QR code bit stream walkthrough, from text to modules

The pipeline at a glance

segments → terminator → pad bytes → blocks → RS codewords → interleave
        → remainder bits → zig-zag placement → mask → format/version info

Each stage is mechanical; run them in order and any two correct encoders produce the same codewords for the same input, version, and error-correction level. (Divergence enters only at mask selection.)

Stage 1: the data bit stream

Each segment is a 4-bit mode indicator, a character count field (8–16 bits wide depending on mode and version), then the encoded characters. After the last segment come the closing formalities:

1. Terminator, up to four 0 bits (fewer if capacity runs out exactly).
2. Pad to a byte boundary with 0s.
3. Pad bytes, the alternating pair 11101100 00010001 (0xEC 0x11) repeated until the version's data capacity is exactly full. The values are not arbitrary: they are balanced bit patterns chosen so padding never creates the uniform areas the penalty rules exist to punish. Spot EC 11 EC 11 in a hex dump and you are looking at a QR code's slack space.

Stage 2: codewords, blocks and error correction

The stream is cut into 8-bit codewords and split into the block structure the standard fixes for this version and level. Each block gains its Reed–Solomon check codewords, and the blocks are then interleaved, data codewords round-robin across blocks, then check codewords the same way.

A final oddity: some versions' module counts are not a multiple of 8, so remainder bits (always 0) top up the stream: 7 of them for versions 2–6, 3 or 4 for various versions from 14 to 34, none elsewhere.

Stage 3: placement: the zig-zag

Function patterns claim their territory first; data fills what is left, in a pattern best described as reading a book backwards:

• Placement works in vertical column pairs, two modules wide, starting at the bottom-right corner.
• The first pair fills upward (within each two-module row of the pair, right cell then left cell), until it hits the top, then the next pair to the left fills downward, alternating all the way across.
• Any cell claimed by a function pattern or reserved field is skipped, the sequence flowing around it; and the entire single column of the vertical timing pattern (column 6) is stepped over, keeping pairs aligned either side of it.

The most significant bit of the first codeword lands in the bottom-right corner cell.

Stage 4: mask, then metadata

The placed data region is XORed with the winning mask pattern. Finally the 15-bit format information (level + mask, BCH-protected, XORed with 101010000010010) is written twice, and from version 7 the 18-bit version information twice more. The symbol is complete.

To watch every stage with real numbers, follow the worked example in encode HELLO WORLD by hand, or generate any payload and decode it back to confirm the pipeline round-trips.

FAQ

What are the pad bytes 0xEC and 0x11 in a QR code?

Filler appended after the terminator to bring the data stream exactly to the version's capacity. They alternate (11101100, 00010001), and their balanced bit patterns avoid creating large uniform areas that would scan poorly.

How is data physically placed in a QR code?

In two-module-wide column pairs, starting at the bottom-right corner, filling upward then downward alternately across the symbol, skipping function patterns and stepping over the vertical timing column entirely.

What are remainder bits?

Zero bits appended after the final interleaved codeword because some versions have a few more data modules than a whole number of codewords fills: 7 bits for versions 2–6, 3 or 4 for several mid-range versions, none for the rest.

What order do the encoding stages run in?

Segments with mode and count headers, terminator, byte padding, pad bytes, block split, Reed–Solomon check codewords, interleaving, remainder bits, zig-zag placement, masking, and finally the format and version information fields.