A message becomes an image. CVP2 uses the full RGB24 spectrum in original digital files. Landmarks frame the data; independent checks verify the recovered bytes.
Visual carries exact text or file bytes; Lexicon first adds English word addresses and optional concept references.
RGB24 · the complete address space
Every possible colour.
4,096 × 4,096 pixels. All 16,777,216 RGB colours, once each. This is an address atlas; the global directory tells you which addresses are assigned.
Preparing the full atlas…
Exact mapping and encoding
Row-major order: n = y × 4096 + x. Red = (n >> 16) & 255; green = (n >> 8) & 255; blue = n & 255; alpha = 255. The original PNG uses exact sRGB channel bytes. Fit view is a scaled preview; inspect coordinates or open the original at 1:1 for individual colours.
The atlas is not a CVP1 message: it has no message header, landmarks, or error correction. CVP1 uses sixteen repeatable transport colours to carry data. A namespace address gains an owner through the global registry.
Your message
Local · no upload
UTF-8 text · every byte is preserved
Every RGB24 value is available: three bytes per data tile, six times CVP1’s gross tile capacity. PNG file-size savings depend on the message. Send the original PNG without resizing or colour conversion.
Auto fit sizes the square from the compressed payload. Frames retain a 12-tile landmark border on each side.
Preparing the first frame…
Original PNGs retain exact colours. Keep the entire square frame and its landmarks.
A visible messageCVP2
—original bytes
—image frames
—gzip stream reduction
01 / 01
The source fingerprint will appear here.
OrientationReference stripProtected data
The whole spectrum is open.
16,777,216 colours
CVP2 stores three wire bytes in each RGB tile. At one pixel per tile, one data pixel carries 24 bits. Framing, error correction and padding reduce the final message capacity; large messages use verified numbered frames.
A million entries require storage proportional to their size. Compression helps when data repeats; the live frame count shows the actual capacity.
The damage bench
Selected transport
This bench injects wire-byte errors, not optical colour drift. RS(255,223) corrects up to 16 unknown byte errors per block. Beyond that, recovery is not guaranteed; the content hash must still match.
Bring the colours back
Verified recovery
Choose the declared transport. Its versioned header must verify; this reader does not infer the format from the colours.
⊞
Drop the PNG frames here
Choose one image or all the images in a sequence. Order does not matter.
The decoder recognizes the layout from the image itself.
CVP2 reads original RGB8 PNG files in canonical orientation with exact integer tile samples. CVP1 retains its existing tested digital-frame compatibility. Neither profile promises recovery after arbitrary editing, screenshots, photography or lossy conversion.
The recovered message
Waiting for images
Decoded content is data. This instrument never runs recovered code or renders recovered HTML.
CVP2 and CVP1 are outer transports for bytes. CVP2 maps three bytes directly to each RGB24 tile; CVP1 retains its sixteen-colour mapping. Carrier and USL1 retain their existing meaning and message rules.
For the sender
Preserve the source bytes and compute SHA-256.
Optionally gzip the source, then split it into bounded chunks.
Add the versioned header, file and sequence metadata, and hashes.
Protect every 223 bytes with 32 Reed–Solomon parity bytes; interleave the blocks.
For CVP2, map consecutive wire-byte triples to R, G and B. Export exact PNG samples with the protocol landmarks. CVP1 instead maps each nibble to its existing calibrated palette.
For the next reader
Select the declared transport and validate its supported geometry and four asymmetric corners.
For CVP2, read original RGB samples directly. For CVP1, use its existing palette decoder.
Recover the wire bytes, deinterleave, and verify the RS blocks and declared profile.
Validate the header and the current chunk’s hash.
Collect the complete matching sequence, verify its stream hash, decompress if needed, and verify the original length and SHA-256.
Protected stream capacity by transport
Data grid
CVP1 bytes / frame
CVP2 bytes / frame
Default image
64 × 64
1,561
10,481
352 × 352 px · 4 px/tile
128 × 128
6,913
42,593
608 × 608 px · 4 px/tile
256 × 256
28,321
171,710
1,120 × 1,120 px · 4 px/tile
1,024 × 1,024
458,265
2,750,705
3,144 × 3,144 px · 3 px/tile
2,048 × 2,048
1,833,729
11,003,489
2,072 × 2,072 px · 1 px/tile
4,096 × 4,096
7,335,585
44,015,294
4,120 × 4,120 px · 1 px/tile
Auto fit selects the smallest multiple-of-eight grid after optional compression. The full 4,096 CVP2 data grid carries 44,015,294 protected stream bytes in a 4,120 × 4,120 image including landmarks. One-pixel tiles are available at every grid. The separate full namespace atlas is 4,096 × 4,096 and contains no message framing.
Instructions that travel
The fixed header carries the grid, page index and count, byte lengths, compression flag, file name and type, message identity, and SHA-256 fingerprints. The selected decoder checks the in-band magic and version before it accepts a frame.
CVP2 RGB24 Exact: reference codec, specification and release hashes. Read original PNG files as integer RGB samples. The sixfold gain compares gross data-tile capacity, not PNG file size, semantic capacity or optical robustness.
The retained CVP1 executable reference is cvp1/1.2.0/cvp1.js, with its normative specification and release hashes. This standalone embeds that same reference bundle. Its frozen version 1 application notes describe the earlier namespace rules; the current global reservation directory adds an allocation layer without changing CVP1 bytes. Agents can inspect window.CarrierVisual and the embedded ucv1-spec JSON. Where supported, the page also registers structured tools for encoding, reading, and inspecting the protocol.
Training material, with ground truth
The paired export contains original PNGs and a manifest linking each frame to exact source bytes, the message hash, page index, palette, and protocol version. The exact source bytes are included as source.bin.
These are reproducible examples for training or evaluation. A model’s visual understanding and decoding accuracy still need to be measured; image encoding alone does not confer either.
The first colour dictionary, audited. The earlier 40-card image recovered all 24,009 bytes exactly. Entry 7 uses two text frames, and entry 8 begins the brighter particle anchors. Its separate layout file motivated this self-contained transport. That historical snapshot is the audit baseline. The dictionary action in this instrument fetches and verifies both current public files, shows their heads, and packages the exact fetched text.
One global colour now belongs to one current entity. The global directory reserves English forms and Carrier concepts together. The semantic-reference example still binds every concept to a registry, tier, entry ID, and exact dictionary head so historical messages remain readable. Historical anchors require their original context. References can point into a much larger shared corpus, but the receiver needs that particular snapshot; compact references do not contain the definitions themselves. Spatial order and typed relations require a declared grammar. This draft application profile preserves references without claiming to resolve philosophical meaning.