Plaza by use case

Find who you need, for what you’re building.

Pick the situation you are actually in — a co-founder, a first contributor, a teammate for the weekend — and the guide answers it in those terms. Plaza ranks builders on evidence measured from their public repositories, not on claims. Every match cites the repositories it read and shows the arithmetic behind the ranking.

The numbers

What does Plaza actually measure about a collaborator?

Only public GitHub evidence, read through the REST API: languages counted in bytes, when each repo was last pushed, repos owned, followers and stars. Nothing is self-reported and no model invents a number — these are constants in the code.

Skill strength
50% recency · 30% measured volume · 20% breadthThe three terms behind one skill's strength (app/lib/match.ts).
Self-declared topics
kept at 60% strengthA repo topic the owner typed is weaker evidence than bytes we counted, so it is discounted rather than trusted.
Final ranking
85% skill coverage + 15% Trust ScoreCoverage dominates on purpose: the question is whether they can do this, not whether they are famous.
Trust Score, out of 100
code depth 28% · repos owned 18% · recency 15% · cadence 12% · account age 12% · stars 8% · followers 7%Seven components (app/lib/github-evidence.ts). Followers and stars — the two numbers you can accumulate without writing code — are worth 15% together.
Recency half-life
60 daysNot a yes/no 'active' badge: someone who pushed yesterday reads differently from someone who pushed on day 89.
Cadence window
90 days, saturating at 30 reposHow much ground an account covered recently, counted from push dates.
Repos-owned band
1 repo to 100 reposLog-scaled between a floor that says nothing and a knee only the top of the real population reaches.
Account-age band
6 months to 15 yearsMeasured from the GitHub account's own creation date, never self-reported.
Language bytes read
the 8 repos pushed most recentlyGitHub allows 60 unauthenticated REST requests in 1 hour, so byte-level detail is spent on the freshest work first.
Cost to search
US$0, no accountSigning in with GitHub is also free; it indexes your own public activity so others can find you.
Contact limit
5 outgoing requests per 24 hoursA per-person cap, so nobody can be volume-mailed through Plaza.
Sign-in session
30 daysA signed cookie, refreshed on sign-in. Plaza stores no password of yours at all.
Comparison

How do you check that a developer really wrote what they claim?

Five things people actually do, and what each one can and cannot establish. Reading a CV proves nothing checkable; an interview measures interview performance; public code proves authorship but only of public code.

Ways to verify a developer’s claimed skill, and what each one costs
MethodWhat it actually provesTypical timeCostWhere it fails
A CV or résuméNothing checkable — every line is written by the candidate1–2 minutes to readUS$0No claim on it points at an artifact you can open
Reading their GitHub by handAuthorship of public repos, if you open each one10–20 minutes per personUS$0Does not scale past a shortlist; forks and starred repos read like their own work
A technical interviewHow they solve one fresh problem while watched45–60 minutes, two people1–2 hours of engineering timeMeasures interview performance, not what they have shipped
A take-home exerciseNew code they wrote for your prompt4–8 hours of theirs, 30 minutes of yoursUnpaid work, or a real feeHigh dropout, and it still says nothing about their past work
Plaza's measurementBytes of each language in their public repos, push dates and stars — with every repo citedUnder 1 minuteUS$0Blind to private and company-internal work, which is most work

Only the last row is measured by Plaza. The times and costs on the other four are typical industry practice, stated as ranges — we did not measure them, and the table says so rather than dressing an estimate up as a finding.

Primary sources

Where can you verify all of this yourself?

Every input Plaza reads is a documented, public GitHub endpoint, and every figure above is a constant in open code. These are the first-party specs — not commentary about them.