Your Type Guard Can Silently Drift from Your TypeScript Type 🔧

Question

Hoi hoi! 👋

I'm @nyaomaru, a frontend engineer just back from a short vacation on Texel, a small island in the Netherlands. 😸🏝️

Today, let's talk about a type guard that looks completely safe.

const isUser = (value: unknown): value is User => {
  // runtime checks...
};

Looks good, right?

TypeScript knows that when isUser(value) returns true, the value is a User.

But there's a small problem

TypeScript trusts that promise.

It doesn't prove that your runtime checks actually validate every field in User.

And that's where type guards can slowly drift away from the types they claim to protect.

Let's take a look! 👀

---

🕳️ A Type Guard Can Become Outdated Without an Error

Imagine we start with this type

type User = {
  id: string;
  name: string;
};

And a hand-written type guard

const isUser = (value: unknown): value is User => {
  if (typeof value !== "object" || value === null) {
    return false;
  }

  const candidate = value as Record<string, unknown>;

  return typeof candidate.id === "string" && typeof candidate.name === "string";
};

So far, everything matches.

Later, we update User

type User = {
  id: string;
  name: string;
  role: "admin" | "member";
};

But we forget to update the guard.

const isUser = (value: unknown): value is User => {
  if (typeof value !== "object" || value === null) {
    return false;
  }

  const candidate = value as Record<string, unknown>;

  return typeof candidate.id === "string" && typeof candidate.name === "string";
};

There is no role check.

But this still compiles. 😿

---

🧠 Why Doesn't TypeScript Catch This?

Because this

(value: unknown): value is User

is a user-defined type predicate.

You are telling TypeScript

Trust me. If this function returns true, the value is a User.

TypeScript can check whether the declared predicate type itself makes sense.

But it cannot generally prove that arbitrary runtime logic actually validates every part of that type.

So this is possible

const isUser = (_value: unknown): _value is User => true;

Terrible guard.

Perfectly valid TypeScript. 😹

The return type is a contract written by us, not a proof generated from the function body.

---

🔄 This Becomes a Maintenance Problem

The annoying part isn't writing the guard once.

It's keeping these two things synchronized over time

TypeScript type
      ↕
Runtime validation

Types change.

Properties get

• added
• removed
• renamed
• made optional
• changed to another type

And every time that happens, we need to remember that some runtime guard somewhere may also need an update.

If we forget, the compiler may not tell us.

That's the kind of bug I really don't want to rely on memory to prevent.

---

✅ What If the Type Could Be the Contract?

This is one of the reasons I added typedStruct to is-kit.

Suppose the application type already exists:

type User = {
  id: string;
  name: string;
  age?: number;
};

We can build the guard against that existing type

import { isNumber, isString, optionalKey, typedStruct } from "is-kit";

const isUser = typedStruct<User>()({
  id: isString,
  name: isString,
  age: optionalKey(isNumber),
});

Now the field map has a type-level relationship with User.

At runtime, it still performs ordinary object validation.

But at compile time, TypeScript can check whether the guards we declared match the object type they're supposed to follow.

---

💥 Now Drift Becomes Visible

Let's add a field again

type User = {
  id: string;
  name: string;
  role: "admin" | "member";
  age?: number;
};

But forget to update the guard

typedStruct<User>()({
  id: isString,
  name: isString,
  age: optionalKey(isNumber),

  // TypeScript error:
  // role is missing
});

Nice.

The runtime bug became a compile-time problem.

The same thing happens if the guard uses an incompatible field type

import {
  isNumber,
  isString,
  oneOfValues,
  optionalKey,
  typedStruct,
} from "is-kit";

typedStruct<User>()({
  id: isString,

  name: isNumber,
  // TypeScript error:
  // User["name"] is string

  role: oneOfValues("admin", "member"),
  age: optionalKey(isNumber),
});

This is the part I care about most.

typedStruct doesn't eliminate maintenance.

It makes forgotten maintenance visible.

---

🧩 Optional and Nullable Are Different

Another place where object guards can get confusing is optional properties.

Consider:

type User = {
  id: string;
  nickname?: string | null;
};

There are two separate ideas here

nickname may be absent

and

nickname may exist with the value null

Those are different runtime contracts.

With typedStruct

import { isString, nullable, optionalKey, typedStruct } from "is-kit";

const isUser = typedStruct<User>()({
  id: isString,
  nickname: optionalKey(nullable(isString)),
});

Now

isUser({ id: "user-1" });
// true

isUser({
  id: "user-1",
  nickname: null,
});
// true

isUser({
  id: "user-1",
  nickname: "Neko",
});
// true

isUser({
  id: "user-1",
  nickname: 42,
});
// false

I like keeping these two decisions explicit:

• optionalKey(...) → the property may be absent
• nullable(...) → the value may be null

They look similar at first, but they describe different things.

---

🌳 Nested Types Don't Need to Be Duplicated Either

Now imagine a larger type:

type Account = {
  readonly id: string;

  readonly profile: {
    readonly displayName: string;
    readonly bio: string | null;
  } | null;

  readonly tags: readonly string[];
};

We could manually copy the profile shape into another type.

But that creates another thing that can drift.

Instead, we can reference the type we already have

import { arrayOf, isString, nullable, typedStruct } from "is-kit";

const isProfile = typedStruct<NonNullable<Account["profile"]>>()({
  displayName: isString,
  bio: nullable(isString),
});

const isAccount = typedStruct<Account>()({
  id: isString,
  profile: nullable(isProfile),
  tags: arrayOf(isString),
});

This is the model I like:

Reuse the existing type at compile time. Compose small guards at runtime.

The application type remains the source we want the guard to follow.

---

🔒 What About Extra Runtime Properties?

There is another distinction worth making.

These are two different questions:

1. Does my guard definition match the TypeScript type?
2. Should a runtime object be allowed to contain additional properties?

By default, an object can still have additional keys.

If you want the runtime object shape to be closed as well, you can enable exact mode:

import { isString, typedStruct } from "is-kit";

type User = {
  id: string;
  name: string;
};

const isExactUser = typedStruct<User>()(
  {
    id: isString,
    name: isString,
  },
  {
    exact: true,
  },
);

Then:

isExactUser({
  id: "user-1",
  name: "Ada",
});
// true

isExactUser({
  id: "user-1",
  name: "Ada",
  debug: true,
});
// false

Whether extra properties should be rejected is a runtime policy decision.

It shouldn't be confused with keeping the guard definition synchronized with the TypeScript type.

---

⚖️ Which Should Be the Source of Truth?

I don't think there is one correct validation style for every project.

The important question is

What already owns the shape of this data?

Manual predicate

const isSomething = (value: unknown): value is Something => {
  // custom logic
};

Great when the validation is unusual or not primarily structural.

Guard-first

const isUser = struct({
  id: isString,
  name: isString,
});

Useful when the guard itself should define the resulting type.

Type-first

const isUser = typedStruct<User>()({
  id: isString,
  name: isString,
});

Useful when User already exists and the runtime guard needs to stay aligned with it.

Schema-first

A schema library or code generation may be the better source of truth when you need things like:

• structured validation errors
• coercion
• transforms
• defaults
• generated artifacts

These solve different problems.

I don't think every boolean validation check needs to become a schema. 😸

---

🚫 What typedStruct Does Not Do

There are some important boundaries.

typedStruct does not generate runtime validation from a TypeScript type.

Types are erased at runtime, so you still need to declare the guards you want to execute.

It also doesn't:

• prove that every custom predicate is honest
• coerce values
• return rich structured validation errors
• replace schema-first workflows
• validate numeric or symbol properties as part of its string-keyed object contract

It's intentionally smaller than that.

The goal is simply to create a typed bridge between

the object type you already have

and

the runtime guards you choose to run

---

🎯 The Important Part

The main point isn't really typedStruct.

It's this

A type predicate is a promise, not a proof.

This

(value): value is User

doesn't mean TypeScript inspected your implementation and proved that every User field was validated.

We made that promise.

So when a TypeScript type is the source of truth, I think it's useful to make the runtime guard structurally depend on that type instead of relying on us to remember every future change.

That's what I wanted typedStruct to help with. 😸

If your guard defines the type, use a guard-first approach.

If an existing TypeScript type should define the contract, connect the guard to that type.

And if you need rich parsing, transforms, coercion, or detailed errors, that's where a schema starts to earn its weight.

I wrote a more complete guide about this on the is-kit documentation site:

https://is-kit.dev/guides/keep-type-guards-in-sync

If you like small reusable TypeScript type guards, is-kit is open source too!

https://github.com/nyaomaru/is-kit

Thanks for reading! 🙌