Frontier Post-Training Specialists: Staff SWEs, Competitive Coders, STEM PhDs & Proof Engineers
Pushing model capabilities past frontier reasoning cliffs requires authentic human domain mastery. Verified post-training labs in our registry recruit credentialed engineers, researchers, and formal verification specialists for proctored, remote SFT, DPO, PRM, and RLVR workflows.
Active Specialist Credential Cohorts & Compensation Bands
Credentialed subject-matter experts are matched to vetted independent AI labs operating proctored SFT, preference ranking, evaluation holdout, and domain audit cohorts.
| Credential Tier | Technical Scope | Typical Compensation Band |
|---|---|---|
| Cohort 01 // Staff & Principal Software Engineers (Repo-SWE & Systems Infra) | Authoring multi-file bug-localization trajectories, Dockerized `FAIL_TO_PASS` unit-test harnesses, distributed systems architecture SFT, and code-review DPO pairs across Python, Rust, C++, Go, and TypeScript. | $110–$195/hr USD |
| Cohort 02 // ICPC World Finalists, IOI Medalists & Codeforces Grandmasters | Constructing hard-horizon algorithmic reasoning prompts (`pass@32 < 0.15`), adversarial edge-case test generators, complexity-optimal C++/Rust reference solutions, and competitive coding RLVR problems. | $105–$210/hr USD |
| Cohort 03 // Mathematics, Theoretical Physics & Quantitative STEM PhDs | Step-by-step graduate derivation authoring, first-error Process Reward Model (PRM) step annotation, SymPy/numeric oracle synthesis, and CBRN/scientific uplift boundary evaluation. | $90–$175/hr USD |
| Cohort 04 // Lean 4, Coq & Isabelle Formal Proof Engineers | Autoformalization of Olympiad and research mathematics into `Mathlib4`, tactic-state trajectory curation, kernel-verified proof refactoring, and formal specification authoring. | $125–$220/hr USD |
Live Proctored Work-Sample Auditions and Zero-LLM Integrity
Every specialist matched to Tier-1 post-training nodes completes a paid or time-boxed live technical audition inside an isolated browser IDE or terminal environment: resolving a real multi-file repository regression, diagnosing the exact invalid step inside a subtle 12-line mathematical proof, or closing a Lean 4 theorem goal without `sorry`.
Production queues enforce strict zero-external-LLM rules backed by Shadow-DOM keystroke telemetry. Authentic human trial-and-error, intermediate hypothesis revision, and terse domain intuition are precisely the high-entropy signals frontier labs pay specialist rates to capture.
Clean Work-for-Hire IP Scope and Scoped Session Telemetry
Contributors execute clear work-for-hire intellectual property assignments scoped strictly to tasks authored inside the annotation workspace, protecting employers' unrelated proprietary codebases. Session keystroke and focus telemetry is restricted exclusively to workspace tabs or dedicated VDI containers—never monitoring personal host filesystems outside the active task window.