MEASURED The field is counted at build time from the 7 sections' own source rather than transcribed, so it cannot go stale when one of them gains a claim. Note what it grades: this proposal, not OpenRouter. It is the one measurement here whose subject is the document you are reading.

You are hiring for a growth engine and a community. They are the same funnel.

This is a Request for Job written as a Request for Proposal: a diagnosis of your public growth surfaces, taken from outside, with a plan attached. It is not a first-90-days plan — I don't have the job, and a plan that assumes it would be presumptuous.

The thesis

Your community can be activated on demand. Whether anything measures what happens next is not visible from outside, and no readout of the activation I could see ever surfaced. The instrument that would settle it — per-surface, cohort-level measurement of what people do after they arrive — is the same instrument the self-serve funnel needs at its top. That's why growth and developer relations aren't adjacent functions here. They're one funnel, measured twice or not at all. And the partnerships req you posted days later says the same thing in your own words: it names marketing as a counterpart it needs, this one never names that motion back, and the revenue partners source lands in the pipeline this role is already asked to measure. Three functions, one owner, one funnel definition.

INFERRED That the readout was available and never taken is reasoning on top of the counts in What I measured, and it is attackable there. MEASURED The two postings are four days apart, the channel vocabulary is absent from this one entirely, and marketing is one open req of twenty-three — all three counts sit in What I measured, and The role shape argues from them. UNKNOWN Whether your funnel already instruments activation at cohort granularity. The plan grades that unknown and the thesis above assumes nothing either way — from outside I can see that no readout surfaced, never that none exists.

Three things I found

MEASURED 200 coupons cleared in about a minute and produced 68% of a channel's lifetime messages in one day — then decayed to baseline in four. INFERRED The readout that would say whether that was acquisition or leakage was available and never taken.
INFERRED Your two support surfaces fail in opposite directions, so no single response-rate metric can see either one. A dashboard built on either alone reports this community as healthy.
MEASURED Per-model usage data already renders and isn't in your sitemap. INFERRED Real proof of what gets built on each model, already constructed, invisible to crawlers — a sitemap change plus a template decision, not a publishing programme.

Read in any order

What I measured Firsthand counts of OpenRouter's public surfaces and community channels, taken before proposing anything. Every claim carries its provenance.
The role shape Why two posted reqs and one unposted function describe a single job, and what it costs to split them.
The plan Seven pillars, each with what I'd do, how I'd know it worked, and what I don't know yet — including a $5M paid budget allocated on paper.
Context for the record Four conditions the proposal has to survive: the acquisition's actual status, what your community was anxious about, what your conference calendar shows, and the large public surface nobody is standing on.
Sequencing and week one Three horizons ordered by dependency rather than ambition, and the seven things I'd need in the first week before any of it is worth starting.
How I counted The sources, the method, and the limits — what the counts can support, what they cannot, and the choices I made where a count could have gone either way.
What I got wrong Hypotheses that died under test while this was being written. Listed because a diagnosis whose errors are hidden is worth less than one whose errors are visible.

How to read the claims

The section called What I got wrong lists hypotheses that died under test while this was being written. It is there on purpose.