Provider Pricing Page Field Audit

Field	Why it matters	Where it appears or may need verification	Serving modes affected	RunPlacement next step
Token or request units	Turns usage into billable API or model calls	Official pricing pages, API docs, tokenizer docs, logs	API inference, serverless models, batch APIs	AI cost per token
Input, cached input, and output	Input and output may price differently; caching can change the estimate	Official model pricing pages, prompt caching docs, request logs	API inference, managed model APIs	Inference cost per request
Failed or filtered request treatment	Not every paid attempt becomes useful output	Provider pricing notes, API logs, error handling docs	API inference, managed inference	LLM API bill too high
Endpoint or compute hours	Always-on endpoints and jobs may bill by time rather than token	Deployment docs, pricing pages, cloud bills	Managed inference, SageMaker, GPU cloud, self-hosted GPU	GPU utilization for inference
Serverless duration or request processing	Intermittent workloads may avoid idle capacity but still have duration and request rules	Serverless inference docs, pricing pages	Serverless inference, API serving	AI inference cost calculator
Batch or async mode	Queueing can change utilization, latency, and pricing mechanics	Batch API docs, asynchronous inference docs, batch transform docs	Batch inference, async inference, offline jobs	Batch inference cost savings
Minimum capacity	Minimum instances or deployment sizes can create a fixed baseline	Deployment docs, provider quote, quota pages	Managed inference, provisioned throughput, GPU cloud	Self-hosted inference break-even
Provisioned throughput or reservation	Reserved capacity changes cost tolerance and traffic requirements	Provisioned throughput docs, reservations, quota pages	Azure Foundry PTU, managed inference, high-scale production	Managed inference vs GPU cloud
Region or global scope	Region, global routing, quota, and currency can change availability or costs	Pricing page selectors, quota pages, deployment docs	Cloud API, managed inference, GPU cloud	Provider quote checklist
Storage and artifacts	Payloads, model artifacts, logs, snapshots, and volumes sit outside headline inference rates	Cloud storage docs, endpoint docs, bills, quotes	Managed inference, batch jobs, GPU cloud, self-hosted GPU	AI inference cost checklist
Data transfer and networking	Ingress, egress, cross-region, private networking, and payload movement can change placement	Cloud networking docs, bills, quotes	Managed inference, GPU cloud, cloud migration	Workload placement worksheet
Support and operations	Support tier, incident ownership, monitoring, rollback, and billing controls affect production cost	Provider docs, support plans, quote terms, team ownership	Managed inference, GPU cloud, self-hosted GPU	API vs self-hosted inference

You are helping me audit provider pricing-page fields for an AI inference cost estimate. Do not rank providers. Do not assume current rates. Do not make benchmark, latency, throughput, or reliability claims unless I provide source data.

Here are the provider source notes:
[Paste official pricing-page notes, deployment docs, quote terms, bill details, or logs here]

Please:
1. Identify which fields are present: token/request units, input/output, endpoint hours, batch/async mode, minimum capacity, provisioned throughput, region, storage, networking, support, and operations.
2. Identify which fields are missing or need direct provider verification.
3. Explain which missing fields could change the calculator result or serving-mode decision.
4. Separate official pricing-page facts, provider quote terms, workload assumptions, and unknowns.
5. Recommend the next RunPlacement page to use: assumptions index, calculator, checklist, cost model, managed inference vs GPU cloud, batch vs realtime, or API vs self-hosted.
6. Keep the output provider-neutral and avoid current pricing, benchmark, or provider-ranking claims unless supplied in the source notes.