Evidence guide
GPU cloud buying becomes unreliable when a catalog price is treated as proof that the same GPU can be provisioned now, in the required region, under the required account and quota. SaaS Sentinel keeps pricing evidence and availability evidence separate so that missing capacity evidence stays visible.
The core distinction
A source-backed price can be perfectly valid and still be insufficient for an immediate deployment. The price tells you that the provider publishes a commercial rate for an identified offer. It does not automatically tell you whether capacity exists now, whether your account has quota, whether the exact GPU variant is present in the chosen region, or whether provisioning will succeed at the desired scale.
Price evidence supports
You can use the published rate as a market observation when its source, unit, currency, GPU identity, and freshness pass the pricing gates. This is the right evidence for price comparison and workload-cost scenarios.
Price evidence does not support
That stronger claim needs separate evidence. A provider can keep a catalog price online during a stock shortage, quota constraint, closed region, account restriction, or temporary scheduling problem.
Availability evidence supports
Depending on the source, availability evidence may support that an offer, product, or region has a qualifying capacity signal. The scope of the evidence matters: a global status cannot automatically prove every region or GPU count.
Even then
Evidence reduces uncertainty; it does not reserve capacity. Final deployment depends on the provider's current scheduler, your account, quota, requested node size, region, and the time between observation and launch.
Evidence ladder
The exact provider implementation differs, but the principle is stable: stronger, narrower, fresher first-party evidence deserves more confidence than a generic marketing page or a price listing that says nothing about capacity.
Weak for capacity
Useful for price and product identity. Treat availability as unknown unless the page contains a separate, explicit, current capacity signal that can be tied to the relevant product or region.
Better
A documented first-party field can show that a product or configuration is offered in a region. SaaS Sentinel labels this deployment_eligibility; it never proves inventory is free at this moment.
Stronger
A current first-party endpoint that explicitly reports stock for the relevant region, GPU, or node shape is labelledstock_status. It can support a live-stock signal, but not the requested quantity unless that is also published.
Stronger still
A first-party quantity tied to a configuration and region is labelled allocatable_capacity. The decision engine checks that quantity against requested GPUs or nodes instead of treating a generic available flag as sufficient.
Operational confirmation
A successful launch is strongest for that account, request, and moment. It still does not prove identical capacity remains available later, so one event is never turned into a permanent market-wide stock label.
Five checks
If availability affects the decision, preserve each dimension that can break equivalence. A price-only ranking can be useful for market discovery, but it should not be the final procurement answer.
Confirm memory, form factor, GPU count, interconnect, and node configuration instead of relying on the family name alone.
Check that the capacity signal applies to the geography you can actually use. A provider-level signal may hide regional shortages.
Spot, on-demand, reserved, community, and managed/serverless capacity can have different stock and scheduling behavior.
Availability decays faster than a stable list price. Compare the observed time with the source's expected refresh interval.
A signal for one GPU does not guarantee eight; a public catalog does not guarantee your quota, credit, verification, or account eligibility.
How SaaS Sentinel behaves
The system is intentionally designed to avoid converting missing evidence into a positive capacity claim. This makes the dataset less cosmetically complete, but more useful for real decisions because uncertainty remains explicit rather than hidden.
The rate can remain comparable for price analysis while availability is shown as unevidenced. The offer is not deleted merely because the capacity layer is incomplete, and it is not labeled available merely because the price is current.
The pipeline can retain last-known-good evidence while degrading freshness instead of silently advancing the timestamp or replacing a validated record with malformed data.
When a source begins exposing stronger structured availability data, the coverage layer can improve without changing the neutral price-ranking policy or pretending historical rows had evidence they did not have at the time.