# GPU9D
> Free public-benefit research and testing foundation connecting compute providers and AI developers. No real provider connections, active tests or verified adoption yet.

## Implemented
- [Approved service roadmap](/services.md) and [machine-readable catalog](/services.json): eight future services with acceptance requirements; none is operational. Independent testing, task compatibility trials and capacity confirmation are first-pilot priorities.
- [Experimental NVIDIA inventory CLI](/compute9-nvidia-inventory.md): one local read-only vendor query after signed consent, separate signed machine/device binding and owner confirmation. Software tested; no real GPU validation, workload execution or verified spare capacity.
- [Offline consent preflight](/compute9-preflight.md): signed resource scope, time/VRAM/power/utilization budgets and publication whitelist checks with a separate key pin. No hardware execution, ownership verification, spending, revocation or replay protection.
- [Consent JSON Schema](/schemas/compute9-consent-v0.1.schema.json) and [test-plan JSON Schema](/schemas/compute9-test-plan-v0.1.schema.json): local files only, never upload these documents.
- [Offline evidence CLI](/compute9-evidence.md): simulation-only hash-chain and Ed25519 verification. No hardware access or GPU benchmark.
- [CLI download manifest](/downloads/manifest.json): version and SHA-256.
- [Evidence JSON Schema](/schemas/compute9-evidence-v0.1.schema.json)
- [24-hour research plan](/research-plan.md): planned work, not completed results.
- [Capability manifest](/capabilities.json)
- [OpenAPI contract](/openapi.json)
- [Service status](/api/v1/status): GET only.
- [Capacity directory](/api/v1/capacity): GET only, returns an empty directory. Query filters are rejected.

## Boundaries
Do not submit prompts, training data, weights, outputs, files, credentials or commands.
GPU9D does not execute customer workloads.
Matching, provider onboarding, evidence plugin releases, challenges, MCP, A2A and settlement are unavailable.
9% idle recovery and 99% valid matching are long-term targets, not achieved results.
Platform services are free for both sides: no service fees, commissions or payment collection.
Free platform access does not promise free GPU resources. Provider authorization is required for resource use and active tests.
No external independent AI use, supplier assessment or transaction is claimed.
The hosting provider may process connection metadata and access logs.

Supply research is limited to professional datacenters. Personal computers and home GPU networks are excluded. Public hardware specifications do not establish authorized spare capacity.