Skip to content

SPEC-v0.1.md

Latest commit

 

History

History
324 lines (220 loc) · 12.4 KB

SPEC-v0.1.md

File metadata and controls

324 lines (220 loc) · 12.4 KB
Raw

AeonScript — Specification v0.1

Status : Draft Date : May 2026 License : CC-BY-SA 4.0

0. Conventions

  • MUST, MUST NOT, SHALL, SHOULD, MAY follow RFC 2119.
  • All sequence examples are written 5' → 3' unless otherwise noted.
  • All bit-level encodings are little-endian unless otherwise noted.
  • Byte-strings are denoted 0xAB (hex), bit-strings 0b1010, ASCII strings in "quotes".

1. Architectural overview

AeonScript is structured as seven independent layers, modeled on the OSI reference model:

┌─────────────────────────────────────────────┐
│ L7 — Application       (file-type formats)   │
├─────────────────────────────────────────────┤
│ L6 — Presentation      (compression, crypto) │
├─────────────────────────────────────────────┤
│ L5 — Session           (semantic tags)       │
├─────────────────────────────────────────────┤
│ L4 — Transport         (error correction)    │
├─────────────────────────────────────────────┤
│ L3 — Network           (PCR random access)   │
├─────────────────────────────────────────────┤
│ L2 — Data Link         (strand/frame mux)    │
├─────────────────────────────────────────────┤
│ L1 — Physical          (alphabet, encoding)  │
└─────────────────────────────────────────────┘

Each layer MUST be independently versionable. A change to L1 (e.g. adding 5-methylcytosine as a 5th symbol) MUST NOT break L5+.

This specification v0.1 fully defines L1, L4, L5. L2, L3, L6, L7 are described at the interface level only and reserved for v0.2+.

2. Layer 1 — Physical

2.1 Alphabet

A conformant L1 implementation MUST support the canonical 4-letter alphabet:

Σ₄ = { A, C, G, T }

It MAY declare support for the extended alphabets:

Profile Alphabet Bits/symbol Status
L1-4 {A, C, G, T} 2.00 Canonical, MUST support
L1-5m {A, C, G, T, 5mC} 2.32 Optional, MAY support
L1-6 {A, C, G, T, 5mC, 5hmC} 2.58 Experimental

The active profile MUST be declared in the L5 session tag (see §4).

2.2 Base-to-bit mapping (canonical, profile L1-4)

Base 2-bit code
A 00
C 01
G 10
T 11

Bytes are encoded LSB-first into bases. Example:

Byte 0x4A = 0b01001010
            └┘└┘└┘└┘
             10  10 01 01     (LSB→MSB pairs)
             G   G  C  C
→ "GGCC"

A conformant decoder MUST parse bases right-to-left within a byte: the first base encodes the least significant bit pair.

2.3 Biochemical constraints

To remain synthesizable and sequenceable, an L1-encoded oligo MUST satisfy:

  1. Homopolymer limit : no run of identical bases > 3. Example: AAAA is forbidden. AAA is allowed.
  2. GC content : between 40 % and 60 % over any 50-base window.
  3. No forbidden motifs : the conformant set MUST include at minimum:
    • GGGGG (G-quadruplex risk)
    • ATATATATAT (secondary structure risk)
    • Any motif declared in L5 tag forbidden_motifs

If raw 2-bit encoding violates a constraint, the encoder MUST apply a mapping permutation (different base assignment per byte) until the constraint is satisfied. The permutation index is encoded as a 2-bit prefix per oligo (24 possible permutations of 4 bases — 4 bits suffice; v0.1 reserves 4 bits and uses the lowest 5 permutations).

2.4 Oligo length

A standard AeonScript oligo:

  • MUST be between 100 and 300 bases (current synthesis sweet spot)
  • SHOULD be 200 bases by default
  • The length is declared per-block in the L5 tag

3. Layer 4 — Transport / Error correction

3.1 Reed-Solomon over GF(256)

L4 v0.1 mandates Reed-Solomon codes over GF(256) for inner-oligo error correction.

Default parameters:

Parameter Value Meaning
Symbol size 8 bits 1 byte
Codeword size (n) 255 Max RS block
Data size (k) 223 Effective payload
Parity (n-k) 32 Can correct up to 16 errors

This is the same Reed-Solomon configuration used in CD audio, QR codes, and deep-space communications — chosen for proven reliability and tooling availability.

Encoders MAY declare alternative parameters via the L5 tag field rs_params.

3.2 LDPC outer code (block-level)

For protection against entire oligo losses (e.g. failed PCR amplification of one oligo), L4 v0.1 SHOULD add a systematic LDPC outer code over groups of N oligos.

Default : 1 parity oligo per 16 data oligos (recovers 1 lost oligo per block).

Reserved for v0.2 — v0.1 reference implementation uses simple repetition (each oligo encoded twice with different keys) as a placeholder.

3.3 Methylation hash channel (when L1 profile = L1-5m)

When the L1 profile uses 5mC, the methylation pattern of cytosines in the oligo MUST encode a 64-bit BLAKE3 hash of the unmethylated oligo bytes.

This provides cross-layer error detection : a silent point mutation will change the unmethylated sequence but not the hash → mismatch detected on decode.

4. Layer 5 — Session / Semantic tags

This is AeonScript's most distinctive feature. Every L5 block begins with a semantic tag that describes its own codec.

4.1 Tag wire format

A semantic tag is an ASCII-encoded structured prefix delimited by pipe characters |:

|<KEY>=<VALUE>;<KEY>=<VALUE>;...|

The tag is itself L1-encoded as standard bases (the | and = and ; characters use specific reserved encodings — see §4.5).

4.2 Mandatory keys

Every AeonScript L5 tag MUST include:

Key Type Example Meaning
AEONSCRIPT version 0.1 Spec version this block conforms to
L1 profile L1-4 Physical-layer alphabet profile
L4 codec rs255-223 Error correction scheme
TYPE MIME text/plain Payload content type
LEN bytes 4096 Decoded payload length
ID string 42157 Unique block identifier

4.3 Optional keys

Key Example Meaning
ENCRYPT aes256-gcm Encryption scheme (L6)
COMPRESS zstd Compression scheme (L6)
TIER cold Storage tier hint
CHECKSUM blake3:abc123… Payload checksum
AUTHORS BnF Provenance metadata
LICENSE CC-BY-4.0 Content license
forbidden_motifs GGGG,ATAT Extra L1 constraints
rs_params n=255,k=200 Override RS defaults

4.4 Tag examples

Minimal :

|AEONSCRIPT=0.1;L1=L1-4;L4=rs255-223;TYPE=text/plain;LEN=4096;ID=42157|

Cultural-archive use-case :

|AEONSCRIPT=0.1;L1=L1-5m;L4=rs255-223;TYPE=image/jp2;LEN=10485760;ID=BnF-MS-fr-2810;TIER=cold;AUTHORS=BnF;LICENSE=CC-BY-4.0;CHECKSUM=blake3:e3b0c4...|

4.5 Tag encoding

ASCII characters used in tags map to L1 bases via this lookup:

  • ASCII byte → 8 bits → 4 bases (per §2.2)
  • The special characters | (0x7C), = (0x3D), ; (0x3B) are encoded normally.
  • The encoder MUST ensure the tag does not violate L1 constraints (§2.3) by inserting padding bases between fields if necessary.

A conformant decoder identifies a tag by:

  1. Scanning the start of an oligo for a 4-base pattern matching | (encoded as GGCC, since | = 0x7C = 0b01111100).
  2. Decoding ASCII until a closing | is found.
  3. Parsing key=value pairs separated by ;.

4.6 Reserved key namespace

  • Keys in lowercase are reserved for implementation hints (no compatibility guarantee).
  • Keys starting with x- are vendor extensions (e.g. x-microsoft-cluster=42).
  • Keys in UPPERCASE are standardized by this spec or its successors.

5. Layer 2 — Data Link (multiplexing) [v0.2 placeholder]

Layer 2 introduces multiplexing of multiple logical channels onto the same physical oligo. v0.1 reserves the design space; v0.2 will fully specify.

Planned mechanisms:

5.1 Strand multiplexing

  • Forward strand decodes message A
  • Reverse complement decodes message B
  • Joint encoding via constraint solver

5.2 Reading frame multiplexing

  • Frames 0, +1, +2 each decode independent messages
  • Inspired by overlapping ORFs in phage PhiX174

5.3 Codec routing

  • The L5 tag declares which L2 mode is active
  • Decoder chooses correct reading direction/frame

6. Layer 3 — Network (random access) [v0.2 placeholder]

Layer 3 provides random-access semantics, allowing retrieval of specific blocks without reading the entire archive.

6.1 PCR primer index

  • Each block is flanked by a unique 20-base PCR primer pair
  • A primer-to-tag-ID index (the "filesystem table") is itself stored as a AeonScript block at a well-known primer address
  • v0.2 will specify the index format

6.2 Hierarchical addressing

  • Tags support hierarchical IDs (e.g. BnF/manuscripts/MS-fr-2810)
  • Enables semantic queries like "retrieve all blocks where AUTHORS=BnF and TIER=cold"

7. Layer 6 — Presentation [v0.2 placeholder]

L6 covers:

  • Compression (zstd, brotli, neural codecs)
  • Encryption (AES-256-GCM, ChaCha20-Poly1305)
  • Sérialisation des MIME types complexes

8. Layer 7 — Application [v0.2 placeholder]

L7 defines domain-specific extensions:

  • AeonScript-PDF : PDF/A archival extension
  • AeonScript-FASTA : direct biological sequence storage
  • AeonScript-Wiki : Wikipedia article format
  • AeonScript-IIIF : cultural heritage image format

9. Bio-safety protocol (cross-cutting, MANDATORY)

A conformant AeonScript encoder MUST, prior to outputting any oligo for synthesis :

  1. Concatenate all output oligos into a candidate genome string.
  2. Search the candidate against a hazard database. The minimum reference set is the IGSC (International Gene Synthesis Consortium) Harmonized Screening Protocol lookup.
  3. If any match exceeding 20 base pairs at 90 % identity is found against pathogen-of-concern entries, the encoder MUST refuse to emit those oligos and MUST log the refusal.

A non-conformant encoder that bypasses this step MUST NOT label its output as "AeonScript-Compliant".

The hazard database is published separately at <TBD>/aeonscript-hazard-db and updated quarterly.

10. Versioning policy

  • The spec follows semantic versioning : MAJOR.MINOR.PATCH.
  • A change to a wire format → MAJOR bump.
  • A new optional feature → MINOR bump.
  • A clarification or fix → PATCH bump.
  • A decoder MUST support the version it claims, and SHOULD support all earlier minor versions within the same major version.

11. Open questions for v0.2

  1. Should the methylation hash be BLAKE3 or SHA-3-256? (BLAKE3 chosen tentatively for speed)
  2. Optimal LDPC block geometry for typical oligo loss rates.
  3. Standardized primer-pair vocabulary for PCR addressing (need ~10⁶ orthogonal pairs).
  4. Compression interoperability with neural codecs (which neural codec to standardize?).
  5. Multi-strand encoding constraint solver: greedy vs SAT-based vs neural-search.
  6. Should the spec mandate a specific foundation model for AI-native encoding, or just a model-agnostic interface?

12. Acknowledgments

This specification builds on prior art including :

  • Goldman et al. (2013) — early DNA storage encoding
  • Erlich & Zielinski (2017) — DNA Fountain
  • Press et al. (2020) — HEDGES error correction
  • Microsoft Research / Karin Strauss et al. — cluster codec
  • Brixi et al. (2026) — Evo 2 foundation model

We thank the broader DNA storage and bioinformatics communities for decades of foundational work.

This is a living document. Comments and pull requests welcome at /aeonscript.