Every developer placed through sourceBOLD has cleared the same four-gate gauntlet, run end-to-end by human reviewers. The funnel is structured so each gate filters for a different signal — language and async clarity, technical depth, architectural judgment, and behavior under real-codebase load. Roughly 96 of every 100 applicants drop out before reaching a client brief.
Here's what each gate actually tests.
G1 — Communication assessment (30 min)
Spoken English fluency, written async clarity, and conflict-handling responses. sourceBOLD is a contractor-engagement platform — every placement integrates into a US engineering team's standups, code review, and async channels. G1 sets the floor: every developer who passes has been signed off by a US-based reviewer.
A developer who codes well but can't run a standup in English doesn't pass G1, no matter how strong G2 would be.
G2 — Technical screen (60 min)
Stack-specific exercises scored against a public rubric. Two reviewers, blind to candidate identity, must agree to pass. Disagreements escalate to a third reviewer; no single reviewer can wave a candidate through.
This is the gate that costs the most reviewer hours, and it's the one we won't compress. The blind double-review pattern is the antidote to single-reviewer bias — a candidate who reminds the reviewer of themselves, or doesn't.
G3 — Systems interview (90 min)
Architecture walk-through with a Staff+ engineer in the candidate's stack. Same set of questions, same rubric, every time — calibrated against developers we've already placed and seen working. A novel architecture question on each call would make the funnel impossible to compare across candidates. Rigorous repeatability is what makes the "4.2% acceptance" number meaningful.
G3 filters for senior judgment, not implementation speed. A candidate who can ship code fast but can't reason about partitioning, idempotency, or backpressure doesn't pass.
G4 — Live pairing (2 hours)
Real-codebase session. The candidate ships a small change against a benchmark repo while we watch the process — testing instincts, debugging behavior, communication under load. This is the gate that catches candidates who interview well but freeze when there's a stack trace on screen.
G4 is the most expensive gate to run (a Staff+ engineer sits with the candidate for two hours), so it's last in the funnel. We only run G4 on candidates who've already cleared G1–G3 — by then the probability of passing is high enough to justify the cost.
Why we publish the numbers
The 4.2% acceptance rate is meaningful only if you trust the funnel. We publish the gate structure, the rubric source (Staff+ engineers in the candidate's stack), and the blind double-review requirement so a buyer can audit our claim before signing an MSA.
Other platforms publish acceptance rates without explaining what the funnel selects for. The number on its own can be gamed. The four-gate structure is what makes ours load-bearing.
For the canonical walkthrough of how the gates fit into the engagement playbook, see How it works.