Review effort: Balanced}

[rambleraptor]

Can we add something about using multiple tests if the parameters are sufficiently complicated?

I've seen a number of tests that parameterize on 3-4+ very complicated fixtures. Even a for loop makes reading them a mess. The loss in readability doesn't make up for the few lines of code savings.