00Live
TS/JS Core Logic
code-workbench · ~60m
Objective
Prove they can plan, implement, and verify TypeScript/JavaScript under fixed model and token caps—not just paste until it compiles.
AI policies don't tell you who can think.The Constraint does.
AI-in-bounds assessment. Same caps. Exportable evidence.
Role pack·Forward Deployed Engineer
[01]Sample run// What you review
See a candidate in the arena.
One module, fixed caps—then the evidence a hiring manager actually reads.
[02]Role pack// AI-native hiring
Forward Deployed Engineer
ai-forward-engineer·AI Native / FDE
TypeScript/JavaScript, React Product UI, agent-assisted delivery, and debugging under constraints—a full assessment for AI-native roles.
~2–3h total · sequential modules · same constraints per candidate
MODEL_LOCK · TOKEN_BUDGET · CONTEXT_CAP · GOALS_VERIFY
01Live
React Product UI
ui-workbench · ~55m
Objective
Ship a product UI with AI in bounds: streaming interfaces, agent controls, and recovery when the first pass fails.
02Planned
Agent workflow
web_skill
Objective
Orchestrate an agent-assisted workflow end to end—scope, tool use, and a verifiable outcome.
03Planned
Debugging
code-workbench
Objective
Diagnose and fix a failing system under the same constraints: isolate the fault, recover, and leave evidence.
Clone a pack → send candidate links → review traces and output. One bar per opening.
This is the Forward Deployed Engineer pack—ready to run. Pilots can use it as-is or map another role as the catalog grows.
[03]Data annotation// labeling · RLHF · eval
Throughput doesn't hire judgment.
Annotation companies hire at volume. Speed tests and rubric quizzes still miss who can think under model pressure—and leave a trail your QA leads can actually review.
Reviewers · AI trainers · eval ops · technical annotators
01
Same bar, every candidate
Fixed caps and the same arena for every hire—so you compare judgment, not who got a longer prompt or a better tool tip.
02
Evidence that maps to QA
Exportable traces: how they interpret instructions, recover from bad first passes, and whether the output would survive your review ladder.
03
Built for data-quality roles
Not another coding screen. Assess the people who decide label quality, preference ranking, and eval pass/fail—under real AI constraints.
Pilots scoped to your ladder · packs adapted to your domains.
Talk annotation pilot →[04]The pilot// How it runs
Live in days, not quarters.
Limited pilot slots. Scoped with your eng team.
- 01
Intro call & req scoping
- 02
Pick a pack (e.g. FDE) + adjust
- 03
Candidate links & trace review
Limited pilot slots · scoped in days, not quarters.
[05]FAQ// Before the pilot
Questions teams ask first.
Yes—that's the point. AI is in-bounds under fixed constraints. You measure how they plan, prompt, recover, and ship—not whether they hid a tab.