Cross-cutting trails

Each trail is a curated reading path. Designed for a reviewer to read one domain end-to-end in 30–45 minutes, blending what's implemented with what's designed.

12 steps

GPU Slice

From SKU through slot map, orchestrator placement, node-agent provisioning, networking, telemetry, and release. The most detailed trail.

9 steps

App Platform

App manifest, artifact trust, runtime profiles, single-allocation adapters (Slurm, RKE2, Jupyter, vLLM), billing alignment.

8 steps

Billing & Payments

Immutable ledger, accrual loop, low-balance/depletion enforcement, Stripe webhooks, refund policy.

10 steps

IAM & Tenancy

Tenant root, projects, membership, service accounts, OIDC, JWKS, scope-aware authz, audit.

10 steps

Node & MAAS

Bootstrap trust, PKI/mTLS, MAAS bare-metal lifecycle, node-agent task model, log gateway, RCAs.

7 steps

Terminal & Sessions

Browser terminal, WebSocket bridge, terminal token mint, SSH key release, RCA on HTTP/2 buffering.

How trails are structured

flowchart LR
    A[1. What this is<br/>plain-English intro] --> B[2. As-built code path<br/>with source citations]
    B --> C[3. Design intent<br/>links to source docs]
    C --> D[4. Operational angle<br/>runbooks + RCAs]
    D --> E[5. Comparative position<br/>vs other clouds]

Each step links into either the portal's curated pages or the source-of-truth docs under /source/.