Skip to lesson
中文
The collection/AN OPEN STUDY ROUTE

WEEK 15 / 24 · Learn, then operate

A controlled runtime

Deploy a read-only runtime: location, authentication and every bill

10 core hours + 2 optionalREAD · BUILD · PRACTISE
Begin the lesson ↓
A controlled runtimeTHE PERSONAL FDE FIELD LIBRARY

The question: Will the task run when your laptop closes, whose account pays, and who can open its results?

Before you begin: P2 works manually with basic evaluations and data boundaries. Understand browser versus server and be able to restore code.

FIG. 15.1THIS WEEK’S REASONING PATH
  1. 01Manual trigger
  2. 02Authenticate
  3. 03Run a mock
  4. 04Store & stop
Teaching diagram · Derived from this week’s lab; a suggested process, not observed results.

Concepts and learning objectives

Deploy a controlled execution environment rather than a chat window on a webpage. Distinguish the development agent, application agent and evaluation tools: their models, identities and budgets may differ. Choose either a manually triggered local tool or an authenticated, read-only hosted service. Both are valid; cloud purchases are not compulsory.

Do not schedule automatically yet. Specify trigger, runtime identity, storage, access and stop controls. Separate subscription, API and hosting costs. Supported subscription automation depends on current official rules; arbitrary applications are not automatically covered by a subscription. Consult R53 and R54.

Reading and selected sections

Priority Source and selected sections Budget
Required selections R24 · Full Stack Open Part 3 — NodeJS and Express — Node services, HTTP interfaces and deployment. 40 min
Required R60 · OWASP — Authorization Cheat Sheet — Identity, roles, object-level access checks and default deny. 35 min
Hosted option R71 · Render — Deploy a Node Express App — Official Node Express deployment steps; check costs separately. 40 min

Studio lab

  1. Draw the runtime: manual trigger, permitted public sources, optional model, controlled result storage. Label network arrows with fields, identity and payer.
  2. Choose local Node with explicit sleep, offline and shutdown limits, or one hosted service using established authentication or managed access control.
  3. Deploy a no-model health check and mock task first. Verify from an unauthenticated session that restricted requests fail without leaking configuration.
  4. Store secrets using the platform's mechanism. Confirm authentication and budget before one or two explicitly authorised read-only calls. Record measured cost or unknown.
  5. Document stopping, credential revocation, output location and cleanup. Keep real configuration separate and public examples in mock mode.

Deliverables and acceptance

Deliverable Evidence of completion
Read-only runtime Location, trigger and behaviour during shutdown are explicit.
Access test An unauthorised session cannot trigger tasks or access private results.
Three-part cost account Subscription, API and hosting have identified sources; unknown cost does not trigger automatic execution.

Independent exercise: Trace a mock HTTP request through executor and output. Remove a configuration value and confirm clear rejection rather than fallback to a real service.

Privacy and safety: Exclude production secrets from frontend builds, screenshots, logs and evaluation attachments. Local code does not imply local model inference. A consumer subscription is not institutional data-processing approval. Review hosting logs and backups separately.

Everyday use: Complete one public-information task through a fixed entry point and retain a manual alternative.

Pass or revisit: Stop the runtime, explain access and payment, and demonstrate denial before adding scheduling. Otherwise keep local manual operation.

A reusable collaboration prompt

Deploy a mock read-only runtime first. Explain local or hosted operation, authentication, storage, stopping and costs. Keep real paid calls disabled and secrets out of the browser. Verify unauthorised access is rejected before expanding visibility.

Sources and verification scope

Reference IDs and scope follow the supplied manuscript, baseline 2026-09-27. External authenticated labs were not all completed; recheck versions and access before implementation.

  1. R24
    Full Stack Open Part 3 — NodeJS and Express ↗

    HTTP servers and deployment selections.

  2. R53
    OpenAI — Billing for ChatGPT and the API platform ↗

    Separate billing systems; subscriptions are not generic API credit.

  3. R54
    Codex — Usage and pricing entry ↗

    Manuscript records a redirect to learn.chatgpt.com/docs/pricing; pricing and capability change.

  4. R60
    OWASP — Authorization Cheat Sheet ↗

    Authentication is not authorisation; enforce server-side and object-level access.

  5. R71
    Render — Deploy a Node Express App ↗

    Optional example; no deployment performed for the reader; current pricing and environment apply.