QEP-5: Code of AI Use (Draft)

[mmcky]

Adds QEP-5: Code of AI Use as a Draft.

AI-assisted contributions are welcome at QuantEcon — our maintainers use these tools daily and in the open. This QEP asks every contributor to accept six norms in exchange: a human submits, meaningful AI involvement is disclosed with a machine-readable marker, the contributor owns and can explain the work, claims about tests and benchmarks are verified before they are made, generated pull requests are not mass-submitted or duplicated, and a lapse is put right by amending the pull request.

It is policy, not infrastructure. The norms bind from acceptance, like the Code of Conduct, without anyone signing anything and without anything being built.

Scope: narrowed from the original discussion

The discussion issue originally proposed a contributor registry and a pull-request gate alongside the policy. Both are deliberately out of scope here, for two reasons set out in What this is not and Alternatives considered: the norms bind whether or not anyone attests to them, so a registry would notify rather than bind; and GitHub shipped native volume controls in 2026 — org-level pull-request limits in particular — that cover the single-account case for free. A gate would also put a bot comment on a newcomer's first contribution, which is the worst possible moment to introduce friction.

If a registry proves necessary later, it is an amendment to this QEP or a QEP of its own covering the maintenance and procedures a registry actually needs. Two constraints from the design work are recorded here so they carry over. The reasoning behind the narrowing is on the discussion issue.

Points worth arguing about

Norm 3 lets tests stand in for reading every line, in code only. The burden moves rather than disappearing — the tests become what you must have read and be able to explain — and there is no equivalent for prose, mathematics, or exposition, where no test tells you an argument is wrong.

Norm 2 is a marker, not a narrative. A structured AI-Usage: free-text line was considered and rejected as cost without signal; the reasoning and the reversal condition are in Alternatives considered.

Organisation-operated automation is carved out explicitly, since norm 1 as first drafted would have prohibited our own dependency bots and link checkers — which QEP-2 already assumes exist. The carve-out applies the accountability rule rather than waiving it: a named maintainer owns the output, the automation is declared publicly, and it carries the automated label.

Process

Opened as a draft. The one-to-two-week comment window under QEP-1 starts when this is marked ready for review; the proposed decision deadline is 2026-08-28, adjustable before then.

qep-checks parity: new QEP, so unversioned (implicitly v0), with the README index row added as standard / Draft / –.