An independent research project in Compositional Data Analysis (CoDa) and entropy-invariant monitoring on the simplex. This is a scientific research repository — not a game engine, not a Unity plugin, not a software library.
Author: Peter Higgins, Rogue Wave Audio, Markham, Ontario, Canada
Core discovery: The Entropy-Invariant Time Transformer (EITT) — Shannon entropy of compositional time series is empirically near-invariant under geometric-mean decimation. Measured at 0.18% variation across a 341:1 compression ratio. Validated across energy systems, chemistry (500,000 data points), hardware reliability, and climate scenarios.
Disciplines: Compositional Data Analysis, Shannon entropy, Aitchison geometry, simplex monitoring, quantum information correspondence, energy transition analysis.
Conference: CoDaWork 2026, Coimbra, Portugal (June 2026). Abstract page 25.
HUF proposes that composition — the internal proportional balance of a system's parts — can be monitored as a primary observable alongside magnitude, identity, and trend. The instrument reads. The human expert decides. The loop stays open.
The framework builds on Aitchison (1982/1986) simplex geometry and Shannon (1948) entropy, applying them to longitudinal monitoring of any system where parts share a conserved whole: energy grids, chemical mixtures, drive fleets, financial portfolios, wetland ecosystems.
Shannon entropy appears empirically near-invariant under geometric-mean block decimation of compositional time series. This is not a theorem — it is an empirical observation awaiting formal proof.
Measured: 0.18% variation across a 341:1 compression ratio (daily to annual European electricity compositions, 8 carriers, 4089 trading days). Confirmed independently on EMBER monthly generation data (6 countries, mean 1.02%, all below 2%) and NGFS Phase 4 climate scenarios (35 scenarios, all below 5%).
April 2026 — Chemistry extension. EITT tested on 500,000 chemical mixture data points (CheMixHub benchmark, 7 datasets). Four diagnostic lenses applied simultaneously (Shannon, Jensen-corrected, Renyi q=2, Aitchison norm). Interior compositions pass at 54-82%. Boundary compositions reveal simplex curvature effects. First empirical decomposition: approximately 50% of the invariance comes from Aitchison geometry, approximately 50% from temporal autocorrelation.
This produced three new frameworks:
| Framework | What It Does | Document |
|---|---|---|
| EITT Findings | Raw science. Four-lens results, failure taxonomy, multi-modal simplex | science/eitt/ |
| HUF-IDX | Development index. What residuals mean. Domain distance from ground zero | science/eitt/ |
| PRISM | Operational layer. Ranked resource allocation targets from residual analysis | science/eitt/ |
Posture: We found this empirically. We can't prove it. Can you?
| Category | Name | Question | Status |
|---|---|---|---|
| MC-1 | Magnitude Monitoring | How much? | Universally deployed |
| MC-2 | Identity Monitoring | Who or what? | Universally deployed |
| MC-3 | Trend Monitoring | Which direction? | Universally deployed |
| MC-4 | Composition Monitoring | What is the balance? | Proposed (HUF) |
HUF/
├── ai-refresh/ # AI STARTS HERE — fast context loading
│ ├── HUF_FAST_REFRESH.json # All canonical values, single file
│ ├── HUF_INTEGRITY_MANIFEST.json # Hash verification, drift patterns
│ └── huf_spec_v2.0.json # Complete framework specification
│
├── science/ # All scientific work by subject
│ ├── reference/ # Core reference (9 docs, ~105 min)
│ ├── core/ # EITT maths, formulas, user handbook
│ ├── methodology/ # Governance scale, confidence index
│ ├── chemistry/ # CheMixHub results, HUF-IDX, PRISM
│ ├── quantum/ # HUF-QIT: 9 isomorphisms, Bell test
│ ├── eitt/ # EITT evidence: 4 proofs, adversarial
│ ├── coda-monitoring/ # Perturbation drift detection protocol
│ ├── spectral/ # Frequency-domain analysis
│ ├── loudspeaker-analogy/ # Origin: crossover networks as CoDa
│ ├── wetlands/ # Ramsar Convention monitoring
│ └── governance/ # 201-country ranking outputs
│
├── huf-gov/ # Active governance
│ ├── governance/ # Standards, protocols, kill tests
│ ├── science/ # Monitoring taxonomy, ontology
│ └── evidence/ # Case studies (energy, backblaze, etc.)
│
├── tools/ # Everything runnable
│ ├── pipeline/ # EITT pipeline, preparsers, scripts
│ ├── diagnostics/ # Validators, dashboards, UML diagrams
│ ├── spectrum-analyzer/ # HUF Spectrum Analyzer (all versions)
│ └── shared/ # Build utilities, styles, glossary
│
├── drafts/ # Conference materials, papers, proposals
│ ├── codawork-2026/ # CoDaWork 2026, Coimbra
│ │ ├── presentation/ # Main talk slide deck
│ │ ├── primers/ # 4 personalized researcher primers
│ │ ├── preparation/ # Q&A prep, conversation guide
│ │ └── extended/ # Extended results, EITT handout
│ ├── papers/ # Paper submissions (7 venues)
│ ├── proposals/ # Unified proposal, integration briefs
│ └── books/ # Book-length QIT treatments
│
├── briefings/ # Session briefings & AI handoffs
│
├── data/ # Datasets by domain
│ ├── backblaze/ # Hard drive failure data
│ ├── energy/ # EMBER/OWID energy data
│ ├── ember/ # EMBER processed results
│ ├── ngfs/ # NGFS Phase 4 scenarios
│ ├── codawork-samples/ # Reproducible CoDaWork samples
│ ├── eitt-lab/ # EITT lab package
│ └── toronto/ # TTC transit data
│
├── dormant/ # Paused branches — sleeping, not dead
│ ├── pre-coda-metrics/ # Pre-CoDa metric formulations
│ ├── early-governance/ # Multi-AI collective experiments
│ ├── planck-case/ # Planck sky map analysis
│ ├── peterson-outreach/ # Peterson letters (paused)
│ ├── hagf/ # Adaptive governance (superseded)
│ └── deceptive-drift/ # Arithmetic hiding composition changes
│
└── archive/ # Superseded work — what failed speaks loudest
Start here: Read ai-refresh/HUF_FAST_REFRESH.json first. It contains every canonical name, number, formula, and structural rule. If anything elsewhere disagrees with FAST_REFRESH, the FAST_REFRESH wins. Then verify with ai-refresh/HUF_INTEGRITY_MANIFEST.json. Then read INDEX.json for the full file map.
Context aggregator: Run python build_context.py --mode seed to generate a single paste-ready text file containing the full AI seed layer. Modes: seed (~9k tokens), science (~232k tokens), full (~346k tokens).
Known drift traps: EITT is "Entropy-Invariant Time Transformer" (never Ternary). Japan drift flag is 2013-2014 (never 2011-2012). Germany is 2023-2024/2024-2025 (never 2011). UK has three specific values (2.98, 3.23, 3.26), never "approximately 3".
| Time | What | Where |
|---|---|---|
| 5 min | The cylinder problem and fuel gauge — one-page HUF | science/reference/ |
| 105 min | Reference Collection — 9 documents, full learning path | science/reference/INDEX.md |
| 15 min | User Handbook — fast summary + guided links | science/core/HUF_USER_HANDBOOK.md |
| 20 min | Chemistry results (the new frontier) | science/chemistry/ |
| 30 min | The kill test — 19 documented failure modes | huf-gov/governance/ |
| 10 min | Quantum correspondence (advanced) | science/quantum/ |
If you want to break it, the kill test is where to start.
- Composition can be monitored directly, not only as a statistical correction.
- In some systems, structural change appears in ratio-state before magnitude-based indicators visibly fail.
- This can be tested across any domain where a conserved whole divides into meaningful parts.
- New simplex mathematics. The foundations are Aitchison (1982), Shannon (1948), Stevens (1946), Amari (1985).
- Universal validity. Cross-domain validity must be earned domain by domain.
- That every compositional change is harmful, actionable, or predictive.
- That HUF replaces domain expertise, causal explanation, or policy judgment.
- That autonomous intervention is justified on compositional readings alone.
Nothing dies here — only goes dormant. The dormant/ folder preserves paused work with documented reasons and conditions for reawakening. The archive/ folder holds superseded and rejected approaches as reference for what was tried and why. What failed always speaks louder.
Standard: RWA-001 (Rogue Wave Audio Corporate Reference)
Protocol: HUF-GOV. Measure, report, file. No intervention on the data.
Multi-AI Review: All core findings subjected to adversarial review by Claude, ChatGPT, Grok, Gemini, and Copilot.
MIT. See LICENSE.
See CITATION.cff.
Peter Higgins | Rogue Wave Audio | PeterHiggins@RogueWaveAudio.com
Repository: github.com/PeterHiggins19/Higgins-Unity-Framework