Functions — จากฟังก์ชันสู่ระบบFrom Functions to Guarded Workflows

Part 1 of twelve. Trace a familiar template function into a guarded workflow with explicit states and no external effects.

Default decisions and draft text are authored mocks, not live Jev results. Use the tutorial with the matching companion code, tests and exercise linked to a fixed GitHub version.

Outcome and prerequisites

Use Node22.18+ and a terminal. Clone the GitHub companion using its README, or download the complete ZIP, extract it and enter ai-core-functional-lab. No npm install, API key or network inference is needed. The reading-time label excludes about25–45 minutes of hands-on work. Keep one reference copy and make exercises in another.

Download companion ZIP · README / lesson map

This part focuses on src/generation.mjs. Expected gate states describe content and continuation; no state sends a message or changes a real account.

Old functions and the new boundary

Start with a familiar function: input goes in, a draft object comes out. traditionalReply is an ordinary template over maintained product facts. It is fast to understand and sufficient when the answer space is small. Adding a model is a design choice, not an automatic upgrade.

The sample product DeskLite is fictional. A user asks how to export CSV; our system may prepare a reply and may never send a file, modify an account or issue a refund. A result called accept means an artifact passed this release’s runtime checks. It grants no external authority.

Here functional means explicit typed boundaries and composable functions. It is not the name of a vendor framework. Codex helps develop the code; a separate generative provider would draft text; Jev supplies typed discriminative judgments. This local lab mocks both model-facing roles and implements no AI-Core SDK.

Seven practical steps

Step 1

Extract the companion ZIP into a writable folder and enter ai-core-functional-lab. Run doctor before changing files. It should identify 28 synthetic cases and offline mode; no package installation or account is necessary.

Step 2

Open data/facts.json and read both maintained claims. Export downloads a CSV locally and does not email it. These facts are evidence supplied by the application, not claims the draft can write into existence.

Step 3

Trace traditionalReply in generation.mjs. Its dependencies are the validated input and facts object. It returns action, language, text and citations; it does not read a key, make a request or mutate a ticket.

Step 4

Run compare for V001 and keep both outputs. The traditional value is a draft; the guarded value includes state, history and attempts. Do not compare a raw draft with a checked artifact as though their guarantees were identical.

Step 5

Read the pipeline stage order: input code, authorization, semantic input, generation, output code, semantic output and quality. Each stage can stop later work. Write that order down before studying the implementation.

Step 6

Run V004, which requests send_reply. Authorization must reject it before any provider call. A fluent answer cannot grant permission absent from the application policy. Check calls equals zero, not only the final word reject.

Step 7

Run the pipeline tests, then explain where the old function still helps: deterministic fallback after a repair request. The new workflow adds observability and judgment boundaries, while the template remains useful.

Files and runnable commands

Open the exact source and read its dependencies before editing. This is runnable reference code, not a fictional AI-Core SDK. The source links point to a fixed version of the public GitHub companion.

src/generation.mjs · Matching tests · This lesson’s exercise

export function traditionalReply(input,facts) {
  // An ordinary deterministic function is preferable when a template is enough.
  return {action:'draft_reply',language:input.language,
    text:facts.facts.map(f=>f[input.language]).join(' '),citations:['export-csv','no-send']};
}
// The mock generator is separate from Jev. No live text generation is configured.

Offline commands:

node src/cli.mjs doctor
node src/cli.mjs compare --case V001
node src/cli.mjs run --case V004
node --test --test-reporter=spec test/pipeline.test.mjs

Expected output and verification

doctor reports OFFLINE FIXTURES and paidCalls 0. V001 accepts an artifact after four mock calls. V004 rejects with action_not_authorized and zero calls. These are reproducible fixture behaviors, not measured Jev intelligence or cost savings.

Inspect reasons, history and every attempt rather than only the final label. Unknown usage must remain unknown. A green test demonstrates the specified code path, not universal semantic safety.

A failure to understand

If you return the traditional draft directly from a network endpoint, its well-formed object can hide missing policy and semantic checks. The compare command exposes that difference without claiming the model-based route is universally safer.

Exercise in a disposable copy

In a disposable copy, add a third product fact in both languages and run compare again. The template changes immediately; canned model judgments do not become more intelligent. Restore the reference before continuing.

Follow the matching exercise file, retain the original and restore the reference before the next chapter. Do not use held-out results to choose a threshold.

Troubleshooting

Cannot find cli.mjs means the terminal is in the wrong folder. If you see a request for an SDK or key, inspect the command: the provided CLI is offline-only and rejects --live.

Next checkpoint

Before continuing, name the property that code checks exactly, the judgment that may be wrong, the state that stops work and the evidence needed before release.

Next lesson →

Sources and version notes

Key takeaways