Compute is disposable

This observatory prices interruptible CPU. It does not schedule work, and it does not treat local disk as part of the machine. A Spot VM can disappear without taking persistent application state with it.

Workload
   │
   ├── Compute ── disposable VM ── AWS Spot / GCP Spot / Azure Spot
   │
   └── Storage ── independent service ── object / block / cache

Intended platform

Ephemeral compute
        │
 AWS Spot  GCP Spot  Azure Spot
        │
 independent cache
     │          │
 object store   block store

A persistent disk in one cloud does not move with a workload to another cloud. Object storage or an application-level cache is the portable place for state. S3, GCS, Azure Blob, Cloudflare R2, a distributed cache, or persistent block storage can sit behind that boundary later. They are outside this dashboard.

CI shape

CI runner
   │
 ephemeral workspace
   ├── disposable files
   └── remote cache → object/cache storage

If the Spot VM disappears: the runner dies, a new runner starts, the cache is restored, and the job retries.

Future scheduler

The observatory is the market-data side of a later scheduler. That scheduler would choose AWS, GCP, or Azure Spot using price, capacity signal, interruption risk, startup latency, region, architecture, and customer constraints. Storage stays independent. Controlled placement probes, which launch a VM only to measure success and then terminate it, are not part of this version.