Skip to content

SPECTRAL v2.0.0 — Deterministic Sonification Workbench & Audible Geometry Platform

Choose a tag to compare

@EmergentMonk EmergentMonk released this 11 Jul 11:07
· 96 commits to main since this release

SPECTRAL v2.0.0

SPECTRAL v2.0.0 is a major architectural and research release.

The centrepiece is the new SPECTRAL Deterministic Sonification Workbench 2.0: a zero-install, browser-native application for transforming data, symbolic systems, images, binary streams, and QEC-style event streams into reproducible WAV source stems with deterministic provenance.

The repository has also expanded substantially across E8-inspired synthesis, astronomical sonification, receipt-bound information-lattice sonification, OMI/XOR experiments, photoacoustic modelling, visualization, research datasets, and publication support.

Compared with v1.0.0, this release contains 84 commits and introduces five major project areas:

APP/
E8/
OMI-ISA/
PHOTOACOUSTIC/
SONIFICATION/
Major addition: browser-native Workbench 2.0

The new APP/ directory provides a complete local DHTML application that runs by opening:

APP/index.html

No installation or local server is required.

The Workbench uses:

no Node.js;
no npm;
no build system;
no bundler or transpiler;
no CDN;
no cloud service;
no telemetry;
no network requests;
and no server-side processing.

The browser is the complete application runtime.

Two determinism modes
Canonical Strict

Canonical Strict is designed for archival, publication, and cross-runtime reproducibility.

It introduces:

signed interleaved PCM16 as the signal ABI;
integer phase accumulators;
explicit fixed-point gains and modulation;
deterministic rounding and clipping;
canonical WAV packing;
bundled bitwise SHA-256;
canonical JSON serialization;
deterministic PRNG operation;
fixed sample-count rendering;
domain-separated recipe, fingerprint, contract, and manifest hashing;
and exclusion of runtime observations from strict render identity.
Replay Safe

Replay Safe provides a deterministic creative workflow while permitting controlled Float64 DSP.

It includes:

explicit PCM16 quantization;
recorded runtime fingerprints;
no hidden entropy;
no mutable engine state;
runtime-scoped replay guarantees;
and mode-separated recipe and contract identities.

Canonical Strict and Replay Safe outputs cannot accidentally share the same SPECTRAL recipe identity.

Initial engine registry

Seven versioned engines are included:

omi_xor_ring
e8_cosmovirus
e8_bell
qec_triality
spectral_algebraics
image_scan
data_mapper

Each engine declares its:

identifier and version;
supported determinism modes;
parameter schema;
presets;
export profile;
claim boundary;
and tone-plan compiler.

The registry is extensible without redesigning the central application.

Data and file inputs

The Workbench can process:

CSV;
TSV;
JSON;
plain text;
pasted numerical text;
arbitrary binary files;
byte streams;
PNG;
JPEG;
and WEBP images.

Input handling now includes:

canonical filename-extension routing;
raw source-byte hashing;
versioned input adapters;
deterministic numeric token extraction;
quoted CSV and TSV support;
JSON-number mapping;
binary-byte mapping;
adapter identity in render recipes;
and separation between compressed image identity and browser-decoded pixel observations.

Browser-native image decoding is intentionally restricted to Replay Safe mode.

Deterministic audio pipeline

The new audio layer provides:

deterministic DSP;
PCM16 stereo synthesis;
explicit WAV serialization;
fixed sample-rate export profiles;
deterministic normalization;
deterministic fades and clipping;
chunked rendering;
cancellation support;
and byte-identical strict rendering when the complete recipe is unchanged.

Three export profiles are included:

Archive — 44.1 kHz PCM16 stereo
Suno Seed Export — 48 kHz PCM16 stereo with conservative headroom
Brutalist — 44.1 kHz PCM16 stereo for aggressive experiments

Suno Seed Export is a technical headroom profile for original material. It is not presented as a detector bypass or upload guarantee.

Provenance and reproducibility system

Every Workbench render can produce:

audio.wav
manifest.json
fingerprint.json
upload_readiness.json
derivation_graph.json
README_ORIGIN.txt

These artifacts can also be packaged into a deterministic, uncompressed ZIP archive.

The ZIP implementation provides:

lexically sorted entries;
fixed DOS epoch fields;
no comments;
no extra fields;
explicit CRC-32 values;
and reproducible bundle bytes.
Render recipes and observation contracts

Identity-bearing render recipes now preserve:

application version;
engine version;
determinism mode;
parser and canonicalization versions;
DSP, PRNG, and WAV-writer versions;
normalized engine parameters;
export-profile configuration;
seed and mutation index;
duration in integer samples;
source hash and byte length;
and Replay Safe runtime identity where applicable.

Observation contracts record:

recipe hashes;
source hashes;
PCM payload hashes;
canonical WAV hashes;
fingerprint hashes;
backend and ABI;
sample format;
replay requirements;
and engine claim boundaries.

The pipeline avoids circular self-hashing by separating the manifest core from its final envelope.

Manifest replay

A new fail-closed replay workflow regenerates audio from an exported manifest.

Replay requires:

a supported manifest schema;
the exact engine version;
matching source bytes when a source was used;
a matching regenerated WAV hash;
and a matching regenerated contract hash.

A replay is accepted as a persisted job only after verification succeeds.

Audio fingerprinting and readiness analysis

The deterministic local fingerprint now includes:

exact content hashes;
peak level;
RMS level;
DC offset;
zero-crossing rate;
stereo correlation;
clipping-sample count;
PCM chunk hashes;
repeated exact-chunk observations;
and RMS-envelope hashes.

Upload readiness provides offline advisory checks for:

clipping;
excessive or insufficient peak level;
DC offset;
exact repeated chunks;
aggressive export settings;
and provenance-bundle completeness.

Local replay detection remains outside deterministic artifacts so browser history cannot change exported identities.

Visualization and playback

The Workbench introduces:

interactive Canvas waveform rendering;
wheel zoom;
drag panning;
selection ranges;
FFT spectrogram rendering;
256, 512, 1024, and 2048-point windows;
linear-frequency display;
logarithmic-frequency display;
local audio playback;
seeking;
looping;
and two in-memory A/B comparison slots.

Visualization settings are observational only and cannot modify generated PCM.

Derivative lineage

SPECTRAL v2.0.0 adds explicit derivative-lineage documentation.

Users can:

preserve an immutable source-render contract;
hash a later derivative locally;
declare the relationship between seed and derivative;
document authorship and transformation notes;
update derivation_graph.json;
embed lineage into the manifest envelope;
and rebuild the deterministic provenance package without pretending the derivative existed at seed-render time.
Local persistence and privacy

The browser application uses IndexedDB to retain:

completed jobs;
presets;
manifests;
contracts;
fingerprints;
canonical WAV files;
and render history.

Additional behaviour includes:

retention of the most recent 24 jobs;
eviction of older jobs before new writes;
in-memory fallback when local-file storage is unavailable;
reconstruction of PCM and ZIP data from stored canonical WAVs;
and continued playback and download when persistence fails.

All source material remains local.

The application creates no cookies and contains no production network client.

Browser test suite

A browser-native known-answer suite is included at:

APP/tests/index.html

Coverage includes:

SHA-256 standard vectors;
UTF-8 encoding;
canonical JSON;
decimal scaling and rounding;
deterministic PRNG vectors;
exact PCM16 WAV construction;
deterministic ZIP output;
engine-registry validation;
strict-mode image safety;
repeated-render equality;
seed and mutation identity changes;
determinism-mode separation;
manifest replay;
receipt-layer tamper rejection;
changed-source rejection;
lineage-manifest replay;
CSV, JSON, text, and binary golden vectors;
cancellable rendering;
browser-history isolation;
IndexedDB persistence;
actual page-reload persistence;
and absence of remote resource loads.
E8 research expansion

The E8/ collection adds new E8-inspired sonification generators, research inputs, demonstration renders, and supporting artifacts.

New generators include:

E8 triality fractal sonification;
Bullet Cluster E8 sonification;
M87 E8 fractal sonification;
simplified M87 sonification;
M81/Hubble sonification;
NGC 2525 sonification;
QEC v167 symbolic sonification generation;
and extended QSOL source preparation.

New research and demonstration material includes:

M81 Hubble WAV and MP3 renders;
M87 full and shortened E8 renders;
Bullet Cluster source imagery and audio;
NGC 2525 audio;
triality-standard renders;
qutrit-4D triality renders;
trinaural triality renders;
QEC v167 MIDI, WAV, JSON, and visualization artifacts;
collective-modes source audio;
molecular sonification examples;
terpene-entourage sonification;
astronomical source imagery;
and associated scientific reference material.
OMI-ISA experiments

A new OMI-ISA/ research area introduces:

OMI/E8 XOR-ring sonification;
deterministic XOR and rotation-state mapping;
E8 fractal-cosmovirus sonification;
Ouroboros-style feedback structures;
demonstration WAV files;
and symbolic engines later incorporated into the browser Workbench.
Photoacoustic sonification

A new PHOTOACOUSTIC/ project introduces a deterministic, simplified photoacoustic sonifier and music generator.

It models:

modulated light intensity;
absorption and thermal inertia;
temperature change;
thermal expansion and pressure generation;
cavity or resonator filtering;
and optional post-processing.

It supports:

WAV-driven light modulation;
synthetic light sources;
tones and chords;
pulse trains;
Fibonacci rhythms;
industrial sequences;
selectable material behaviour;
resonator presets;
distortion;
reverb;
EQ;
and deterministic NumPy/SciPy processing.
Receipt-bound sonification research

The new SONIFICATION/ area contains implementation and media associated with:

Collective Modes, Stable Supports, and Observation-Dark Subspaces in Deterministic Information Lattices: Graph Dynamics and a Reproducible Protocol for Receipt-Bound Sonification

Included material comprises:

the v1.2 sonification script;
a 30-second deterministic prototype;
tagged paper sonification audio;
the Collective Modes [Dark Subspaces] video;
and the Recursive Receipt video.

Associated Zenodo records:

v1.2: https://doi.org/10.5281/zenodo.21293821
v1.0: https://doi.org/10.5281/zenodo.21292906
Publication integration

The repository now formally incorporates:

Slade, T. (2026). Spectral Algebraics: Audible Geometry via E8-Inspired Signal Synthesis and 3D Visualization [Dataset]. Zenodo.

https://doi.org/10.5281/zenodo.21308248

The main README now includes:

DOI badges for all three archived records;
complete formatted citations;
BibTeX entries;
ORCID information;
direct project-area navigation;
reproducibility guidance;
claim boundaries;
and clear relationships between repository materials and publications.

Zenodo metadata and citation-maintenance tooling have also been updated.

Licensing change

SPECTRAL is now distributed under the MIT License.

This permits use, copying, modification, merging, publication, distribution, sublicensing, sale, commercialization, and incorporation into software, scientific instruments, embedded systems, manufactured products, and physical devices, provided that the required copyright and permission notices are retained.

MIT licensing is now aligned across:

the root LICENSE;
APP/LICENSE.txt;
the main README;
the APP README;
.zenodo.json;
and citation-maintenance tooling.
Repository organization

SPECTRAL v2.0.0 establishes the following primary structure:

SPECTRAL/
├── APP/ Browser-native deterministic sonification workbench
├── E8/ E8-inspired generators, datasets, and audio
├── OMI-ISA/ OMI/XOR/E8 symbolic sonification experiments
├── PHOTOACOUSTIC/ Photoacoustic modelling and synthesis
├── SONIFICATION/ Receipt-bound sonification research artifacts
├── LICENSE MIT License
└── README.md Project overview, citations, and documentation
Breaking and architectural changes

This is a major-version release because SPECTRAL is no longer only a collection of standalone Spectral Algebraics experiments.

Important changes include:

the browser-native Workbench becoming the primary interactive application;
formal separation between Canonical Strict and Replay Safe rendering;
versioned engine and input-adapter registries;
manifest- and contract-based artifact identity;
a new provenance bundle format;
local IndexedDB persistence;
expanded repository organization;
updated publication relationships;
and migration to the MIT License.

Existing standalone Python experiments remain available, but their dependencies and reproducibility guarantees are script-specific.

Getting started

Clone the repository:

git clone https://github.com/QSOLKCB/SPECTRAL.git
cd SPECTRAL

Launch the browser Workbench by opening:

APP/index.html

Run its browser-native test suite by opening:

APP/tests/index.html

Python experiments can be run individually after installing their script-specific dependencies.

Citation

If you use the general SPECTRAL and E8-inspired audible-geometry material, cite:

Slade, T. (2026). Spectral Algebraics: Audible Geometry via E8-Inspired Signal Synthesis and 3D Visualization [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21308248

For the deterministic information-lattice and receipt-bound sonification material, cite v1.2:

Slade, T. (2026). Collective Modes, Stable Supports, and Observation-Dark Subspaces in Deterministic Information Lattices: Graph Dynamics and a Reproducible Protocol for Receipt-Bound Sonification (Version v1.2). Zenodo. https://doi.org/10.5281/zenodo.21293821