Throughput proxies that survive scrutiny

Recruiters are right to be skeptical of throughput claims. "10× engineer" language is meaningless. What helps is a repeatable measurement with explicit caveats — and the discipline to omit figures that cannot survive a staff interview.

Prefer proxies you can re-run

A defensible activity proxy usually has three properties:

1. Reproducible — another person can pull the same window with the same tool and land near the same number.
2. Scoped — personal vs team, public vs private, commits vs reviews are not mixed silently.
3. Caveated — the write-up says what the number is not (business outcomes, quality, impact).

GitHub's trailing-twelve-month contribution graph is one such proxy: commits, pull requests, and reviews as the platform counts them. It is useful for sustained engineering activity. It is useless as a claim about revenue, customer outcomes, or "AI 10×."

A method worth copying

1. Fix a trailing 12-month window and record the measurement date.
2. State which surfaces are included (for example: personal public + private work you control) and which are excluded.
3. Publish the method next to the figure — not in a footnote nobody reads.
4. Refuse to blend contract delivery counts, open-source stars, and agent run counts into one vanity index.

Caveats that should stay visible

• Contribution counts are a proxy for activity, not a productivity religion.
• Agent-assisted commits still count as delivery work — the honest claim is orchestrated velocity, not typing speed.
• Volume without gates is noise: lint, types, tests, and disclosure review matter more than the raw number.

If a metric cannot survive that level of scrutiny, it should not appear on a senior résumé or public profile.