Roadmap

Concrete things donations unlock. Costs are honest estimates, not aspirational targets. Smaller donations stack toward the same goals.

• GPU rental for the next benchmark run

  $200

  Roughly 40–60 hours on an A100 via a cloud provider, enough to re-run the full evaluation harness against a frontier model and add two new adversarial categories.

• Developer laptop

  $1,500

  A reliable machine for day-to-day work. The current laptop throttles on long inference runs and has failed twice during evaluations.

• Sustained cloud subscription

  $600 / year

  Keeps the evaluation API access and storage running without interruption. Covers 12 months of moderate usage.

• Desktop workstation

  $3,000

  A local machine with enough VRAM to run 13B-class models for the constitution layer offline. Removes a recurring per-run cost and unblocks longer experiments.