Scientific Computing & Numerical Methods
PDE solvers, inverse problems, sensor fusion, and ill-conditioned numerical pipelines verified against physical invariants.
Physics · Applied Math · SimulationQualify in this domainCrossing Hurdles · Benchmark Research
The next frontier needs human-crafted tasks.
Turn deep domain engineering into reproducible Terminal-Bench 4.0 environments that measure what autonomous agents can genuinely solve and where they break down.
Schema 1.4 · Non-Root IsolationLive PR & Stage Tag SyncLocal File Upload + Stripe Express
CROSSING HURDLES · PR LIFECYCLE
- Qualify & claim your domain01 / NEW TASK
Pass the human-authored TB 4.0 studio with local file upload, complete 7-step onboarding, and claim an open task.
- Build reproducible environments02 / BUILD
Author instruction.md, non-root Linux containers, clean inputs, and an independent reference solver.
- Automated PR tags & Stage 15 gate03 / IN REVIEW
Every pushed SHA updates CI: 14/14 Passed and Abomination: Passed tags live across GitHub and reviewer queues.
- Super Reviewer merge & payout04 / DELIVERED
Ready to Deliver tasks sync in real time, squash-merge cleanly to main, and settle via Stripe Express.
Every verdict and payout binds to an immutable Git commit SHA
14 + 1 Verification GatesSequential CI checks followed by Reviewer Abomination replay.
Zero-Leakage IsolationSeparate no-network containers transfer only declared /app artifacts.
Source-Bound EvidenceOracle, NOP, and frontier pass@k trials lock to your exact SHA.
Automated Stripe PayoutsDirect settlement triggered automatically on Super Reviewer delivery.
Domain tracks
Bring a problem where shallow heuristics fail.
Projects seek realistic multi-step engineering tasks with unambiguous specifications, clean input artifacts, and programmatic verification free of hidden generator coupling.
Systems, Infrastructure & Security
Kernel diagnostics, distributed recovery, binary forensics, and hardened sandboxes inside reproducible Linux containers.
Linux · Networking · AppSecQualify in this domainData Engineering & Knowledge Graphs
Schema reconciliation, streaming entity resolution, temporal joins, and corrupt-log recovery with exact output contracts.
ETL · SQL · Graph AnalyticsQualify in this domainHardware, CAD & Manufacturing
Geometric tolerance verification, netlist synthesis, toolpath scheduling, and physical DRC constraints.
EDA · Geometry · CAMQualify in this domainOperations Research & Supply Chain
Multi-echelon routing, crew and fleet recovery, combinatorial dispatch, and stochastic capacity allocation.
MILP · Scheduling · OptimizationQualify in this domainAdversarial Verifier & QA Design
Anti-tamper grading harnesses, isolated no-network evaluation, and mutation-tested check suites resistant to shortcuts.
AST Rigor · Witness ReplayQualify in this domainEnd-to-end workflow
From live qualification to delivered benchmark.
One connected workspace links your onboarding compliance, task claim, GitHub Pull Request, 14-stage CI card, review feedback, and payout ledger.
- 01
Qualify via Live Studio
Complete the human-written Terminal-Bench 4.0 assessment by authoring prose in the studio and uploading local workspace files or folders.
- 02
Onboard & Claim a Task
Finish the 7-step identity, agreements, Substitute W-8BEN, and Stripe Express setup, then claim an open domain slot in your cohort.
- 03
Push & Sync PR Stage Tags
Develop on your provisioned GitHub branch, calibrate Oracle = 1.0 and NOP = 0.0 locally, and earn CI: 14/14 Passed automatically on PR.
- 04
Abomination & Live Delivery
Reviewers run Stage 15 Abomination and promote Ready to Deliver in real time; Super Reviewers squash-merge to main and release payout.
Ready to contribute
Start the live assessment or inspect the published standard.
Passing the onboarding assessment immediately provisions your Crossing Hurdles expert account and unlocks cohort task claims, GitHub repository provisioning, and weekly Stripe payouts.