diff --git a/CHANGELOG.md b/CHANGELOG.md
index fca9b994e..651577513 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -53,6 +53,22 @@ The sections do not signal obligations; the prefixes do.
 
 ## Unreleased
 
+### Changed
+
+- The `offset` of a segment counts bytes from the most significant byte of the
+  slot: byte `0` is the first byte of the slot's big-endian word. The
+  description now says so, says how to convert from a layout that counts from
+  the low-order end (`$wordsize - o - n`), and allows writing that conversion as
+  an expression. New segment examples pack an `address` and a `uint32` into
+  one slot, each written with literal offsets (`12` and `8`) and with an
+  expression that takes the value's size from the region's own `length`
+  (`{ ".length": "$this" }`) ([#309]).
+  - Schemas: **ethdebug/format/pointer/scheme/segment**
+  - Producers: no change needed. The reference implementation and bugc already
+    resolve offsets this way; the text states existing meaning.
+  - Consumers: no change needed. A consumer that resolved `offset` against the
+    big-endian word already follows this.
+
 ## 0.1.0-draft.0 — 2026-09-21
 
 The version scheme changed: prerelease versions of the specification are now
@@ -616,3 +632,4 @@ First published version of the specification.
 [#285]: https://github.com/ethdebug/format/pull/285
 [#286]: https://github.com/ethdebug/format/pull/286
 [#303]: https://github.com/ethdebug/format/pull/303
+[#309]: https://github.com/ethdebug/format/pull/309
diff --git a/packages/pointers/src/segment.examples.test.ts b/packages/pointers/src/segment.examples.test.ts
new file mode 100644
index 000000000..ea2dc7794
--- /dev/null
+++ b/packages/pointers/src/segment.examples.test.ts
@@ -0,0 +1,87 @@
+import { expect, describe, it } from "vitest";
+
+import type { Pointer } from "@ethdebug/format";
+import { schemas } from "@ethdebug/format";
+
+import type { Machine } from "#machine";
+import { Data } from "#data";
+import { dereference } from "./dereference/index.js";
+
+const { examples } = schemas[
+  "schema:ethdebug/format/pointer/scheme/segment"
+] as {
+  examples: Record<string, unknown>[];
+};
+
+// the packed examples all address slot 2
+const packed = examples.filter((example) => example.slot === 2);
+
+// slot 2 packs an address (bytes 12 to 31) and a uint32 (bytes 8 to 11); a
+// byte's number counts from the most significant end of the word
+const address = Uint8Array.from({ length: 20 }, (_, i) => 0xa0 + i);
+const count = new Uint8Array([0xde, 0xad, 0xbe, 0xef]);
+const word = new Uint8Array(32);
+word.set(address, 12);
+word.set(count, 8);
+
+const state = {
+  stack: { length: Promise.resolve(0n) },
+  storage: {
+    read: async ({
+      slot,
+      slice,
+    }: {
+      slot: Data;
+      slice?: Machine.State.Slice;
+    }) => {
+      expect(slot).toEqual(Data.fromNumber(2));
+      return Data.fromBytes(
+        slice
+          ? word.slice(
+              Number(slice.offset),
+              Number(slice.offset + slice.length),
+            )
+          : word,
+      );
+    },
+  },
+} as unknown as Machine.State;
+
+async function resolve(example: Record<string, unknown>): Promise<Data> {
+  const pointer = { location: "storage", ...example } as Pointer;
+  const cursor = await dereference(pointer);
+  const { regions, read } = await cursor.view(state);
+  return read(regions[0]);
+}
+
+describe("segment schema examples (packed)", () => {
+  it("include literal and expression offsets", () => {
+    expect(packed.map((example) => example.offset)).toEqual([
+      12,
+      8,
+      {
+        $difference: ["$wordsize", { ".length": "$this" }],
+      },
+      {
+        $difference: ["$wordsize", { $sum: [20, { ".length": "$this" }] }],
+      },
+    ]);
+  });
+
+  it("count offsets from the most significant byte", async () => {
+    const [literalAddress, literalCount] = packed;
+
+    expect(await resolve(literalAddress)).toEqual(Data.fromBytes(address));
+    expect(await resolve(literalCount)).toEqual(Data.fromBytes(count));
+    expect(await resolve({ slot: 2, offset: 0, length: 1 })).toEqual(
+      Data.fromBytes(new Uint8Array([0x00])),
+    );
+  });
+
+  it("resolve each expression like its literal twin", async () => {
+    const [literalAddress, literalCount, exprAddress, exprCount] = packed;
+
+    expect(await resolve(exprAddress)).toEqual(await resolve(literalAddress));
+    expect(await resolve(exprCount)).toEqual(await resolve(literalCount));
+  });
+});
diff --git a/packages/web/spec/pointer/concepts.mdx b/packages/web/spec/pointer/concepts.mdx
index da47aa746..4e41e1c70 100644
--- a/packages/web/spec/pointer/concepts.mdx
+++ b/packages/web/spec/pointer/concepts.mdx
@@ -283,6 +283,9 @@ This format currently defines two such addressing schemes:
 single continuous byte array and addressing a slot or collection of slots
 in a word-arranged locaton (respectively).
 
+Within a slot, the segment scheme numbers bytes from the most significant
+byte: byte `0` is the first byte of the slot's big-endian word.
+
 <details>
 <summary>
 **Example**: Slice-based region
diff --git a/schemas/pointer/scheme/segment.schema.yaml b/schemas/pointer/scheme/segment.schema.yaml
index 487795bf5..4404a431d 100644
--- a/schemas/pointer/scheme/segment.schema.yaml
+++ b/schemas/pointer/scheme/segment.schema.yaml
@@ -24,9 +24,32 @@ properties:
     description: |
       The starting byte index within the slot.
 
+      Bytes within a slot are numbered from the most significant byte. A
+      slot's value is its `$wordsize`-byte big-endian word, and byte `0` is
+      the first byte of that word, as if the word were written to memory. An
+      `offset` of `0` therefore addresses the most significant byte of the
+      slot, and an `offset` of `$wordsize - 1` addresses the least
+      significant byte.
+
       This field is **optional**. If unspecified, it has the default value of
       `0`, indicating that the segment begins at the start of the specified
-      slot.
+      slot (its most significant byte).
+
+      A data layout that counts bytes from the low-order end of a slot must
+      convert: a value of `n` bytes that sits `o` bytes from the low-order end
+      is at `offset` `$wordsize - o - n`. An emitter may write that number as
+      a literal, which is the simplest form to read and resolve. It may also
+      write the conversion as an expression, such as
+
+      ```json
+      {
+        "$difference": ["$wordsize", { "$sum": [o, { ".length": "$this" }] }]
+      }
+      ```
+
+      which can take `n` from the region's own `length`, keeps the layout's
+      own numbers visible, needs no arithmetic in the emitter, and does not
+      depend on a fixed word size.
 
       This field's expression must resolve to a non-negative value. It is
       **not** bounded by the word size: an offset that meets or exceeds
@@ -77,3 +100,27 @@ examples:
   - slot: 0
     offset: $wordsize
     length: 4
+  # packed values: an `address` (20 bytes) at the low-order end of slot 2,
+  # and a `uint32` (4 bytes) just above it, written with literal offsets
+  - slot: 2
+    offset: 12
+    length: 20
+  - slot: 2
+    offset: 8
+    length: 4
+  # the same two values with the conversion `$wordsize - (o + n)` written as
+  # an expression that takes `n` from the region's own length
+  - slot: 2
+    offset:
+      $difference:
+        - $wordsize
+        - .length: $this
+    length: 20
+  - slot: 2
+    offset:
+      $difference:
+        - $wordsize
+        - $sum:
+            - 20
+            - .length: $this
+    length: 4