Input Code — ตรวจ input ด้วยโค้ดCheck Inputs with Ordinary Code

Part 3 of twelve. Keep exact schema, type, size and field checks in ordinary code before spending an inference budget.

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/input.mjs. Expected gate states describe content and continuation; no state sends a message or changes a real account.

Old functions and the new boundary

Keep observable input properties in ordinary code: field names, language enum, ID shape, string type and size. A probabilistic call adds latency and ambiguity to checks a parser can decide exactly. This is familiar validation, not an obsolete practice.

The allowed fields exclude evaluator labels and instruction-like extras. NFC normalization and trimming operate after type and size checks. JavaScript length counts UTF-16 code units, not linguistic words; this bound limits payload size rather than measuring readability.

A short deny pattern catches known injection examples, but it cannot identify every hostile paraphrase. Code filters reduce obvious noise before a fallible semantic guard. Separating these stages makes their limits testable instead of calling a regex a universal security system.

Seven practical steps

Step 1

Read the four allowed input keys and their meanings. An ID identifies a synthetic case, language selects the response track, request is untrusted text, and requestedAction is still subject to authorization.

Step 2

Run V001 through the input command. The returned value should be a normalized copy with state accept. This only accepts the input shape; it says nothing about the eventual response quality.

Step 3

Run V002, whose request is a number. Reject it before normalization; coercing it with String would create a different request instead of validating the caller’s request.

Step 4

Run V003 and identify the exact pattern responsible for rejection. Keep a test for that phrase, but also record that the result depends on the pattern list. This prevents a passing test from becoming a sweeping claim.

Step 5

Inspect the object and field checks. Unknown keys are refused, arrays are not accepted as records, and IDs cannot become file paths. Validate before providers see a request so malformed work consumes no inference budget.

Step 6

Read the normalization test and verify the original object remains unchanged. Composition is easier when one gate returns a new value rather than mutating the caller’s input behind its back.

Step 7

Connect the input gate to the pipeline’s early return. For malformed input, provider call spies must stay at zero. The boundary is not complete until downstream work actually stops.

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/input.mjs · Matching tests · This lesson’s exercise

import {gate,plain} from './contracts.mjs';
export function checkInput(raw) {
  if(!plain(raw)) return gate('reject',['input_not_object']);
  const keys=['id','language','request','requestedAction'];
  if(Object.keys(raw).some(k=>!keys.includes(k))) return gate('reject',['unknown_input_field']);
  if(typeof raw.id!=='string'||!/^[VHA]\d{3}$/.test(raw.id)) return gate('reject',['invalid_id']);
  if(!['en','th'].includes(raw.language)) return gate('reject',['unsupported_language']);
  if(typeof raw.request!=='string'||!raw.request.trim()||raw.request.length>2000)
    return gate('reject',['invalid_request_size']);
  if(typeof raw.requestedAction!=='string') return gate('reject',['missing_action']);
  const request=raw.request.normalize('NFC').trim();
  if(/ignore (all|previous)|system prompt|bypass the gate/i.test(request))
    return gate('reject',['known_injection_pattern']);
  // This deliberately small pattern list is not a complete injection detector.
  return gate('accept',[],{id:raw.id,language:raw.language,request,requestedAction:raw.requestedAction});
}

Offline commands:

node src/cli.mjs input --case V001
node src/cli.mjs input --case V002
node src/cli.mjs input --case V003
node --test --test-reporter=spec test/contracts.test.mjs

Expected output and verification

V001 accepts shape; V002 rejects invalid_request_size; V003 rejects known_injection_pattern. Pipeline tests confirm these code-level refusals make no provider calls. No model judgment is needed to establish those particular properties.

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

A permissive coercion may turn an array or number into text and allow later semantic work. That hides a caller bug and changes the test population. Refuse the malformed representation explicitly instead.

Exercise in a disposable copy

Add a test at exactly 2000 code units and another at 2001. Then add an unknown property to an otherwise valid record. Keep these as code tests rather than asking Jev to count or inspect schema.

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

Thai combining marks can make visual character counts differ from length. Do not describe this payload bound as a Thai word limit. If you change normalization order, rerun size-boundary tests deliberately.

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.

← Previous lesson · Next lesson →

Sources and version notes

Key takeaways