AXIOM — AI Certainty Middleware for Industrial Decision Systems

Certainty infrastructure for industrial AI decisions. NICE Challenge 2026 — AI & Communications | Yangtze Delta

Architect's Note on AI Use

AXIOM was designed, architected, and directed by Sheldon K. Salmon. AI tools (including large language models) were used as instruments — the same way a carpenter uses a saw. The intellectual core — the CDI Engine, the Confidence Deployment Index, the four scoring dimensions, the Uncertainty Mass architecture, the Epistemic Debt Score, the graduated model trust system, and the STP audit trail integration — is wholly human‑originated.

AXIOM is not a generative AI output. It is a human‑built reliability infrastructure layer, built on three years of independent research into certainty engineering for AI systems. Every formula, every threshold decision, and every architectural choice in the CDI Engine reflects human intent validated through systematic adversarial red-teaming.

Quick Start

Get the AXIOM CDI Engine running in under a minute.

git clone https://github.com/AionSystem/AXIOM.git
cd AXIOM
pip install -r requirements.txt

Score your first AI decision:

from engine.cdi_engine import CDIEngine

engine = CDIEngine()
engine.register_model(
    model_id="your-model-id",
    calibration_status="UNCALIBRATED"
)

result = engine.score({
    "model_id": "your-model-id",
    "output_confidence": 0.82,
    "input_context": {"process": "quality_control", "line": "A3"},
    "timestamp": "2026-04-02T17:00:00Z"
})

print(result["cdi"])              # 0.71
print(result["validity_status"]) # VALID
print(result["um"])               # 0.28
print(result["action"])           # PROCEED

Note: This runs the CDI Engine locally. The api/ FastAPI server and Supabase sync are separate deployments and will not be active in local mode. All core CDI scoring and audit trail sealing work without them.

For the full API server and enterprise deployment, see Installation & Deployment.

Repository Structure

Key files for evaluators: engine/cdi_engine.py (scoring logic) · engine/uncertainty_mass.py (UM computation) · engine/epistemic_debt.py (EDS scoring) · docs/cdi-formula.md (CDI formula derivation) · concept/NICE-PROPOSAL.md (full NICE Challenge proposal)

AXIOM/
│
├── public/                         ← GitHub Pages landing page
│   ├── index.html                  ← AXIOM concept landing page
│   └── assets/                     ← Images, icons, diagrams
│
├── engine/                         ← CDI scoring engine (core)
│   ├── cdi_engine.py               ← CDI Engine v0.1.0 (scoring logic)
│   ├── uncertainty_mass.py         ← Uncertainty Mass (UM) computation
│   ├── epistemic_debt.py           ← Epistemic Debt Score (EDS) — 0–100 org metric
│   └── model_registry.py           ← AI model calibration registry + trust scoring
│
├── audit/                          ← Audit trail infrastructure
│   ├── stp_seal.py                 ← Sovereign Trace Protocol integration
│   └── audit_log.py                ← Decision log structure + export
│
├── connectors/                     ← Enterprise AI platform connectors
│   ├── openai_connector.py         ← OpenAI API connector
│   ├── anthropic_connector.py      ← Anthropic API connector
│   ├── generic_rest_connector.py   ← Generic REST API connector (any model)
│   └── README.md                   ← Connector documentation
│
├── api/                            ← REST API layer (FastAPI)
│   ├── server.py                   ← FastAPI server entry point
│   └── routes/
│       ├── score.py                ← POST /score — CDI scoring endpoint
│       ├── audit.py                ← GET /audit — audit trail retrieval
│       └── calibrate.py            ← POST /calibrate — model calibration update
│
├── dashboard/                      ← Web dashboard (optional)
│   ├── index.html                  ← CDI monitoring dashboard
│   └── assets/                     ← Dashboard assets
│
├── supabase/                       ← Database schema
│   └── schema.sql                  ← CDI scores, model registry, audit log tables
│
├── docs/                           ← Documentation
│   ├── cdi-formula.md              ← CDI formula derivation and weight rationale
│   ├── eds-formula.md              ← Epistemic Debt Score formula (5 components)
│   ├── architecture.md             ← System architecture and integration guide
│   └── deployment.md               ← Enterprise deployment guide
│
├── concept/                        ← Concept and proposal documentation
│   ├── NICE-PROPOSAL.md            ← NICE Challenge 2026 full proposal
│   ├── MARKET-ANALYSIS.md          ← Yangtze Delta industrial AI market analysis
│   └── USE-CASES.md                ← Industrial sector use cases
│
├── tests/                          ← Test suite
│   ├── test_cdi_engine.py          ← CDI scoring unit tests
│   ├── test_uncertainty_mass.py    ← UM computation tests
│   ├── test_eds.py                 ← EDS scoring tests
│   └── test_audit_trail.py         ← STP seal integration tests
│
├── AXIOM.md                        ← CDI Engine full technical documentation
├── NOTICE
├── COMMERCIAL-LICENSE.md
├── TEST_SUITE_AXIOM.md             ← Test suite documentation
├── NICE_PROPOSAL.md                ← NICE Challenge 2026 proposal (top-level copy)
├── LICENSE                         ← GPL-3.0
└── README.md                       ← This file

Overview

Most industrial AI deployments stop at the output. They compute, they score, they recommend — and then they hand an operator a decision with no way to know how much to trust it.

AXIOM is a model-agnostic middleware layer that sits between any deployed AI system and its downstream action triggers. Before a quality control model passes or rejects a component, before a predictive maintenance system flags or clears equipment, before a logistics optimizer routes or holds a shipment — AXIOM intercepts the AI output, scores its epistemic reliability with the Confidence Deployment Index (CDI), checks it against validated thresholds, and either certifies it for automated execution or gates it for human review. Every decision, every score, every gate is sealed to an immutable audit trail.

How it works — three steps:

An AI model produces an output with a confidence value. AXIOM's CDI Engine intercepts it, scores four reliability dimensions, and computes a CDI and Uncertainty Mass in real time.
An operator opens the monitoring dashboard and sees a reliability-weighted view of all AI decisions — green decisions are executing automatically, amber decisions are flagged for review, red decisions are blocked.
Every decision, every CDI score, and every audit event is permanently sealed with a cryptographic timestamp. If a decision is challenged, the sealed record proves exactly what the AI knew and how confident it was.

The CDI Engine

The CDI Engine (v0.1.0) is the core of AXIOM — an epistemic scoring system that tells operators how much to trust each AI decision, and how much to trust the trust score itself.

Scoring Dimensions

Dimension	Weight	Description
Output Confidence Score (OCS)	35%	Model-declared confidence, adjusted for calibration history against verified ground truth
Corroboration Score (COR)	30%	Agreement with recent historical decisions on comparable inputs within the same process context
Temporal Validity (TVR)	20%	Model relevance decay since last calibration event or training update
Context Consistency (CCI)	15%	Cross-check of input parameters against declared domain constraints and historical normal ranges

Weight rationale: OCS and COR carry the highest weight because they are the two externally-verifiable dimensions — model confidence can be calibrated against ground truth, corroboration requires independent decision history agreeing. TVR and CCI are internally-derived signals and are weighted accordingly. Full derivation: docs/cdi-formula.md

CDI Formula

CDI = (OCS × 0.35) + (COR × 0.30) + (TVR × 0.20) + (CCI × 0.15)
UM  = 1 − min(calibration_coverage / 3, 1) × (1 − score_variance)

Output

CDI Range	Validity Status	Action
≥ 0.70	VALID	Decision executes automatically
0.40–0.69	DEGRADED	Decision flagged — human review required before execution
< 0.40	SUSPENDED	Decision blocked — human authorization required

Uncertainty Mass (UM)

Every CDI score carries an Uncertainty Mass (UM) — a measure of how much the CDI score itself should be trusted:

UM	Meaning
< 0.35	CDI is reliable — act on it
0.35–0.60	CDI is useful but uncertain — review before high-stakes actions
≥ 0.60	Do not rely on this CDI score

SUSPENDED decisions remain visible in the monitoring dashboard with a blocked status and human authorization prompt. They are never silently dropped — their presence is itself operational information.

Graduated Model Trust

The CDI Engine does not assume any AI model is trustworthy without evidence. It uses a graduated model trust score [0.0–1.0] derived from calibration history against verified ground truth, which directly reduces the OCS uncertainty penalty as evidence accumulates.

Trust Score	Calibration Status	OCS UM Penalty	Measurement Class
0.0	UNCALIBRATED (no ground truth)	0.20	INFERENTIAL
0.01–0.59	PARTIAL (1–49 validated decisions)	0.08–0.20	EVALUATIVE_PARTIAL
0.60–0.85	PARTIAL (50–249 validated decisions)	0.03–0.08	EVALUATIVE_PARTIAL
1.0	VERIFIED (formally calibrated)	0.00	EVALUATIVE_CERTIFIED

The engine becomes more confident as evidence earns it — without requiring code changes. The trust score is declared at initialization, logged to the audit trail, and surfaced in every scored output. Call engine.update_model_calibration(n, 'PARTIAL') as ground truth accumulates.

Epistemic Debt Score (EDS)

An organizational-level metric (0–100) measuring cumulative AI epistemic risk across all deployed systems:

Component	Weight	What It Measures
Calibration Coverage	25%	Proportion of deployed models with verified calibration
SUSPENDED Decision Rate	25%	Rate of blocked decisions across all systems
DEGRADED Override Rate	20%	Rate at which flagged decisions are manually overridden without review
Audit Trail Completeness	20%	Proportion of decisions with sealed audit records
Model Age Distribution	10%	Staleness of deployed models relative to calibration events

EDS > 80: AXIOM Certified — eligible for regulatory pre-clearance documentation. EDS 50–80: Active improvement required. EDS < 50: Systemic reliability risk — remediation engagement recommended.

AI Model Integration

AXIOM is model-agnostic. Any AI model that produces an output with a confidence value can be registered and scored by the CDI Engine. Built-in connectors are provided for common enterprise AI platforms.

Connector Configuration

Priority	Connector	Use Case
Primary	Generic REST (`generic_rest_connector.py`)	Any AI model with a REST API endpoint
Built-in	OpenAI Connector (`openai_connector.py`)	OpenAI API models (GPT series)
Built-in	Anthropic Connector (`anthropic_connector.py`)	Anthropic API models (Claude series)

How It Works

AI model produces output with confidence value → AXIOM connector intercepts it
CDI Engine scores OCS, COR, TVR, CCI across the four dimensions
CDI and UM are computed — VALID / DEGRADED / SUSPENDED gate fires
Decision sealed to STP audit trail with SHA-256 binding
If model calibration data available → update_model_calibration() called to reduce UM penalty without code changes

The engine registers every model at initialization with a declared calibration status. New models start as UNCALIBRATED — the full UM penalty applies. As ground truth validation accumulates, calibration status upgrades automatically.

NICE Challenge Alignment

AXIOM was developed as the primary submission for the NICE Challenge 2026 — AI & Communications field. Full proposal: concept/NICE-PROPOSAL.md

Requirement	Status
Cutting-edge technical innovation — AI & Communications	✅
Independent IP rights	✅ — DOI-registered foundation (AION Stack, CERTUS Engine)
Intention to establish entity — Yangtze Delta	✅
Detailed business plan — introduction through risk analysis	✅
Market analysis — Yangtze Delta industrial AI	✅
Financial and implementation plan (3-year)	✅
SWOT and risk analysis	✅
Project Manager introduction	✅ — Sheldon K. Salmon, ORCID: 0009-0005-8057-5115
Technology Readiness Level	TRL 5 — validated in relevant environment
Potential customers / partners in China	🔄 In progress via NICE ecosystem

The AXIOM Ecosystem

AXIOM is a unified platform with three core capabilities built into a single service offering:

Service	Purpose	Access
AI Decision Audit	One-time reliability assessment for deployed AI systems	Engagement-based
Continuous CDI Monitoring	Real-time per-decision scoring via middleware	License-based
Epistemic Debt Score Report	Quarterly organizational AI reliability report	Retainer-based

All three share the same CDI Engine, the same UM architecture, and the same STP audit trail infrastructure.

Sovereign Trace Protocol Integration

AXIOM integrates with the Sovereign Trace Protocol — a permanence infrastructure with cryptographic sealing. Every CDI decision score, every UM value, every gate event, and every model calibration update is permanently sealed with a SHA-256 bound audit trail.

Audit integrity note: Sealed records are filed to the STP ledger repository. This provides immutable timestamping — if a decision is challenged during a regulatory audit, quality review, or legal proceeding, the sealed record proves exactly what the AI knew, how confident it was, and whether uncertainty was properly flagged before the action triggered. The audit trail cannot be retroactively altered.

How It Works

Event	Trigger	Sealed Record
Decision Scored	Every CDI computation	CDI score, UM, validity status, model trust level, input context hash
Gate Fired	DEGRADED or SUSPENDED decision	Full decision context + gate reason + human review outcome
Calibration Updated	`update_model_calibration()` called	Old trust score, new trust score, sample count, timestamp
EDS Report Generated	Quarterly report cycle	Full EDS breakdown, all five components, organization identifier

Verification

Anyone can verify a sealed audit record by:

Recomputing the SHA-256 hash of the decision record
Comparing it to the hash sealed in the STP ledger
If they match, the record has not been altered since sealing

Technical Stack

Layer	Technology	Why
CDI Engine	Python 3.11+	Industrial-grade, enterprise-deployable, stdlib-compatible
API Layer	FastAPI	High-performance REST API, async-native, auto-documented
Model Registry	SQLite (dev) / PostgreSQL (prod)	Calibration history, trust scores, decision log
Audit Trail	Sovereign Trace Protocol + SHA-256	Immutable, verifiable, tamper-evident
Connectors	Python REST clients	Model-agnostic, any AI platform
Dashboard	HTML + JavaScript	Lightweight, deployable anywhere
Backend Sync	Supabase (optional)	Real-time, row-level security
Deployment	Docker + GitHub Pages (`public/`)	Container-first, scalable, CDN-delivered landing page
License	GPL-3.0 with Commercial option	Open core for research, commercial for enterprise

Three Core Services

🏭 AI Decision Audit — Enterprise Reliability Assessment

For a manufacturing enterprise that has deployed AI in production and needs to know how reliable those decisions actually are — without waiting for a failure to find out.

CDI baseline scoring across all deployed AI decision systems
Uncertainty Mass profiling per model, per process
Epistemic Debt Score (EDS) organizational baseline (0–100)
Sealed audit trail for every assessed decision
Calibration gap identification — which models need ground truth validation
Remediation roadmap with prioritized interventions
Deliverable: CDI Baseline Report + EDS Report + sealed ledger entry

📊 Continuous CDI Monitoring — Real-Time Decision Scoring

For an enterprise that wants ongoing reliability scoring for every AI decision, automatically, without per-engagement consulting fees.

AXIOM middleware deployed between AI model and action trigger
CDI computed per decision in real time
VALID / DEGRADED / SUSPENDED gates fire automatically
Real-time monitoring dashboard with CDI distribution
DEGRADED decisions routed to human review queue
SUSPENDED decisions blocked pending human authorization
Automatic model calibration updates as ground truth accumulates
STP seal on every decision — continuous audit trail
Monthly EDS report generated automatically

📜 Epistemic Debt Score — Organizational AI Reliability Report

For an executive team that needs to understand and report on the reliability of their organization's AI decision infrastructure — before regulators require it.

Quarterly EDS report (0–100) across all deployed systems
Five-component breakdown: calibration coverage, SUSPENDED rate, DEGRADED override rate, audit trail completeness, model age distribution
Benchmark comparison against industry calibration datasets (accumulated from audit engagements)
Regulatory compliance documentation package
EDS improvement roadmap with quarterly milestone tracking
AXIOM Certified status for organizations maintaining EDS > 80
Sealed report — SHA-256 bound, tamper-evident, submittable to regulators

Data Security & Audit Integrity

No personal data collected — AI decisions are scored by input context hash, not by individual identity
Model outputs processed locally where possible — API-based models use encrypted transport (TLS 1.3)
Calibration data stored with row-level security (Supabase RLS) when cloud sync is enabled
Audit trail sealed with SHA-256 — tamper-evident, cannot be retroactively altered
Enterprise deployments support air-gap operation — CDI Engine runs fully offline with local SQLite registry
STP ledger entries are public-readable but write-authenticated — no unauthorized sealing

Note: Cloud infrastructure providers (Supabase, hosting platforms) may log request metadata per their own policies. Review their privacy documentation for deployment-level considerations.

Installation & Deployment

1. Clone the repo

git clone https://github.com/AionSystem/AXIOM.git
cd AXIOM

2. Install dependencies

pip install -r requirements.txt

3. Register your AI model

from engine.cdi_engine import CDIEngine

engine = CDIEngine()
engine.register_model(
    model_id="your-model-id",
    model_type="rest",           # or "openai", "anthropic"
    calibration_status="UNCALIBRATED",
    calibration_samples=0,
    registered_by="your-deployment-id"
)

As ground truth validation accumulates, call engine.update_model_calibration(n_samples, 'PARTIAL') — no code changes required. The CDI Engine reduces the UM penalty automatically.

4. Configure Supabase (optional — for cloud sync and dashboard)

Create a free Supabase project
Run supabase/schema.sql to create the CDI scores, model registry, and audit log tables
Set SUPABASE_URL and SUPABASE_KEY in your environment

5. Run the API server

uvicorn api.server:app --host 0.0.0.0 --port 8000
# POST /score   → CDI scoring endpoint
# GET  /audit   → Audit trail retrieval
# POST /calibrate → Model calibration update

6. Deploy with Docker

docker build -t axiom .
docker run -p 8000:8000 axiom

License

AXIOM is dual-licensed:

GNU General Public License v3.0 — for academic, research, and open-source use
Commercial License — for enterprise deployment, proprietary integration, and OEM applications

User Type	License
Academic / Research	GPL-3.0 (Free)
Open-source projects	GPL-3.0 (Free)
Internal R&D (non-commercial)	GPL-3.0 (Free)
Enterprise deployment (commercial)	Commercial License (Fee)
OEM / white-label integration	Commercial License (Fee)

See LICENSE for GPL terms and COMMERCIAL-LICENSE.md for commercial licensing information.

For commercial licensing and enterprise engagement inquiries: aionsystem@outlook.com

Acknowledgments

AION Constitutional Stack — foundational certainty engineering frameworks
CERTUS Engine — deployed precedent (VERITAS, UNDP Innocentive Challenge 2026)
Sovereign Trace Protocol — cryptographic audit trail infrastructure
FSVE (Formal Scoring and Validity Engine) — CDI scoring architecture origin
FastAPI — API layer
Supabase — optional cloud sync backend

"The code is open source. The architecture is not replicable."

This is an application of the AION Constitutional Stack — applied to industrial AI decision systems and enterprise reliability infrastructure. The method travels. The judgment behind it doesn't.

CDI Engine v0.1.0 — NICE Challenge 2026 submission. Epistemic Debt Score — organizational AI reliability, measured. AXIOM — Every AI decision scored. Every gate sealed. Every audit trail permanent.

Name		Name	Last commit message	Last commit date
Latest commit History 47 Commits
api		api
audit		audit
concept		concept
connectors		connectors
dashboard		dashboard
docs		docs
engine		engine
public		public
supabase		supabase
tests		tests
AXIOM.md		AXIOM.md
COMMERCIAL-LICENSE.md		COMMERCIAL-LICENSE.md
LICENSE		LICENSE
NICE_PROPOSAL.md		NICE_PROPOSAL.md
NOTICE		NOTICE
README.md		README.md
TEST_SUITE_AXIOM.md		TEST_SUITE_AXIOM.md

Folders and files

Latest commit

History

Repository files navigation

AXIOM — AI Certainty Middleware for Industrial Decision Systems

Table of Contents

Architect's Note on AI Use

Quick Start

Repository Structure

Overview

The CDI Engine

Scoring Dimensions

CDI Formula

Output

Uncertainty Mass (UM)

Graduated Model Trust

Epistemic Debt Score (EDS)

AI Model Integration

Connector Configuration

How It Works

NICE Challenge Alignment

The AXIOM Ecosystem

Sovereign Trace Protocol Integration

How It Works

Verification

Technical Stack

Three Core Services

🏭 AI Decision Audit — Enterprise Reliability Assessment

📊 Continuous CDI Monitoring — Real-Time Decision Scoring

📜 Epistemic Debt Score — Organizational AI Reliability Report

Data Security & Audit Integrity

Installation & Deployment

1. Clone the repo

2. Install dependencies

3. Register your AI model

4. Configure Supabase (optional — for cloud sync and dashboard)

5. Run the API server

6. Deploy with Docker

License

Acknowledgments

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 1

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages