API Evangelist enhancements for the Standard Compute Inference API

overlay: 1.0.0
info:
  title: API Evangelist enhancements for the Standard Compute Inference API
  version: 1.0.0
extends: ../openapi/standard-compute-openapi.json
x-provenance:
  generated: '2026-09-02'
  method: generated
  source: >-
    Derived from artifacts in this repo (authentication/, errors/, rate-limits/,
    conventions/, lifecycle/) against the verbatim spec fetched from
    https://api.stdcmpt.com/openapi.json on 2026-09-02.
  note: >-
    The upstream document is an unedited FastAPI default — info.title "FastAPI",
    info.version 0.1.0, no servers[], no securitySchemes, no requestBody, no
    components.schemas, and a lone 200 on every operation. This overlay records what
    the provider's own documentation already states but the contract omits. It does
    NOT mutate openapi/standard-compute-openapi.json, and the scorer reads the
    original, so this improves derived artifacts and gives the provider a ready-made
    patch list — it does not inflate their content score.
actions:
  - target: $.info
    description: Name the API and the company, per the provider's own site.
    update:
      title: Standard Compute Inference API
      description: >-
        Flat-rate, smart-routed LLM inference. One endpoint speaks both the OpenAI
        Chat Completions wire format and the Anthropic Messages wire format; set the
        model to "standardcompute" and the router selects the underlying model per
        request. Billing is a fixed monthly compute budget rather than per-token.
      contact:
        name: Standard Compute
        url: https://standardcompute.com
        email: contact@standardcompute.com
      termsOfService: https://standardcompute.com/terms
  - target: $
    description: >-
      Add the production server. The spec ships with no servers[] at all, so a
      generated client has no host to call.
    update:
      servers:
        - url: https://api.stdcmpt.com
          description: Production gateway (OpenAI-compatible paths live under /v1)
  - target: $.components
    description: >-
      Declare the Bearer API key. Every /v1 completion path requires it, and its
      absence is the most misleading thing in the published contract — a client
      generated from this spec would send no Authorization header and receive 401.
    update:
      securitySchemes:
        bearerApiKey:
          type: http
          scheme: bearer
          description: >-
            Standard Compute API key from the dashboard, prefixed sc_live_. Supplied
            as "Authorization: Bearer <key>". Shown in plaintext once, at creation.
  - target: $
    description: Apply the security requirement globally.
    update:
      security:
        - bearerApiKey: []
  - target: $.paths['/v1/models'].get
    description: >-
      GET /v1/models is genuinely anonymous — the provider documents it as the
      self-service uptime check — so it overrides the global requirement.
    update:
      security: []
      tags: [Models]
      description: >-
        Returns the current smart-routing pool. Each entry carries context_length,
        supported_parameters, anthropic_family_tier and is_family_default, which is
        the only capability-discovery surface this API exposes.
  - target: $.paths['/v1/chat/completions'].post
    description: Document the failure modes the spec omits entirely.
    update:
      tags: [Inference]
      responses:
        '401':
          description: >-
            Invalid or missing API key. Body is the OpenAI error envelope:
            {"error": {"message": "Invalid API key", "type": "invalid_request_error"}}
        '402':
          description: >-
            The plan's monthly compute budget has been consumed. Requests resume when
            the billing period renews or the plan is upgraded. This is the designed
            ceiling of the flat-rate model, not a fault.
  - target: $.paths['/v1/messages'].post
    description: Same failure modes on the Anthropic-compatible path.
    update:
      tags: [Inference]
      responses:
        '401':
          description: Invalid or missing API key.
        '402':
          description: Monthly compute budget consumed.
  - target: $.paths['/healthz'].get
    description: >-
      Record a contract/deployment divergence. The spec declares GET /healthz, but a
      live request on 2026-09-02 returned an HTTP 404 Google Cloud Run error page —
      the route is described but not served at this host.
    update:
      deprecated: false
      x-api-evangelist-probe:
        fetched: '2026-09-02'
        url: https://api.stdcmpt.com/healthz
        http_status: 404
        note: declared in the spec, not served in production
x-recommendations-to-provider:
  - >-
    Set info.title and info.version. "FastAPI" / "0.1.0" is the framework default and
    means no consumer can tell one release of the contract from another.
  - >-
    Declare securitySchemes and apply security. Today the contract says the API is
    anonymous, which is false for eight of its nine operations.
  - >-
    Declare requestBody and response schemas (or $ref the OpenAI and Anthropic
    schemas). Empty bodies make code generation impossible against this spec.
  - >-
    Declare the 401 and 402 responses. 402 in particular is the ceiling of the whole
    product and appears nowhere in the machine-readable contract.
  - >-
    Emit budget headers on every response (remaining budget, period reset). Budget
    state is rendered to humans in the dashboard and to agents nowhere — this is the
    single largest agent-readiness gap on the API.
  - >-
    Serve /.well-known/security.txt pointing at https://standardcompute.com/security.
    The responsible-disclosure policy already exists and meets RFC 9116's substance;
    only the file is missing.