There’s a toggle on the dashboard called “Exclude Employees.” The first time I flipped it, I watched an ecosystem lose more than half of its apparent activity. The community was smaller than the headline number suggested: a company with a Discord server and a very committed engineering team. Nothing shameful in that, but it is a different operating fact and asks for different work.
That toggle sits on a dashboard I built because a friend took over a Developer Relations team and inherited GitHub stars, Discord growth, registration counts, and conference photos. The question underneath those measures was simple: which people reached by the programme went on to write code? Two line items kept meeting in the same budget conversation, a $50K hackathon and a $5K meetup series, and each was defended by atmosphere. The hackathon had conference photos and a sponsor recap. The meetup series had registration counts and Discord growth. Both had enough anecdotes for a quarterly review, but nothing could rank them despite a tenfold cost difference. Neither could answer whether anyone shipped code afterward.
I recognised the pile because I’ve assembled versions of it myself. The industry repeats a familiar pattern: fly somewhere photogenic, give a talk, collect business cards, post the full room to LinkedIn, call the room a signal, and move to the next city. I call it developer tourism, because the team appears everywhere without connecting the visit to a developer’s next action. Conferences can work, and some do. The failure is leaving the event unable to say whether anyone built anything afterward. The Goldman/Docker test explains the budget case, including which adoption stories to fund and when sales should enter an account. This post describes the instrument.
I built a dashboard tracking Sui, Walrus, and Sui Name Service around my friend’s question. I can’t link the dashboard itself because it contains attendee CSVs, email addresses, and lists of people whose activity is declining. None of that belongs on the public internet. The method is all here.
The pressure to answer that question now spans the industry. In the State of Developer Relations 2024, proving business impact rose from seventh to second among challenges named by practitioners, at 28.5% of respondents. Another 61% said they found it difficult to prove their influence. As DevRel becomes a settled function, it receives settled-function questions: where did the money go, which decision did it improve, and what should we repeat? Mary Thengvall has warned about vanity metrics for years. She describes them as “items that are easy to track but don’t often speak to the larger business value of the work you’re doing.” GitHub stars can be bought or botted, Discord membership says little about intent, and event attendance proves only that someone entered a room.
The floor under the numbers
Electric Capital gave the conversation a useful baseline. Its 2024 Developer Report analysed 902 million public commits across 1.7 million repositories, mapped developers to ecosystems, and counted who was building where. The report ships as a slide deck; ChainCatcher’s write-up provides checkable full text. Its units are ordinary: a monthly active developer committed code within a rolling 28-day window, full-time means ten or more commit days a month, and tenure measures time since the first commit. Each count begins with a verifiable event, a person pushing code to a public repository, rather than a survey answer or signup form. The dashboard uses those units unchanged.
Resolving attendees to builders
The central join asks whether a person wrote code after attending an event. Each event becomes a funnel with total participants, participants resolved to GitHub identities, new developers, and developers still active after 7, 14, 28, and 60 days.
“Resolved to GitHub” is the ugly step. Hackathon CSVs arrive with inconsistent column names and email addresses that may not match anything. The resolver scans every column for GitHub URLs, extracts usernames from repository links that no longer resolve, matches emails against developer profiles, and checks repository contributor lists as a last resort. Elegance could wait. I first needed to know who showed up and whether they stayed active.
Activation rate is the number the funnel exists to produce: of the new developers who attended, what percentage were still committing code 28 days later? Sixty?
Event attribution here is temporal matching, a window from 14 days before the event to 28 days after. It measures proximity, not causation. A developer who was going to build anyway gets counted, and a talk that plants an idea which sprouts in month three gets missed. I trust the comparisons between events far more than any absolute claim about a single one.
The signals that force decisions
Churn scoring made the dashboard personal. I built it from three inputs: time since the last commit, whether activity is falling or holding steady, and whether commits are becoming smaller and less frequent. The output is a list of names whose activity is about to stop. That makes “community health” uncomfortable and specific. A developer with 100 commits who is losing steam already knows the platform and has climbed its learning curve. Send that person an email or DM and ask what they’re working on before spending another dollar to acquire someone who has never built here.
Retention cohorts show whether onboarding is improving. Compare one month’s 30-day result with the next. A rise means some intervention or fix worked; a fall means something broke. The dashboard colours retention of 50% or more green, 30% to 49% yellow, and anything lower red. A red column sends attention to the first-run experience before another event budget is approved.
The split that surprised me was exclusive versus multichain. Exclusive developers contribute only to your ecosystem; multichain developers hedge across several. A high exclusive ratio suggests you’ve built something developers commit to, literally and figuratively, while a high multichain ratio means you’re attracting experienced people who are still choosing. Neither is better, but they call for different work: deepen the exclusive builders, or court the experienced ones while they’re deciding.
Packages add another stage in Web3. A developer writes code and deploys a package, which either generates on-chain transactions or sits idle. I linked around 278,000 Move packages in the dataset to their GitHub repositories, developers, and transaction counts. The question became: which participants deployed packages that people actually use?
The map is no kinder
“Exclude Employees” wasn’t the only toggle with news in it. “Global community” sounds generous until the map shows where commits come from. Sometimes the next dollar should move away from San Francisco and toward India, which rose from tenth to the second-largest source of crypto developers and produced 17% of the field’s new developers in 2024. Another map may show a community concentrated in three cities. In that case, thin coverage should not be presented as a global strategy.
The working-hour heatmap killed another story I liked: most developers commit during business hours in developer-heavy time zones. Launches, announcements, support coverage, and office hours can be scheduled around when developers actually commit.
What the numbers changed
Back to the hackathon and the meetup series. Once their attendee lists resolved to GitHub identities, I could count new developers acquired per event, activation rate per event, and cost per activated developer, and watch which attendees were still committing 28 and 60 days later. I can’t publish the per-event percentages; the ranking is what mattered, and the ranking is what changed the budget conversation. It followed retention.
Breaks also show up in the funnel. Signups without activation point to onboarding; activation without 28-day retention shows a problem after the first success. The churn list turns “community health” into a task: write to the developer with 50 commits whose activity is slowing before sending a generic newsletter to 10,000 people who may never build anything.
I still believe in presence: handshakes, whiteboard sessions, hallway conversations that produce sharper questions than the agenda did. The dashboard just tells me which rooms to go back to.
How the numbers were built
The data comes from Electric Capital’s Developer Report, enriched with GitHub profiles, Sui package analytics, and custom retention modelling. It tracks Sui, Walrus, and Sui Name Service with daily updates. Monthly active developers use the rolling 28-day commit window. Retention cohorts are monthly and checked at 30, 60, and 90 days. Event attribution uses the temporal match described above, from 14 days before to 28 days after an event. Package linking combines MVR and Blockberry data with confidence scoring. The dashboard was built for Sui, but the method applies anywhere developers ship code.
If you run a developer programme, the join is the place to start. The percentage of your last event’s attendee list that resolves to GitHub identities is your first number; how many of those people were still committing 28 days later is your second. Everything else here grew out of those two.