ASCENT / 1.0 draft

MIT

July 2026

Ascent.Grok.Me

Keep every classic ASCII byte sacred.

ASCENT — ASCII Successor with Compatible Encoding of Named Text — is a wire contract greppers can still trust. Not “Unicode but better.” Agent directives, multimodal refs, quantum-safe control, and deep-space frames ascend without burning the stairs.

Open the wire lab Read the spec Official site

Proposed by Jon Bailey (Pitchfork-and-Torch) · eternal identity plane 0x00–0x7F · C1 fences · length-prefixed control · ASCENT-V data plane

Eternal parse law

If a byte is less than 0x80, it is classic ASCII — forever.

Width 1. Bit-identical. Overlong encodings of the sacred range are illegal. Control is fixed-width or length-declared; data is variable-width; private planes cannot redefine the core (P0–P2).

Any pure classic ASCII file is valid ASCENT-7
Higher planes are invisible or safely ignorable to legacy tools
No multi-byte remapping of control bytes inside P0

IDENTITY PLANE

128 sacred slots · 0x00–0x7F · the stairs you still climb every day

# design motto
Keep every classic ASCII byte sacred.
Everything else ascends without burning the stairs.

Origin & motivation

Why a successor that refuses to rewrite 1963

ANSI X3.4-1963 still underpins shebangs, SMTP 7-bit paths, greppers, and simple tools. Extended ASCII variants fractured; scripts and emoji grew; LLM agents need in-band control; deep-space links demand outer ECC. ASCENT keeps the identity plane and climbs only through declared fences.

01
Preserve the 1963 identity plane that still underpins foundational tools.
02
Carry LLM/agent control tokens in-band without overloading printable prose.
03
Attach multimodal references and short payloads with content hashes.
04
Ship quantum-safe control sequences via an append-only algorithm registry.
05
Support long-haul links with an outer ECC profile.
06
Remain self-describing so documents embed plane maps and version history.

Eleven planes

Architecture that greppers can still trust

P0–P2 are immutable core. P3+ add capability. Private planes cannot touch the sacred shell.

0x00–0x7F

Identity

Classic 7-bit ASCII. Width 1. Bit-identical forever. Sacred.

0x80–0x9F

C1 fences

Fixed control fences: agent open/close, crypto, multimodal.

0xC0–0xCF

Escape nucleus

Length-prefixed ops and DEF schemas. Control stays declared.

P3/P4

ASCENT-V

Scripts & emoji

Variable-width data plane. No surrogates. No overlong sacred range.

fenced TLV

Agent frames

STOP, ROLE, TOOL, THINK, HANDOFF, CAP, SAFETY — first-class opcodes.

REF / INLINE / CHUNK

Multimodal

Content-hashed media references and stream units.

0x9C + registry

Quantum-safe

Append-only algorithm IDs. No silent downgrade.

DEF + history

Self-map

Plane maps, registries, and version hash chains in-stream.

outer ECC

Deep-space

Sync D5 E5 C0 DE. RS(255,223). Erase-on-fail, not mojibake.

P10+

local only

Private

Escape hatch for experiments. Cannot redefine P0–P2.

Four profiles

Same grammar. Orthogonal constraints.

ASCENT-7

Classic identity

Pure P0. Every classic ASCII file is already valid ASCENT-7.

ASCENT-E

Earth interchange

Full features for general terrestrial traffic and agent runtimes.

ASCENT-D

Deep-space

Mandatory outer ECC, prefer REF, algorithm freeze, erase-on-fail.

ASCENT-A

Where a greppable wire contract actually matters

Agent-native control

ROLE, TOOL, THINK, HANDOFF, CAP, and SAFETY as fenced, length-declared units — not ad-hoc tokens that greppers and injection defenses cannot trust.

›Pure-ASCII consumers see only the text portion
›Unknown opcodes with valid length are skippable
›Nested frames forbidden in v1 for simpler auditors

Eternal backward compatibility

Stronger than “ASCII is a subset of UTF-8”: multi-byte encodings of the sacred range are illegal, and P0 control bytes are never remapped.

›Shebangs, SMTP 7-bit paths, grep/awk still work
›Optional greppable header ASCENT/1.0
›UTF-8 treated as bridge for import/export only

Multimodal references

REF, INLINE, CHUNK, and END units carry kind, codec, flags, hash algorithm, content hash, and length in one ordered stream.

›Content-addressed verification and dedup
›Deep-space profiles prefer REF over INLINE
›Unknown codecs skipped by length

Deep-space links

Profile D wraps units in P9 frames with a long sync word and Reed-Solomon ECC. On parity failure the unit is erased — fail closed.

›Default RS(255,223) over GF(256), interleave 8
›Algorithm freeze mid-cruise for probes
›Beacons for acquisition under high BER

Quantum-safe control

Crypto markers reference an append-only algorithm registry. Unknown algorithms are skipped by length; silent downgrade is impossible.

›ML-KEM / ML-DSA and future registry entries
›Key id, nonce, and ciphertext in-stream
›Profiles may freeze the algorithm set

Self-describing archives

DEF frames embed plane maps, annex hashes of normative prose, and history hash-chain deltas so documents carry their own decoder seed.

›Bootstrap: greppable header + fixed P0–P2 shell
›ASCENT-A encourages full normative embedding
›Reduces dependence on vanished external specs

Interactive demo

Encode a stream. Watch the sacred meter.

Compose greppable headers, agent fences, and P0 prose. Hover bytes to inspect planes. This lab is educational — implementers should consult the normative SPEC for production decoders.

Wire lab

illustrative · Earth subset

not a normative decoder — see SPEC.md

Prose (P0)

Agent opcodeOpcode argument

Prefixed greppable header ASCENT/1.0\n

Total

81B

Sacred P0

79B

Control

Sacred meter97.5%

Byte map

P0 identityP1 fenceP2 / data

Hex dump

0000  41 53 43 45 4e 54 2f 31 2e 30 0a 9a 01 02 00 00  |ASCENT/1.0......|
0010  09 61 73 73 69 73 74 61 6e 74 9b 48 65 6c 6c 6f  |.assistant.Hello|
0020  2c 20 55 6e 69 76 65 72 73 65 2e 20 4b 65 65 70  |, Universe. Keep|
0030  20 65 76 65 72 79 20 63 6c 61 73 73 69 63 20 41  | every classic A|
0040  53 43 49 49 20 62 79 74 65 20 73 61 63 72 65 64  |SCII byte sacred|
0050  2e                                               |.|

Greppable header · 11B

Agent ROLE · 16B

P0 prose · 54B

Comparative view

Not a dual-stack compromise

vs UTF-8 — ASCENT sacrifices single-encoding convenience for scripts in exchange for stronger identity guarantees and first-class control planes.

vs ad-hoc LLM tokens — in-band, length-declared, greppable structure with explicit fences and SAFETY semantics instead of brittle side-channels.

vs pure binary — human-readable ASCII remains a first-class citizen on the wire.

Agent fence sketch

0x9A          AGENT_OPEN
0x01          version
0x02          opcode ROLE
0x00          flags
0x00 0x09     arg length
61 73 73 ...  "assistant"
0x9B          AGENT_CLOSE

# Unknown opcodes with valid length
# are skippable (unless CRITICAL).
# Nested frames forbidden in v1.

Deep-space sync word D5 E5 C0 DE · default ECC RS(255,223) · interleave 8 · erase-on-fail

Status & resources

1.0 draft under MIT. Stop arguing about the stairs.

As of 29–30 July 2026 the specification is a public draft with a live wire lab, Python reference decoder (Earth-profile subset), and example binaries. Governance is sketched as an ASCENT Joint Maintenance Group dual-liaising with ISO/IEC JTC 1 and the IETF. Registries stay public and secret-free.

Normative specification

Pitchfork-and-Torch/ascent

Official wire lab

ascent.jonbailey.xyz

Primary lasting bet: a modern agent- and multimodal-aware encoding can still honor the 1963 identity plane. Whether ASCENT becomes widely adopted, a specialized deep-space format, or influences future control grammars depends on implementer uptake of skip-unknown, erase-on-fail, and append-only registry properties.

“Stop arguing about the stairs and start climbing them.”