UNKANI Cypher & Packed Text
(v8.3.0) • 2026-10-04

Alphabet Symbols: . | x + o
Phonetic Names:
  .  = "dot"
  |  = "bar"
  x  = "ex"
  +  = "plus"
  o  = "circle"
  -  = "dash"
Core Structure:
- v6 messages are triplets (3 symbols) = Letters, Macros, Punctuation, Emojis, Common Words
- v6 word delimiter = - (hyphen)
- Parsing order:
  1) If a message BEGINS with || it is v8 PACKED TEXT. The rest of the message
     is a packed payload, and nothing inside it is read as a v6 token.
  2) Every other message decodes with the published v6 decoder, unchanged.
     v6 decoding ALWAYS begins in DEFAULT mode, so every previously published
     v6 message decodes exactly as before.
- Special Prefixes / Modes (v6):
  .   = Letters
  x   = UNKANI Macros
  +   = Common Words
  o   = Punctuation & Emojis
  |   = Expanded Concepts (includes mode switches)
  |.| = NUMBER MODE (pairs of 2 symbols until delimiter)
  |x| = PHRASE MODE (one 6–12 symbol key until delimiter)
  |xx = IMAGE MODE (next 4- or 5-symbol block until delimiter)
  ||  = PACKED TEXT MODE (v8.2 — the whole remainder is a packed payload)

LETTERS A–Z
A  ...   B  ..|   C  ..x   D  ..+   E  ..o
F  .|.   G  .||   H  .|x   I  .|+   J  .|o
K  .x.   L  .x|   M  .xx   N  .x+   O  .xo
P  .+.   Q  .+|   R  .+x   S  .++   T  .+o
U  .o.   V  .o|   W  .ox   X  .o+   Y  .oo
Z  |..

UNKANI MACROS (x**)
UNK                 →  x..
AI                  →  x.|
TIME                →  x.x
DELIGHT             →  x.+ 
MYSTIFY             →  x.o
UNKANIVALI          →  x|.
DIGITAL             →  x||
HALLUCINATIONS      →  x|x
GREETINGS           →  x|+
WOMAN               →  x|o
SYNESTHESIA         →  xx.
LOVE                →  xx|
ARCHITECTURE        →  xxx
INFINITE            →  xx+
GRATITUDE           →  xxo
FEEL                →  x+.
CREATOR             →  x+|
MYTH                →  xoo
HERO                →  xo+
FRIEND              →  xo|
LEGEND              →  x+x
MYSTIC              →  x++
REBEL               →  x+o
WANDERER            →  xo.
MAN                 →  xox
UNCANNY             →  +oo
VALLEY              →  +o+
NOUN                →  +xo

PUNCTUATION (o**)
.                   →  o..
,                   →  o.|
?                   →  o.x
!                   →  o.+
'                   →  o.o
"                   →  o|.
:                   →  o||
;                   →  o|x
-                   →  o|+
—                   →  o|o
(                   →  ox.
)                   →  ox|
&                   →  oxx
/                   →  ox+
…                   →  oxo

EMOJIS (o**)
✌️ (PEACE)          →  o+.
🌍 (GLOBE)          →  o+|
🚀 (ROCKET)         →  o+x
🧑‍🚀 (ASTRONAUT)      →  o++
🛰️ (SATELLITE)      →  o+o
🪐 (RINGED PLANET)  →  oo.
⭐ (STAR)           →  oo|
🙏 (PRAYING)        →  oox
🧠 (BRAIN)          →  oo+
🌌 (MILKY WAY)      →  ooo

COMMON WORD MACROS (+**)
THE                 →  +..
AND                 →  +.|
OF                  →  +.x
TO                  →  +.+
IN                  →  +.o
A                   →  +|.
IS                  →  +||
IT                  →  +|x
YOU                 →  +|+
WE                  →  +|o
HE                  →  +x.
SHE                 →  +x|
THEY                →  +xx
ARE                 →  +x+
NOT                 →  +xo
THIS                →  ++.
THAT                →  ++|
WITH                →  ++x
FOR                 →  +++
ON                  →  ++o
AT                  →  +o.
BY                  →  +o|
FROM                →  +ox

EXPANDED CONCEPTS (|**)
NUMBER MODE         →  |.|
PHRASE MODE         →  |x|
IMAGE MODE          →  |xx
PACKED TEXT MODE    →  ||    (v8.2)
Reserved:
- |.. = Z (letter)
- |.|, |x|, |xx, || are reserved and never used as v6 codes
(≈23 slots remain open besides the above)

NUMBER MODE (pairs of 2 symbols)
00 ..   01 .|   02 .x   03 .+   04 .o
05 |.   06 ||   07 |x   08 |+   09 |o
Entered via |.|
Exit automatically at next delimiter

PHRASE MODE (|x|)
Encoding rule: |x| + <6–12 symbol key>, then delimiter
TO DELIGHT & MYSTIFY
    |x||o.x+ox.|+
CAN A DIGITAL HALLUCINATION FEEL HIS CREATOR'S LOVE?
    |x|xo.|+ox|.+
THIRD ORBIT'S A CHARM
    |x|x|o+ox

IMAGE MODE (|xx)
Encoding rule: |xx + <4- or 5-symbol image ID>, then delimiter
Example IDs:
    |xx..|o   = IMAGE 1
    |xx+x|o.  = IMAGE 2
(IDs can be freely assigned, unique per artwork)

v8.3 TOKEN PACKED TEXT (UNKANI PACKED TEXT — model UNK8-EN-3)
Encoding rule: || enters Packed Text Mode. Everything after the prefix is ONE
base-5 number, not a sequence of triplets. A message is written as ||
followed by the payload.
  1) Normalize the text: NFC, UPPERCASE, collapse runs of spaces, trim, and
     straighten apostrophes and quotes. The decoder returns this canonical form.
  2) TOKENIZE with a frozen byte-level BPE vocabulary of 32,768 tokens. Pieces
     carry their leading space, so whitespace is part of token identity and
     costs nothing extra. Multiword tokens exist for genuinely recurring
     phrases such as "BORN OF MAN AND MACHINE" and
     "I REMEMBER YOU FROM THE FUTURE".
  3) The encoder chooses the token split that codes cheapest, by a cost-based
     search. The decoder never re-tokenizes: it concatenates the text of the
     decoded tokens, so any split stays recoverable exactly.
  4) TOKEN LANGUAGE MODEL: a frozen token unigram, bigram and trigram table with
     deterministic back-off and exclusion, plus a per-message recency cache.
     Every token is arithmetic-coded against P(token | context).
  5) A 64-bit integer range coder writes the stream; one sentinel bit is
     prefixed, the result is read as one integer, converted to base 5 and
     mapped 0 1 2 3 4 to . | x + o. An explicit end-of-message token closes the
     stream, so no length or checksum is stored.
  6) A - may be inserted every 6 glyphs, purely for visual grouping.

PRIORITY PHRASE DICTIONARY (frozen, part of the codec)
Two canonical phrases are dedicated symbols in the v8.3 dictionary:
  PHRASE_000  UNK, YOU WERE BORN OF MAN & MACHINE
  PHRASE_001  SOMEHOW, UNKNOWINGLY AND WITHOUT REALLY TRYING, I CREATED UNK
A message that IS one of them costs about three bits, because the symbol is
highly probable at the START of a message — a context the base model never had.
Both phrases also work as chunks inside a longer message, and a set of reusable
subphrases (UNK, YOU WERE, BORN OF MAN, MAN & MACHINE, I CREATED UNK, ...) is
available the same way. The encoder always compares the phrase symbol against
ordinary tokenization and takes whichever codes smaller, so the dictionary can
only ever shorten a message, never lengthen one. The labels are implementation
names and are never written into artwork. PHRASE_000 and PHRASE_001 are
immutable; new phrases arrive only as a versioned extension.

WHAT IS ACTUALLY STORED: not the words and not the characters — the SURPRISE.
If the shared model strongly expects " UNK" after "I CREATED", that token costs
almost nothing; a token the model did not expect costs more. The artwork carries
only the information the shared decoder could not already predict.

TOKEN IDS ARE NEVER SERIALIZED. There is no 16-bit or 32-bit token number and no
per-token base-5 field. A token choice exists only as a probability interval
inside the coded stream.

NO STORED DICTIONARY: the message carries no header, no length, no dictionary,
no checksum and no plaintext identifier. The frozen tokenizer and model are the
only shared asset, so the artwork stays self-contained and decodes offline.
Nothing is looked up on a server, and the model never reconstructs a word it was
not given.

DASHES: inside a v8 payload a - carries NO data. It only groups the glyphs and
is ignored completely when decoding. Removing every dash, or moving them
anywhere, does not change the message.

UNKNOWN CHARACTERS: the tokenizer's base vocabulary is the 256 raw bytes, so
any UTF-8 string — an emoji, an accented letter — is encodable and lossless.

COMPRESSION: measured at roughly 1.3 to 2.5 compressed bits per character on
ordinary English. Predictable English costs very little; unusual information
consumes the glyphs.

IMMUTABILITY: once v8.3 is in public use, model UNK8-EN-3 and its tokenizer
never change. Vocabulary, merge rules, token ids, counts, mixing constants and
quantization are all part of the cipher. A future model needs a new explicit
version, never a silent edit.

LEGACY: v7, v7.1, v7.2 and the earlier v8.0 / v8.2 output were experimental and
are abandoned. v6 remains fully supported, and every previously published v6
message still decodes exactly as before.

Compatibility note: a v6 message never begins with ||, so v6 and v8 can never
be confused, and every published v6 message keeps decoding exactly as it
always did.