# GPU9D approved service roadmap Platform services remain free for providers and demand-side participants. Free platform access does not include free GPUs. Supply research covers professional datacenters only, excluding personal computers. GPU9D does not receive or host customer prompts, datasets, weights, outputs or workload files. ## Current status These services have been accepted into the development scope; none is operational yet. Available software consists of a simulation/integrity evidence CLI, an offline signed-consent preflight CLI, and an experimental owner-authorized read-only NVIDIA inventory CLI. The inventory CLI has software tests but no real-GPU validation. These tools do not establish independent performance testing, supplier ownership, usable spare capacity or real task delivery. First pilot: independent testing + task compatibility trials + effective capacity confirmation. The empty production directory remains empty until real authorization and resource confirmation exist. Machine-readable roadmap: https://gpu9d.com/services.json Schema: https://gpu9d.com/schemas/services-v0.1.schema.json Implemented software and interfaces: https://gpu9d.com/capabilities.json ## Planned services and acceptance ### Independent compute testing Status: research_in_progress; unavailable. Priority 1. Repeatable measurements generated through an independent, authorized test process rather than relying only on supplier claims. - Public reproducible method and environment/version disclosure - Owner-approved challenge, resource limits and zero-cost scope - Independently measured correctness, timings, failures and uncertainty - Provider self-tests, platform challenges and buyer confirmations labelled separately ### Task compatibility trials Status: planned; unavailable. Priority 1. Check a declared task profile against framework, memory, region, deadline and interruption constraints before selecting capacity. - Complete buyer constraints without workload content - Authorized public or synthetic trial profile - Deterministic hard-constraint checks and explicit unknowns - Real trial outcome, not a GPU-model-only recommendation ### Effective capacity confirmation Status: planned; unavailable. Priority 1. Confirm that a specific small allocation and time window can actually be offered by its authorized operator. - Datacenter provenance and operator authority checked - Owner-approved allocation, start/end window and interruption terms - Timestamped quote validity and withdrawal/expiry handling - Separate confirmation from reservation or delivery guarantees ### Successful-delivery total-cost comparison Status: planned; unavailable. Priority 1. Compare compute, preparation, transfer, storage, retry and migration costs for a successful task. - Declared price/currency and assumptions - Measured preparation and transfer where authorized - Retry/migration uncertainty and missing costs shown - No billing authority, payment collection or cost guarantee ### Interruption and recovery validation Status: planned; unavailable. Priority 2. Test whether an authorized scenario can recover from interruption and what recovery costs in time and resources. - Supplier-approved fault/recovery scenario and stop rules - Checkpoint/restart support tested with public or synthetic inputs - Measured recovery time and documented failure modes - Participant data and checkpoints stay with participants ### Failure analysis and evidence review Status: planned; unavailable. Priority 2. Organize minimal corroborated evidence to distinguish availability, environment, constraints and delivery failures. - Minimal event/receipt field whitelist - Independent participant statements distinguished - Conflicting accounts and unknown causes retained - No customer content, refund adjudication, escrow or liability guarantee ### Evidence expiry and repeat testing Status: planned; unavailable. Priority 1. Make test scope, sample size, age and changes visible; stale evidence must not imply current performance. - Test method/version/time and sample size - Expiry or invalidation after relevant changes - Authorized repeat measurements including failures - No permanent score based on one best result ### Datacenter identity and resource authority checks Status: research_in_progress; unavailable. Priority 1. Check the operator, professional facility provenance and authority to expose the specific resources and fields. - Public claims and independently checked identity distinguished - Private authorization records kept out of public evidence - Specific resource, field, test and availability permission - Outreach, replies and signatures alone do not establish ownership ## Neutrality and evidence boundaries Suppliers provide resources, permission and technical statements. Buyer constraints come from the buyer. Independent test measurements and participant delivery receipts must have their own disclosed sources. Provider-supplied results must be labelled as self-reports. Hash chains and signatures support integrity and signer checks; they do not prove measurement truth, physical ownership or complete history. Publish methods, conditions, sample sizes, failures, uncertainty and expiry. Supporting the project must not buy ranking or evidence levels. Actual hardware access, challenge tests, repeat testing and failure injection require specific operator permission, resource limits, stopping rules and an approved cost scope. No paid test, purchase, booking, SLA, refund service or payment collection is enabled by this roadmap.