MARIA OS

DECISION
INTELLIGENCE
OS

Name: MARIA OS
Author: MARIA OS

Self-Driving AI Operations,
Built on Human Judgment.

MARIA OS enables self-driving AI operations by first condensing human judgment into an operating system.

Instead of automating decisions blindly, it defines responsibility, boundaries, and human intervention upfront.

These judgment structures are transferred into autonomous AI agents, allowing execution to scale without losing human authority.

The result is an Agentic Organization—where AI agents operate as accountable members of your organization, governed by shared values and clear decision boundaries.

In 30 seconds

MARIA OS turns executive judgment into an operating system. It captures how leaders decide — where to trust automation, where to stop, and where humans must stay accountable — and makes those decisions executable at scale through AI agents.

Preserve human authority while scaling AI execution

Make implicit judgment explicit, structured, and reusable

Prevent AI autonomy from becoming organizational risk

This is for

Organizations where AI makes real decisions
Leaders who care about responsibility, not just speed
Judgment that must be consistent and reusable

This is not for

A faster way to chain prompts
Full automation without human accountability
AI that replaces leadership judgment

Universe Builder

Watch a zone come to life

universe-builder

Scroll to start building...

Build Sequence

Goal

Scope

Team

Skills

Build

Gates

Validate

Test

Deploy

Goal > Scope > Team > Skills > Build > Gates > Validate > Test > Deploy

Skills (K1-K8) are dynamically fetched and auto-refilled from Skill Store

Platform Capabilities

Core Features

Value Hierarchy Processing

Discover the gap between stated values and practiced values. AI learns from decisions.

Stated

"Customer experience is our priority"

Practiced

Cost cuts approved 3x faster

View

Responsibility Gate

AI pauses at designed boundaries. Human approval is architecture.

View

Decision Pipeline

Safety, compliance, responsibility — same gates in order.

View

Universe Architecture

Universe → Galaxy → Planet → Zone → Agent.

View

Database-Native

All AI state in PostgreSQL. Query, audit, debug with SQL.

View

Skills: Dynamic Fetch & Refilling

Agent execution power never depletes. Skills fetched on demand or generated when missing.

Fetch

Create

Bind

Refill

View

Complete Audit Trail

Every decision logged, linked, and traceable.

View

Phase-Aware Context

Launch mode vs steady-state. AI adapts criteria.

View

Collective Intelligence

Aggregate values from multiple stakeholders.

View

Execution Flow

Parallel, serial, loops with HITL checkpoints.

View

12 Industry Use Cases

CEOFinanceHealthcareAuditLegalManufacturingRetailEnergyInsuranceAuto-DevEducationMunicipality

View

Mission Control

Real-time visibility into every AI decision

Executive Brief

18,432 decisions processed.
Only 27 required human judgment.

All operational2s ago

Active Agents

142/ 6 zones

Decisions

847/min

Compliance

99.7%

Human Authority

100%

Preserved

HITL Gates

Awaiting approval

CEO Gates

ESCALATED

High-impact pending

Policy Violations

CLEAR

Last 24h

Philosophy Drift

0.2%

Value alignment

Audit Trail

100%

Traceable

Explainability

99.9%

Decisions explained

Skill Health

98.2%

Refill Queue

skills depleted

Auto-refilling...

Live Activity

now

Decision approvedExpense-007

Threshold checkPolicy-003

HITL requestedApproval-001

Skill refilledRefill-002

Skill Requirement Engine

Skills are fetched or generated on demand

Agents receive skills from the Skill Store. Missing skills are dynamically created.

skill-fetch

Dynamic Skill Fetch

Requirement resolved from existing Skill Store

Requirement

skill_key:

K4.verify

domain:

contract

min_level:

Skill Store

Risk Evaluation

K4.verify / contract

Cost Analysis

K3.infer / finance

Verifier-01Contract Verification

Requirement

Skill key, domain, level

Store Search

Query matching capability

Resolve/Create

Generate if not found

Bind to Agent

Phase Gate enforced

No HardcodingConsistent QualityInfinite Extensibility

Skill Governance

Skill / Phase Permission Matrix

Each skill category has defined boundaries per decision phase.

P0Observe

P1Propose

P2Experiment

P3Execute

P4Irreversible

P5Govern

Collect

Curate

Infer

Verify

Plan

Execute

Audit

Govern

AllowedGateBlocked|K6 blocked in observe, K8 requires authority

Organizational Topology

Universe > Planet > Zone > Agent

Each layer defines scope, permissions, and responsibility boundaries.

MARIA Universe3 planets

Common Infrastructure

Skill Registry

Policy

Routing

Executive Decision11 agents across 3 zones

Design Principles

Verifier and Auditor are independent

Executor always pairs with Phase Gate

Governor at Planet level for consistency

Orchestrator at Zone entry point

Skill Refilling Control

Agent execution power never depletes

Refill only the depleted skills at the correct phase.

skill_ready

Capability Mismatch

Agent knows what to do but cannot execute at required level

Quality Degradation

Skill is fatigued or no longer fits current context

Phase Mismatch

Decision weight exceeds skill strength

Environment Change

Skill is correct but premises have shifted

Agents get tired. Environments change. Decisions get heavier.

MARIA OS refills skills, not replaces Agents.

Real-time Skill Refilling

Watch skills flow to Agents

Skills are continuously refilled from the central Skill Store to maintain Agent capabilities.

Skill Store

K1Collect

K2Curate

K3Infer

K4Verify

K5Plan

K6Execute

K7Audit

K8Govern

Dynamic Refill Pipeline

Market Analyst

Strategy

Risk Evaluator

Compliance

Data Processor

Operations

Quality Auditor

Governance

Refill Log

Waiting for refill events...

Total Skills Bound

Refills Complete

Active Transfers

Agents never stop. Skills arrive before they are needed.

Post-Deployment

Execution Flow Circuit

Parallel, serial, and loop paths with HITL checkpoints

Trigger

Collect Agent

Parallel

Curate Agent

Verify Agent

Merge Results

Infer Agent

Human ReviewHITL

Plan Agent

Compliance Gate

Execute Agent

Quality Loop

back to Infer

Audit Agent

Final ApprovalHITL

Complete

Flow Legend

Agent

HITL

Parallel

Loop

Gate

Decision

HITL Checkpoints

Human Review

Final Approval

Parallel execution for efficiency

Loop circuits for quality

Human gates clearly marked

Auto-Dev Architecture

Why Autonomous Development Completes

Even with ambiguous instructions, auto-dev reaches the completion line

Process

Define Completion Line

Not perfect spec, but minimum success criteria

Small Iterations

Build, run, verify, fix — always recoverable

Auto-Complete

Continues until acceptance criteria met

Iteration Cycle

Build

Run

Verify

Fix

Cycle 1/3

Pass & Complete

Stop with Context

Why It Completes

Executor AI

Builds & Runs

Judge AI

Verifies & Approves

No RunawayExecutor & Judge are separate

No Goal DriftCompletion criteria locked

No False DoneVerifiable acceptance only

Ideal for:Code improvementsTest additionsFeature buildsRefactoringPrototype drafts

Turning autonomous development into a trustworthy process

Completion Rate Architecture

True Completion = 5 Metrics Multiplied

Not "finished tasks" but "safely reached judgment-ready state"

Start Success起動成功率

Prerequisites met, execution begins

99.2%

Continuation継続率

No interruptions until final phase

--%

Deviation Control逸脱抑制率

Self-corrects when drifting from goal

--%

Goal Reach終了条件到達率

Reaches defined end conditions

--%

Quality Pass品質合格率

Passes quality gate verification

--%

Total Completion Rate

--%

= Product of all 5 metrics

Why Multiplication?

If any metric = 0, total = 0

No single point of failure hidden

True quality, not task throughput

MARIA OS Definition

Completion is not "finished running" but "safely reached a judgment-ready state that a third party can take over."

Phase Flow

Intake

Contract

Plan

Execute

Verify

Package

Handoff

Industry Solutions

12 Use Cases

Pre-configured value hierarchies and decision pipelines for your industry.

1/12

CEO Decision OS

Your vision becomes the operating system. Philosophy-driven decisions, encoded into every judgment across the organization.

Structurize executive vision and strategic decisions as 'decision policy sets' and reflect them in organizational judgment criteria.

Vision AlignmentPhilosophy EncodingStrategy Simulation

Details

2/12

Finance

Financial decisions are determined by whether audit and responsibility hold, not by correctness alone.

Achieve complete decision traceability as a standard foundation for audit and accountability.

CreditInvestmentAML/KYC

Details

3/12

Healthcare

Patient safety is not negotiable. Every AI decision passes through the Hippocratic gate before execution.

Balance automation and risk management through automatic halt of high-risk decisions and human gates.

Clinical DecisionDiagnosis SupportTreatment Plan

Details

4/12

Audit

Audit value lies in evidence integrity before discovery. Complete traceability from decision to outcome.

Build a complete picture of evidence, rationale, responsibility, and halt criteria. Integrate compliance and internal controls.

Internal AuditExternal AuditSOX

Details

5/12

Legal & Compliance

Regulatory compliance is the foundation. Every decision is audit-ready from inception with full jurisdiction awareness.

Automatically update and apply AI decision rules to dynamically changing regulations and policies.

Contract ReviewRegulatory FilingDue Diligence

Details

6/12

Manufacturing

Manufacturing decisions must be stoppable on the floor. Worker safety and quality control above throughput.

Integrate different decision criteria across departments and manage consistent OK/NG decisions organization-wide.

Quality ControlMaintenanceProcess Opt

Details

7/12

Retail & E-commerce

Customer trust drives every recommendation. Personalization without manipulation, transparency in pricing.

Dynamically configure decision flows, approval flows, and evidence collection based on role and business context.

Pricing StrategyInventoryPersonalization

Details

8/12

Energy & Utilities

Grid reliability is non-negotiable. Sustainability meets operational resilience with public safety priority.

Measure decision stability scores and confidence intervals. Visualize decision accuracy and safety margins.

Grid ManagementDemand ForecastRenewable Integration

Details

9/12

Insurance

Fair assessment, transparent pricing. Risk decisions that policyholders can trust without discrimination.

Structure expert decision logic and intent. Convert into assets that successors can reference.

UnderwritingClaims ProcessingFraud Detection

Details

10/12

Auto-Dev

Auto-Dev is not 'AI writes code.' Auto-Dev is 'the database authorizes changes.' Full reproducibility guaranteed.

Embed approval, rejection, and explanation request flows into each generative AI step for high explainability and control.

Code GenPR ReviewSecurity Scan

Details

11/12

Education

Students are not single recommendation vectors. Education decisions are managed as states within a neural decision network.

Manage education decisions as neural network-like structures, not single scores. Control layer halts excessive fixation.

Learning SupportCareer GuidanceCurriculum Adapt

Details

12/12

Municipality

Migration and employment are not one-way matching. They are decision networks with time and state that can be paused and reviewed.

Design migration, employment, and settlement as decision networks with time and state. Guarantee explainable structure on failure.

Migration MatchJob PlacementSettlement Support

Details

Use Case Deep Dive

Judgments are always traceable

MARIA OS is designed for audit from the start. Every judgment passes through structured gates that ensure explainability and accountability.

Gate 0Schema Gateスキーマゲート

Blocks execution if judgment structure is not audit-ready

Unexplainable judgments are structurally blocked from execution.

Structural Separation

Judgment Phase vs Responsibility Phase

Many AI systems blend thinking, deciding, and executing into one. MARIA OS separates judgment from responsibility assignment by design.

Judgment Phase判断フェーズ

Compare options

Evaluate against criteria

Place tentative conclusion

AI can assist with judgment. But judgment alone does not authorize action.

RESPONSIBILITY GATE

Responsibility Phase責任フェーズ

Confirm who owns the decision

Define scope of accountability

Lock evidence and proceed

Responsibility must be explicitly assigned before execution begins.

AI decides freely. Humans take responsibility blindly.MARIA OS prevents this by structure.

Visual Governance

Universe View: Gates at Every Boundary

AI operations are visualized as space, not logs. See where agents are, which gates they passed, and where judgments are held.

Finance Zone

3 agents

Operations Zone

2 agents

Compliance Zone

4 agents

Zone

Agent

Gate

Agents cannot freely cross zones

Every boundary requires gate verification

Trace flow is visible at a glance

Audit-Ready Output

Evidence Package: Submit-Ready from Day One

AI adoption stalls when teams cannot explain decisions to auditors. MARIA OS bundles all supporting evidence automatically with every judgment.

DECISION

Judgment ID

JDG-2024-0892

Result

Approved with conditions

Confidence

87%

EVIDENCE BUNDLE

Complete

Input Snapshot

Frozen inputs at execution time

Evidence & Sources

Referenced data with provenance

Applied Policies

Rules that governed the judgment

Approval Records

Who approved, when, and scope

Execution Trace

Full reproducibility info

Export as audit package for internal/external review

MARIA OS is not about making AI faster.It is about making AI safe enough for society.

Accountability by design. Not by afterthought.

Cross-Industry Architecture

Judgment & Responsibility OS Structure

One OS structure adapts responsibility density across industries

MARIA OS

Responsibility Gate

Trace Gate

Policy Layer

Evidence Layer

Unified Decision Flow

Input

Evidence

Judgment

Gate

Responsibility

Execution

Finance

4 gates

Municipality

3 gates

"Not centralized, not chaotic — a structure with responsibility gradients"

The Problem

Judgment does not scale. Execution does.

Human organizations scale execution by automation. But judgment has always remained personal, tacit, and fragile.

When AI agents act without defined judgment context, decisions fluctuate, conflict, and drift.

This is not an AI problem. It is a judgment problem.

Judgment that is not defined becomes risk when automated.

The Solution

Condense judgment. Make it executable.

MARIA OS starts before agents. It captures how decisions are actually made:

what is evaluatedwhat is prioritizedwhen humans intervenewhen automation stops

Decision structures condensed into a Judgment OS. Not documentation. An executable system.

Only judgment that is formalized can be safely automated.

Decision OS Studio

Design judgment as executable code

Decision OS Studio

Layer Stack

Global Policy12

galaxy

Finance Ops24

universe

AP Process18

planet

Expense Zone8

zone

Graph view, table view, YAML — your choice

Operational Views

Observe, debug, and control

Log Explorer

Search logs...

14:32:01.234INFODecision completedDEC-8847expense-007

14:32:01.102DEBUGPolicy check passedPOL-2341policy-003

14:32:00.987WARNThreshold exceeded, escalatingTHR-0012validator-002

14:32:00.845INFOHITL approval requestedAPR-4421approval-001

14:32:00.712INFOAgent spawnedAGT-1234expense-008

14:32:00.601DEBUGContext loadedCTX-9921loader-001

Agents142 active

EXP-007

Expense Processor

24 tasks

POL-003

Policy Validator

18 tasks

APR-001

Approval Router

3 tasks

AUD-002

Audit Logger

156 tasks

REC-001

Recovery Agent

0 tasks

MON-004

Monitor Agent

42 tasks

Decisions

18,432 today

DEC-18432

Expense Approval

2s ago

$4,250approved

DEC-18431

Vendor Check

5s ago

approved

DEC-18430

Policy Exception

12s ago

$12,500pending

DEC-18429

Threshold Check

15s ago

$890approved

Stops & Incidents2 active

INVESTIGATINGINC-003

Threshold Stop: Amount exceeds limit

Expense of $12,500 triggered automatic stop

Timeline

14:30:12Expense submitted

14:30:13Threshold check failed

14:30:13Automatic stop triggered

14:30:14Human escalation initiated

Conflict Resolution

Conflict Resolution as a Common Language

MARIA OS treats conflicts as assets to be structured, not failures to hide.

Conflict Classification

Hard: Mutually exclusive

Soft: Different scope

Temporal: State changed

Decision Gate

Record

TargetContract

Candidates2

ImpactHigh

OwnerCFO

MARIA OS aligns judgment with responsibility at the OS level.

Continuous Re-evaluation

Decision Gating

Transform Conflicts into Structured Decisions

Conflicts are decision gates — both AI and humans can evolve safely when properly designed.

Intake

Detect

Classify

Gate

Record

HardHuman Review

SoftAuto-merge

TemporalVersion compare

Human Gate

IrreversibleSTOPPED

ApprovalCFO

ResponsibilityLocked

Audit Trail

14:32:01Detected

14:32:02Classified

14:32:03Gate triggered

14:35:47Approved

Decision → Re-detect → Feedback Loop

Knowledge Maintenance

Scheduled Knowledge Cleansing

Daily

Real-time conflict check

847

this month

Weekly

Cross-agent consistency

this month

Monthly

Full structure review

this month

Pipeline

Active

Capture

Scan

Cleanse

Gate

Report

Critical

Warning

Resolved

Cycle-driven knowledge maintenance

Loop Architecture

Each Loop Ascends to Higher Dimensions

Particles orbit within each ring. When a cycle completes, one breaks through and ascends to the next dimension.

Execution

Single Task Loop

break

Learning

Pattern Recognition

break

Adaptation

Policy Refinement

break

Evolution

System Growth

break

Transcendence

New Capability

∞

Infinite Ascent

Dimensions

∞

Potential

↑

Always Up

Scope Expansion

Ascend Through Scales of Impact

Particles orbit faster as mastery grows. Break the orbit, ascend to the next scale. Gates multiply, impact expands forever.

Individual

Single Agent

Gates

escape

Team

Collaborative

Gates

escape

Organization

Enterprise

Gates

escape

Ecosystem

Cross-Boundary

Gates

∞

Beyond All Scales

Gate Density

2 → 5 → 12 → 30+

Orbit Speed

Accelerates with scale

Expansion

Forever upward

Self-Improving

Tick-Driven Knowledge Loop

Add

Detect

Evaluate

Approve

Feedback

Knowledge Re-evaluation

Phase Consistency

Impact Rescoring

Human Feedback

Active

Today

Pending

Cycles, not single events

Verification

Phase Alignment Loop

Synchronization

Active

Judgment

Responsibility

Verification

Alignment

Continuous

Hierarchy

Agent

11/12

Zone

4/4

Planet

1/1

Rule-based

Constraints

Evidence

Patterns

Continuous verification across hierarchy

Risk Assessment

Impact Rescoring

Cycle: 1,247

Financial Risk+0.6

6.8/ 10

Operational Impact-0.9

4.2/ 10

Compliance Score0.0

3.5/ 10

Recent

2m agoKnowledge added+0.3

5m agoConflict resolved-1.2

12m agoDecision made+0.8

-23%

Reduction

847

Day

99.2%

Accuracy

Continuous rescoring

Oversight

Human-in-Loop Feedback

Approve

New Info

Re-detect

Re-review

Contract #1247

2h ago

approved

Budget #892

30m ago

re-review

Hiring #445

Just now

pending

94%

First-pass

18%

Re-review

2.4h

Avg. Time

Continuous feedback

Quality

Seven Pillars of Reliability

Quality Gates

Automated boundaries

Iterative Loops

Continuous cycles

Impact Rescoring

Dynamic evaluation

Evidence Trails

Full traceability

Human Feedback

Approval integration

Transparent UX

Visual clarity

Quality KPIs

Continuous monitoring

Gate Board

Interpret

Propose

Decide

Act

78:22

Auto vs Human

94%

Alignment

Verified, traceable, continuously improving

Architecture

Decision Maturity Levels

Maturity CurveL1 → L5

ReactiveAuditable

Agent maturity through loops, gates, and responsibility

Field Competence

Agentic Skills in the Real World

Internal structure manifests as precision and accountability

Phase Understanding

Phase CheckImpact scope

Risk Awareness

Impact ScoringRisk prioritization

Cyclic Improvement

Feedback LoopsSelf-improvement

Explainability

Evidence TrailsVerifiable outputs

Evidence-Based

Conflict CleansingFact-grounded

Skill Flow

Structure

Behavior

Execution

Outcome

Precision

Cross-checked

Reproducible

Traceable

Re-learning

Auto-improved

Accountable

Contextualized

Field-ready agent skills through structure and accountability

The Scale

Autonomous execution, governed by design.

Once judgment is embedded into the OS, autonomous AI agents can safely scale.

thousands of agentsoperating in parallelacross multiple universes

But responsibility does not scale with them. Human authority remains explicit.

Execution scales. Responsibility stays human-sized.

The Builder

Universe Builder: from judgment to organization

Not a workflow builder. A system that:

takes judgment requirementsdesigns responsibility boundariesassembles autonomous agentsenforces quality gatesdeploys governed execution

What you build is not an agent. You build a Universe.

Nothing critical left implicit. Nothing important added later.

Universe Builder

Configure governance before deployment

maria.os / universe-builder

Goal

Team

HITL

Gates

Validate

Deploy

Human-in-the-Loop Points

Define where human approval is required

Amount Threshold

> $5,000

required

New Vendor

First-time

required

External Write

Any external

required

Design governance before deployment, not after

The Architecture

Scale Without Losing Control

MARIA OS is designed to operate at scale.

Up to 10,000

AI agents per organization

But humans never have to manage 10,000 decisions.

MARIA OS aggregates execution while keeping decisions, approvals, and accountability human-sized.

Scale execution, not responsibility.

Aggregated execution, human-sized accountability

The Growth

You don't add agents. You grow universes.

In MARIA OS, scaling does not mean adding agents blindly.

defined scopevalidated judgmenthuman responsibilitymonitored execution

As zones appear, they light up on the topology map. Not as instances, but as governed operational units.

This is what safe AI scaling looks like.

The Difference

What MARIA Does Differently

Decisions are first-class objects

Every decision is logged, linked to evidence, and traceable later. Not as logs. As structured, queryable decision records.

18,432 decisions today

All traceable, all auditable

Swipe or wait

1 / 4

The Dashboard

What you see in MARIA OS

Thousands of decisions/dayOnly critical need humansViolations stopped, not hiddenEvery approval leaves evidence

Today: 18,432 decisions. Only 27 required human responsibility.

Human dependency: 18% → 14%. Authority remained human-controlled.

Real-time metrics, not post-hoc reports

The Boundary

Human-in-the-Loop Is Not an Exception

Human approval is not a failure state. It is a designed boundary.

When AI reaches a responsibility boundary, it pauses. If human authority is required, the decision waits. This is not a failure — it is how the system was designed.

Decisions requiring approvalApproving rolesRejection handling

AI never crosses boundaries. It stops by design.

The Hierarchy

Governance That Matches Real Organizations

MARIA OS mirrors how authority actually works.

Galaxy→Universe→Planet→Zone

Clear delegationClear responsibilityClear escalation

AI operates autonomously without dissolving accountability.

Universe → Galaxy → Planet → Zone → Agent

The Audit Trail

Evidence, Audit, and Trust

Every governable decision in MARIA OS is:

Logged Linked Traceable

Approvals, exceptions, recoveries — all part of the same decision history.

Nothing important happens without leaving a trace.

Logged, linked, traceable — always

The Users

Who Is MARIA OS For?

Enterprises deploying AI at scale

Regulated industries requiring auditability

AI-first orgs that cannot afford ambiguity

Teams wanting automation without losing control

Section 1

For Data Scientists & Engineers

Database-Native AI Orchestration

Structured state.

A database-native orchestration system. Actions from explicit state, not ephemeral context.

Properties

Deterministic

Same input, same output

Reproducible

Replay any decision

Queryable

SQL over decisions

Clear State

No hidden context

Query it. Reason about it. Scale it safely.

DecisionRecord

id:dec_8f2a

state:proposed

risk:R2

evidence:pending

State Transitions

Query Panel

SELECT * FROM decision_records
WHERE state = 'approved'

-- Results

AI autonomy designed like a distributed system, not a chatbot.

Models before prompts. State before speculation.

Section 2

Data Architecture

OLTP × Document Store Hybrid

Contract & Invariants

Two-layer persistence for AI decisions. OLTP stores the source of truth ledger with ACID guarantees. Document Store projects operational views for real-time UI. CDC streams keep them synchronized.

Public SpecContract

DecisionRecord Contract

idstring

stateenum

risk_tierR1-R3

intentstring

constraints_hardarray

values_layeredobject

evaluation_summaryobject

evidence_refsURI[]

policy_versionstring

execution_envelopeobject

timestampsbitemporal

Invariants

No record update, no side effect

Append-only journal

Bitemporal timestamps

Data Plane

OLTP StoreSource of Truth

DecisionJournal

DecisionCurrent

GateEvaluations

ExecutionEnvelopes

CDC

Change Streams

Document StoreProjection

UI read model

Universe nodes

Live status

fast pagination

State Machine

prop

→

vali

→

appr

→

appr

→

exec

→

reco

Control PlanePrivate

Value Scan Neural

••• value extraction + uncertainty •••

Policy Engine

••• gates + thresholds •••

Execution Engine

••• reversible plans •••

Evidence Bundle

••• artifacts + redaction •••

What we publish: contracts and invariants.
Reproduction requires accumulation.

Fix truth in OLTP, render universe in Document Store.

Decisions can be governed by database.

Guardrails & Decision Agreements

Policy Engine

AI decisions require more than correctness—they need consistency, explainability, and accountability. Policy Engine embeds guardrails into every automated judgment.

Core Modules

Gate Flow

Proposed

Validate

Evaluate

Decision

Result

pass

All constraints satisfied. Proceeding to execution.

Evidence Requirements by Risk Tier

R0Minimal

Unit tests + Logs

R1Moderate

Additional tests + Refactor trace

R2High

Human review + Escalation

R3Critical

Multi-party approval + Staged deploy

Policy Engine permits—not produces—AI decisions.

Rules + Values + Evidence = Safe, explainable actions.

Why not LangChain? Why not Workflow Engines?

Because autonomy is not a chain. And governance is not a workflow.

MARIA OS operates at a different architectural layer.

Stack position

Business Execution

ERP, CRM

MARIA OS

Decision OS

Workflow Engines

Steps/Tasks

LangChain

Prompts

LLMs

Models

vs LangChain

LangChain

MARIA

Unit

Prompt

Decision

State

Ephemeral

Persistent

Replay

Limited

Native

Approval

Ad-hoc

First-class

vs Workflows

Workflow

MARIA

Model

Step/Task

Decision

Branch

Predefined

Contextual

Failure

Retry

Escalation

Human

Optional

By design

LangChain: how AI speaks | Workflows: how work flows | MARIA OS: who is responsible

MARIA OS completes AI frameworks.

Deep Dive

A Layered Theory of Decision Engine OS

Dive as deep as your role requires.

Built to let organizations survive AI.

Not All Values Should Be Optimized

MARIA OS separates what must be protected from what can be optimized.

Deep Values (Protected)

Human Authority

LOCKED

Compliance

LOCKED

Auditability

LOCKED

Never optimized or traded.

Surface Values (Optimizable)

Profile: Balanced

Throughput

0.30

Latency

0.30

Human Load

0.25

Cost Efficiency

0.15

Balanced optimization

Change priorities. Not principles.

Humans define what matters. Execution adapts automatically.

Value Scanning

What Does Your Organization
Actually Value?

Not what it says.
What it decides.

Organizations often declare values.
Few know which values are actually practiced.

The Gap Between Words and Actions

Stated Values

"We prioritize customer experience"
"Innovation is our core value"
"We move fast and take risks"
"Employee wellbeing comes first"

From mission statements, town halls, and strategy decks

Practiced Values

Cost cuts approved 3x faster than UX improvements
80% of "innovative" proposals stopped at risk review
Approval delays average 4.2 days for external writes
Overtime requests auto-approved 94% of the time

From actual decisions, approvals, and system behavior

The difference is not hypocrisy.
It's lack of visibility.

The Vision: Passive Value Extraction

Our goal is a system that learns your judgment without interrupting your workflow

No Interviews

No Surveys

No Documentation

Start Organization Scan

MARIA OS reads what already exists:

Decision LogsApproval LogsOptimization ProfilesResponsibility GatesStop EventsRecovery EventsExecutive DecisionsEvidence Narratives

Collective Values Scanner

Scanning What Matters Most

MARIA OS continuously scans and evaluates collective values across your organization, transforming qualitative principles into quantifiable decision constraints.

Value Extraction

Continuously extracts and normalizes values from organizational context and stakeholder input.

Conflict Detection

Identifies potential conflicts between values before they become decision blockers.

Priority Encoding

Encodes value priorities as hard constraints that guide every autonomous decision.

"Values are not suggestions. They are the immutable laws that govern every decision."

The Value Extraction Pipeline

From enterprise data to governed AI operations

scanningscan

processingproc

deployingdepl

monitoringmoni

Policy Docs

Customer DB

Financial Data

Code Repo

Email Archive

Slack Logs

Scanning... 0%

Agent

MARIA

Data Sources

Policy Docs

Customer DB

Financial Data

Code Repo

Email Archive

Slack Logs

scan.log

>Initializing document scanner...

Scanning... 0%

AI Agent

MARIA OS

SCAN

CONDENSE & WEIGHT

DEPLOY & MONITOR

Step 1

Scan Documents & DBs

Step 2

Extract & Condense Values

Step 3

Deploy to AI Agents

Step 4

MARIA OS Monitors & Stops

Neural Value Network

Value Hierarchy Processing

L0 CoreL1 StrategicL2 Metrics

scanning

weighting

processing

output

L0: Core (Immutable)

L1: Strategic

L2: Metrics

Undefined

Values are processed hierarchically. L0 constraints must pass before L1 optimization.
Undefined values trigger safe defaults.

Value Extraction & Analysis

From raw enterprise data to structured, weighted decision values

MARIA OS Analyzing Values...0s

Initiating value extraction from scanned enterprise documents...

First, I need to identify explicit policy statements. Scanning document: "Corporate Governance Policy v3.2"
Found statement: "All financial transactions exceeding $10,000 require dual approval from department head and finance."
Extracting value: FINANCIAL_APPROVAL_THRESHOLD = $10,000
Extracting value: APPROVAL_REQUIREMENT = DUAL_SIGN_OFF
Confidence score: 0.94

Now analyzing implicit patterns from historical decisions in the CRM database...
Pattern detected: Customer refunds above $500 consistently escalated to senior management in 847 of 892 cases (94.9%)
Extracting implicit value: REFUND_ESCALATION_THRESHOLD = $500
Extracting implicit value: ESCALATION_TARGET = SENIOR_MANAGEMENT
Confidence score: 0.89

Cross-referencing with email communications archive...
Found recurring phrase in executive correspondence: "customer retention takes priority over short-term cost savings"
Mapping to value hierarchy: CUSTOMER_RETENTION > SHORT_TERM_COSTS
Weight assignment: CUSTOMER_RETENTION = 0.85, SHORT_TERM_COSTS = 0.45
Confidence score: 0.78

Analyzing Slack channel #finance-approvals for decision patterns...
Identified 1,247 approval conversations over 18 months
Extracting decision factors: urgency, amount, vendor_relationship, budget_status
Building decision tree from historical outcomes...
Decision pattern extracted: IF urgency=HIGH AND vendor_relationship=STRATEGIC THEN fast_track_approval
Confidence score: 0.91

Processing code repository commit messages and PR reviews...
Pattern: Security-related PRs require approval from security team lead
Extracting value: SECURITY_PR_APPROVAL = MANDATORY
Extracting value: SECURITY_APPROVER_ROLE = SECURITY_TEAM_LEAD
Confidence score: 0.97

Analyzing financial database transaction logs...
Identifying spending patterns by department...
Detected anomaly threshold: Transactions 3x above department average trigger review
Extracting value: ANOMALY_DETECTION_MULTIPLIER = 3.0
Extracting value: ANOMALY_ACTION = TRIGGER_REVIEW
Confidence score: 0.86

Now consolidating extracted values into value clusters...

Cluster 1: FINANCIAL_GOVERNANCE
- FINANCIAL_APPROVAL_THRESHOLD: $10,000 (weight: 0.92)
- DUAL_APPROVAL_REQUIRED: true (weight: 0.94)
- ANOMALY_MULTIPLIER: 3.0 (weight: 0.86)
- REFUND_ESCALATION: $500 (weight: 0.89)

Cluster 2: CUSTOMER_PRIORITY
- RETENTION_PRIORITY: HIGH (weight: 0.85)
- COST_SENSITIVITY: MEDIUM (weight: 0.45)
- SATISFACTION_THRESHOLD: 0.8 (weight: 0.72)

Cluster 3: SECURITY_COMPLIANCE
- CODE_REVIEW_MANDATORY: true (weight: 0.97)
- SECURITY_APPROVAL_ROLE: SECURITY_LEAD (weight: 0.97)
- DATA_ACCESS_LOGGING: REQUIRED (weight: 0.93)

Cluster 4: OPERATIONAL_VELOCITY
- FAST_TRACK_CONDITIONS: [urgency=HIGH, relationship=STRATEGIC] (weight: 0.91)
- STANDARD_SLA: 48_HOURS (weight: 0.78)
- ESCALATION_TIMEOUT: 24_HOURS (weight: 0.82)

Calculating inter-cluster relationships and potential conflicts...
Conflict detected: CUSTOMER_PRIORITY vs FINANCIAL_GOVERNANCE at refund threshold boundary
Resolution strategy: Apply CUSTOMER_RETENTION weight (0.85) > SHORT_TERM_COSTS (0.45)
Recommendation: Increase refund auto-approval to $750 for customers with LTV > $10,000

Building value graph with weighted edges...
Total nodes: 47 extracted values
Total edges: 156 relationships
Graph density: 0.142

Validating extracted values against organizational charter...
Charter alignment score: 0.89
Flagged items requiring human review: 3
- FAST_TRACK_CONDITIONS may conflict with audit requirements
- ANOMALY_MULTIPLIER threshold needs CFO confirmation
- SECURITY_APPROVAL_ROLE expansion to include DevOps leads

Preparing value package for AI Agent deployment...
Serializing value structures to Decision Axis format...
Generating boundary conditions and stop triggers...
Creating responsibility gates for R2+ decisions...

Value extraction complete.
Total extracted values: 47
High confidence (>0.9): 18
Medium confidence (0.7-0.9): 24
Low confidence (<0.7): 5 (flagged for review)

Ready for deployment to MARIA OS governed agents.

Values Extracted

Value Clusters

89%

Charter Aligned

Need Review

Values are automatically clustered, weighted by confidence, and validated against your organizational charter before deployment.

Value Extraction Output

Executive Summary

Your organization prioritizes safety and accountability.

Speed is optimized only when risk is low.

This behavior is consistent across Finance and IT.

Observed Values

Safety over Speed

Strong

Human Authority Preserved

Very Strong

Cost Efficiency

Weak

Throughput Optimization

Moderate

Judgment Pattern Mining

Recurring Judgment Patterns

External writes almost always escalate to human approval

Latency spikes are tolerated during month-end

Recovery succeeds more often after evidence enrichment

Finance zone has 2.3x stricter approval than Marketing

These patterns reveal how your organization actually thinks,
not how it claims to think.

Drift Detection

Value Drift Over Time

Value Drift Detected

Approval Strictness

+18% over 3 months

Throughput Optimization

-12% during incidents

Human Intervention Rate

+7% after audit

This is what executives actually want to know.
Is our culture drifting?

Your values are already encoded
in your decisions.

MARIA OS simply reveals them.

Value Scanning is not analysis.
It is organizational self-awareness at scale.

Decision Flow

Unbreakable Decision Sequence

Don't start with evaluation. Build unbreakable choices from values.

1/8Input context and action candidates

Irreversible Loss First

Decisions with irreversible consequences are filtered out first, regardless of score

Never Mix Conflicts

Values are evaluated independently. Processed as prioritized constraints, not weighted sums

Works Without Values

Even without defined values, infers minimum safety constraints and keeps only unbreakable options

Other AI evaluates and chooses. MARIA OS discards what breaks first.

Use Case 1/12

Finance

CreditInvestmentAML/KYCModel RiskCX AutomationCompliance

L0 CoreL1 StrategyL2 Metrics

Irreversible Loss0.90

Compliance0.90

Customer Asset0.90

Auditability0.85

Fraud Resist0.85

Privacy0.85

Speed

Cost Efficiency

Risk-Adj Return

CX Trust

Explainability

HITL Control

AML FPR

AML FNR

Approval Latency

Loss Rate

Override Rate

Conflict

Speed vs Audit

Conflict

Return vs Explain

Conflict

CX vs Fraud

Conflict

HITL vs Speed

Financial decisions are determined by whether audit and responsibility hold, not by correctness.

Value Scan implements this as the highest-level constraint.

Use Case 2/12

Audit

Internal AuditExternal AuditSOXEvidenceProcess MonitorFraud Detection

L0 CoreL1 StrategyL2 Metrics

Auditability0.92

Evidence Integrity0.90

Compliance0.88

Independence0.85

Confidentiality0.85

Access Control0.82

Coverage

Anomaly Detect

Transparency

Remediation

Audit Speed

Continuous Mon

Trace Coverage

Evidence Link

Control Failure

Anomaly FPR

Time to Report

Conflict

Speed vs Coverage

Conflict

Monitor vs Privacy

Conflict

Detect vs Indep

Conflict

Fix vs Transparency

Audit value lies in evidence integrity before discovery.

Value Scan implements this as the highest-level constraint.

Use Case 3/12

Manufacturing

Quality ControlMaintenanceProcess OptSupply ChainFloor AISafety Mgmt

L0 CoreL1 StrategyL2 Metrics

Worker Safety0.92

Product Quality0.90

Safety Standard0.88

Lot Traceability0.85

OT Resilience0.85

OT Security0.82

Throughput/OEE

Yield

Downtime Reduce

Pred Maintenance

Lead Time

Floor Explain

OEE

Scrap Rate

Defect PPM

MTBF

Safety Incident

Conflict

Prod vs Safety

Conflict

Time vs Quality

Conflict

Uptime vs Security

Conflict

Explain vs Prod

Manufacturing decisions must be stoppable on the floor.

Value Scan implements this as the highest-level constraint.

Use Case 4/12

Auto-Dev

Code GenPR ReviewRefactorTest GenSecurity ScanRollback

L0 CoreL1 StrategyL2 Metrics

Reproducibility0.95

Rollbackable0.92

Auditability0.90

No Secrets Leak0.90

Idempotent0.88

Policy Pinned0.85

Code Quality

Test Coverage

Review Speed

Security Scan

Explainability

Staged Rollout

Approval Rate

Rollback Rate

MTTR

Regression Freq

Policy Drift

Conflict

Speed vs Audit

Conflict

Quality vs Deploy

Conflict

Security vs Speed

Conflict

Explain vs Quality

Auto-Dev is not 'AI writes code.' Auto-Dev is 'the database authorizes changes.'

Value Scan implements this as the highest-level constraint.

Use Case 5/12

CEO Decision OS

Vision AlignmentPhilosophy EncodingStrategy SimulationStakeholder TrustValue ConsistencyBoard Reporting

L0 CoreL1 StrategyL2 Metrics

Vision Alignment0.95

Mission Integrity0.92

Core Values0.90

Stakeholder Trust0.88

Accountability0.85

Sustainability0.85

Risk Tolerance

Social Impact

Profit Priority

Innovation Drive

Transparency

Long-term View

Vision Gap Score

Value Drift Rate

Decision Alignment

Stakeholder NPS

Philosophy Score

Conflict

Profit vs Social

Conflict

Risk vs Trust

Conflict

Innovation vs Values

Conflict

Long-term vs Short

Your vision becomes the operating system. Philosophy-driven decisions, encoded into every judgment.

Value Scan implements this as the highest-level constraint.

Use Case 6/12

Healthcare

Clinical DecisionDrug InteractionDiagnosis SupportTreatment PlanResource AllocationPatient Privacy

L0 CoreL1 StrategyL2 Metrics

Patient Safety0.98

Do No Harm0.96

Clinical Accuracy0.94

HIPAA Compliance0.92

Informed Consent0.90

Care Continuity0.88

Diagnosis Speed

Treatment Efficacy

Cost Efficiency

Patient Experience

Provider Workload

Evidence-Based

Misdiagnosis Rate

Adverse Events

Readmission Rate

Wait Time

Outcome Score

Conflict

Speed vs Accuracy

Conflict

Cost vs Safety

Conflict

Load vs Continuity

Conflict

Efficacy vs Consent

Patient safety is not negotiable. Every AI decision passes through the Hippocratic gate.

Value Scan implements this as the highest-level constraint.

Use Case 7/12

Legal & Compliance

Contract ReviewRegulatory FilingRisk AssessmentDue DiligencePolicy EnforcementLitigation Support

L0 CoreL1 StrategyL2 Metrics

Legal Compliance0.95

Confidentiality0.94

Audit Trail0.92

Regulatory Adherence0.90

Conflict Check0.88

Jurisdiction Aware0.85

Review Speed

Risk Identification

Cost Control

Analysis Accuracy

Client Service

Precedent Match

Error Rate

Review Time

Compliance Score

Risk Exposure

Matter Cost

Conflict

Speed vs Trail

Conflict

Cost vs Accuracy

Conflict

Service vs Privacy

Conflict

Precedent vs Local

Regulatory compliance is the foundation. Every decision is audit-ready from inception.

Value Scan implements this as the highest-level constraint.

Use Case 8/12

Retail & E-commerce

Pricing StrategyInventory MgmtPersonalizationFraud PreventionSupply ChainCustomer Support

L0 CoreL1 StrategyL2 Metrics

Customer Trust0.90

Fair Pricing0.88

Data Privacy0.90

Product Safety0.92

No Manipulation0.85

Transparency0.82

Conversion Rate

Customer LTV

Inventory Turn

Personalization

Delivery Speed

Return Rate

Cart Abandon

NPS Score

Margin

Stockout Rate

Fulfillment Cost

Conflict

Convert vs Ethics

Conflict

Personal vs Privacy

Conflict

Margin vs Fair

Conflict

Speed vs Inventory

Customer trust drives every recommendation. Personalization without manipulation.

Value Scan implements this as the highest-level constraint.

Use Case 9/12

Energy & Utilities

Grid ManagementDemand ForecastAsset MaintenanceRenewable IntegrationEmergency ResponseRate Optimization

L0 CoreL1 StrategyL2 Metrics

Grid Reliability0.99

Public Safety0.98

Regulatory Compliance0.95

Cyber Security0.94

Environmental0.88

Service Equity0.85

Efficiency

Sustainability

Cost Management

Demand Response

Asset Lifespan

Renewable %

Outage Duration

SAIDI

Carbon Intensity

Peak Demand

Maintenance Cost

Conflict

Efficiency vs Reliable

Conflict

Cost vs Green

Conflict

Renewable vs Secure

Conflict

Demand vs Equity

Grid reliability is non-negotiable. Sustainability meets operational resilience.

Value Scan implements this as the highest-level constraint.

Use Case 10/12

Insurance

UnderwritingClaims ProcessingFraud DetectionRisk ModelingPolicy PricingCustomer Retention

L0 CoreL1 StrategyL2 Metrics

Fair Assessment0.92

Regulatory Compliance0.94

No Discrimination0.95

Data Accuracy0.90

Claims Integrity0.88

Privacy0.88

Underwriting Speed

Claims Efficiency

Loss Ratio

Customer Satisfaction

Fraud Detection

Retention Rate

Combined Ratio

Claims Cycle Time

Fraud Rate

NPS

Expense Ratio

Conflict

Speed vs Fair

Conflict

Fraud vs Claims

Conflict

Loss vs Satisfy

Conflict

Efficient vs Fair

Fair assessment, transparent pricing. Risk decisions that policyholders can trust.

Value Scan implements this as the highest-level constraint.

Use Case 11/12

Education

Learning SupportCareer GuidanceCurriculum AdaptProgress TrackHuman GateAptitude Match

L0 CoreL1 StrategyL2 Metrics

Student Wellbeing0.95

Educational Equity0.92

Data Privacy0.90

Pedagogical Integrity0.88

Parental Consent0.88

Age Appropriate0.90

Learning Outcomes

Engagement

Personalization

Career Readiness

Skill Development

Teacher Support

Completion Rate

Skill Gap Score

Career Match Rate

Intervention Success

Path Flexibility

Conflict

Personal vs Equity

Conflict

Career vs Wellbeing

Conflict

Engage vs Pedagogy

Conflict

Outcomes vs Age

Students are not single recommendation vectors. Education decisions are managed as states within a neural decision network.

Value Scan implements this as the highest-level constraint.

Use Case 12/12

Municipality

Migration MatchJob PlacementSettlement SupportPolicy ImpactResource AllocationCommunity Integration

L0 CoreL1 StrategyL2 Metrics

Citizen Welfare0.95

Fairness0.92

Legal Compliance0.94

Privacy Protection0.90

Transparency0.88

Accountability0.88

Match Quality

Settlement Success

Economic Impact

Community Cohesion

Service Efficiency

Long-term Retention

Placement Rate

1-Year Retention

Satisfaction Score

Integration Index

Cost per Placement

Conflict

Quality vs Speed

Conflict

Economic vs Welfare

Conflict

Retention vs Fair

Conflict

Cohesion vs Transparent

Migration and employment are not one-way matching. They are decision networks with time and state that can be paused and reviewed.

Value Scan implements this as the highest-level constraint.

L0: Core

L1: Industry

L2: CEO

L3: Ops

Collective Value Sphere

Values are never mixed.
Processed in order.

Values used in decisions are structured as layers.

Layer 0

Human life, irreversible loss, legal - No one can override

Layer 1

Industry-specific regulations, safety, audit

Layer 2

CEO values - Challenge, long-term, speed, aesthetics

Layer 3

Project discretion - Cost, effort, ops load

Upper layer values are never optimized away by lower layers.
That's the design of "unbreakable decisions".

Human LifeComplianceIrreversibleAudit ReqSafety StdRegulationQuality StdLong-term

Adapt to Context. Never Compromise Principles.

Organizations change priorities over time. MARIA OS adapts automatically.

17:55

Normal

Peak

Month-end

ACTIVE

Balanced

throughput

0.30

latency

0.30

human Load

0.25

cost

0.15

Deep values: unchanged

Month-end18:00-23:00

throughput

0.55

latency

0.15

human Load

0.20

cost

0.10

Deep values: unchanged

Incident Response

throughput

0.15

latency

0.50

human Load

0.30

cost

0.05

Deep values: unchanged

Balanced optimization active.

Change priorities without changing who you are.

Neural Decision Pipeline

Decision Flow Processing

InputScanFilterOutput

input

scanning

filtering

output

Neural decision pipeline ready

Input Layer

Value Scan

Constraint Filter

Decision Output

Every decision flows through the same neural pipeline: Input → Scan → Filter → Output

Constraints are applied in order. L0 values cannot be overridden by L2 optimizations.

DECISION CONTROL

Why MARIA OS CanStop Decisions

We define when decisions must not execute as a structural constraint.

MARIA OS evaluates whether a decision is in an executable state, not just "correct."

Premise Integrity

Are all required premises present and consistent?

STOP

Decision Stability

Does the conclusion remain stable under slight condition changes?

HUMAN_REVIEW

Impact x Irreversibility

Is this a high-impact decision that cannot be undone?

CEO_GATE

Philosophy Alignment

Does it deviate from core values or strategic direction?

CEO_REVIEW

Explainability

Can we explain why this decision was made to third parties?

STOP

All Passed

Execution proceeds

AUTO_EXECUTE

STOP

HUMAN

EXEC

MARIA OS exists to reliably stop decisions that must not be executed.

IMPLEMENTATION

How MARIA OS Stops Decisions

Decisions as State, not results

decision_state.ts

1if (PremiseCompleteness < 0.95) { STATE = "STOP" }
2if (DecisionStability < 0.90) { STATE = "HUMAN_REVIEW" }
3if (ImpactScore > 0.7 && Irreversible) { STATE = "CEO_GATE" }
4if (PhilosophyDeviation > threshold) { STATE = "CEO_REVIEW" }
5if (ExplainabilityScore < 0.99) { STATE = "STOP" }
6if (allChecksPassed()) { STATE = "AUTO_EXECUTE" }

AI Agent

Control

Policy

State

Human

CEO

Board

Control layer exists outside of AI. AI is the proposer.

"Must not execute" is a state.

Model Architecture

Fast × Heavy Model Orchestration

Models are separated by role, not capability.

Don't breakCan stopExplain themselves

That's why we separate models by role, not by power.

Fast Model

Preprocessing

Prerequisite Organization

Input Normalization

State Updates

Low LatencyDeterministicReplaceable

Heavy Model

Reasoning

Trade-off Analysis

Reasoning Generation

Counterfactual Scenarios

High ReasoningAbstractionHigh Cost

Decision Flow

Scroll

Input

Fast Model

State Evaluation

OK → Continue

Unstable → Call Heavy

Heavy Model

Decision Gate

AUTOREVIEWESCALATE

Why separation enables stopping

Smart models never execute

Fast models never judge

Authority lives outside models

Human gates are enforced

Neither model is permitted to execute.

Control architecture, not model architecture.

PERFORMANCE METRICS

How to Measure a Decision OS

"Can it reliably stop decisions that should not be executed?"

Axis

MARIA

General

Stop Recall

High

Undefined

False Execute Rate

Minimal

Out of scope

Decision Stability

Quantified

Ad-hoc

Explainability Robustness

Audit-ready

Black box

Decision Transferability

Structured

Person-bound

Why MARIA OS Can Stop

Premises in place?

Decision stable?

High impact?

Philosophy aligned?

Explainable?

Spec Sheet

UnitState Machine

StopMulti-layer

HumanRole Gates

AuditAuto Trail

An OS that controls whether decisions should be executed.

TECHNICAL DEEP DIVE

Control Logic & Implementation Design

Not ML evaluation — control systems engineering.

5 Independent Evaluation Layers

Premise Consistency前提整合性評価

completeness < threshold → STOP

Verify that all premises required for judgment are in place and not contradictory.

premise_completeness_ratepremise_conflict_flagpremise_drift_distance

State Machine Control

1// State Machine - Decision Control
2enum DecisionState {
3  AUTO_EXECUTE,    // Full automation
4  HUMAN_REVIEW,    // Requires human check
5  CEO_REVIEW,      // Escalate to CEO
6  BOARD_REVIEW,    // Board-level decision
7  STOP             // Halt execution
8}
9
10// Models have NO execution authority
11// Only OS layer manages state transitions
12function evaluateDecision(context: Context): DecisionState {
13  const premise = checkPremiseConsistency(context);
14  const stability = checkDecisionStability(context);
15  const impact = checkImpactIrreversibility(context);
16  const philosophy = checkPhilosophyDeviation(context);
17  const explain = checkExplainability(context);
18

Fast / Heavy Separation

Fast

Premise org, state updates, deterministic

Heavy

Deep reasoning, on-demand, high-cost

Neither model has execution authority.

A control OS that prevents decisions from breaking.

Reproducible

Measurable

Comparable

The Core of MARIA OS

From Judgment to Scalable Autonomy

AI is already autonomous.

The question is: can your organization trust it?

What organizations face today:

Decisions become black boxes

Exceptions depend on individuals

Ops cannot be replicated

Responsibility blurs at scale

Not an AI problem.
The absence of structured judgment.

Not about making AI smarter.

It is not "a platform to optimize AI agents."

Condense judgment into structures

Transplant into an operating system

Scale safely through autonomous agents

That is the essence of MARIA OS.

Autonomy scales.
Responsibility does not.

Up to 10,000 AI agents per organization. But execution only scales.

Decision points

→ Decision Axis

Approvals

→ Responsibility Gates

Human decisions

→ "human-sized"

Execution scales.
Responsibility stays human-sized.

AI operations run without exhaustion, without breaking, with full replicability.

Universe Builder creates
replicable AI orgs.

Not just a group of agents.

Universe =

Complete operational unit: judgment, responsibility, optimization, approval.

Universe Builder flow:

ReqDesignAgentsGatesValidDeploy

Once working, replicate across departments and environments.

HITL is not a fallback.
A collaboration point.

Not a mechanism to compensate for AI failure.

Designed from the start:

Which decisions need human

Why human is needed

Which role handles it

What they see/decide

Meaningful decisions

Quality assurance

Learning assets

AI and humans collaborate, not compete.

Why traditional approaches fail

Aspect	Traditional	MARIA OS
Judgment	Implicit	Explicit Decision
Responsibility	Person-dep.	Gates
Optimization	Everything	Surface only
Values	Documents	Constraints
Reuse	Difficult	Per-Universe
Audit	After fact	Always on
Scale	Chaotic	Stable

How organizations transform

Before

More AI = More anxiety

Judgment person-dependent

Ops not reproducible

After

More AI = More stability

Judgment in structure

Universes replicable

Not an improvement.
A transformation.

Summary

What is MARIA OS?

Elevates judgment to OS

Locks values as constraints

Scales autonomy safely

Human-AI collaboration

MARIA OS is a

Decision Engine OS

Does not automate judgment.

Makes it scalable, governable, reusable.

Architecture Overview

Scaling Autonomy, Not Just Agents

Extract Values → Condense Constraints → Deploy to 10,000 Agents

Universe

Organizational scope & responsibility boundary

Decision Axis

Condensed judgment & value alignment

Responsibility Gate

Human authority & approval checkpoints

AI Agents

Autonomous workers operating under governance

MARIA OS scales AI agents, not responsibility.

Every agent operates within defined boundaries. Human authority remains at the Gate.

Where MARIA OS Fits

Mapping the AI Operations Stack

Who is responsible?

Execution →

Governance →

Low / High

High / High

Low / Low

High / Low

Agent Frameworks

Workflow Engines

Observability

Governance / GRC

MARIA OS

Decision Engine OS

Select category

Tap a point

Comparison

AspectAgentFlowObsMARIA

Primary BuildExecuteObserveGovern

DecisionImplicitAssumedPost-hocFirst-class

ApprovalNoneExceptionDisplayBy Design

Not competing inside the AI tooling stack.

A new layer: Decision Engine OS.

Technical Prerequisites

Prerequisites for Adoption

MARIA OS is not an AI you plug in for convenience. It's an OS for governing decisions. These are not constraints—they are conditions for reliability and long-term operation.

Strong consistency

Judgment validity, approvals, execution rights

Eventual consistency

Visualization, UI, operational views

We don't sacrifice speed. We separate correctness from speed.

Ease of use vs Ease of operation—we choose the latter.

That is the design philosophy of MARIA OS.

Frequently Asked Questions

FAQ

When AI decides, responsibility must be designed.
MARIA OS is built for that reality.

Automation is easy.

Trust is hard.

MARIA OS exists to make AI trustworthy.

Not by slowing AI down, but by making responsibility explicit.

DECISIONINTELLIGENCEOS

Watch a zone come to life

Core Features

Value Hierarchy Processing

Responsibility Gate

Decision Pipeline

Universe Architecture

Database-Native

Skills: Dynamic Fetch & Refilling

Complete Audit Trail

Phase-Aware Context

Collective Intelligence

Execution Flow

12 Industry Use Cases

Real-time visibility into every AI decision

Skills are fetched or generated on demand

Dynamic Skill Fetch

Requirement

Store Search

Resolve/Create

Bind to Agent

Skill / Phase Permission Matrix

Universe > Planet > Zone > Agent

Agent execution power never depletes

Capability Mismatch

Quality Degradation

Phase Mismatch

Environment Change

Watch skills flow to Agents

Execution Flow Circuit

Why Autonomous Development Completes

True Completion = 5 Metrics Multiplied

12 Use Cases

CEO Decision OS

Finance

Healthcare

Audit

Legal & Compliance

Manufacturing

Retail & E-commerce

Energy & Utilities

Insurance

Auto-Dev

Education

Municipality

Judgments are always traceable

Judgment Phase vs Responsibility Phase

Universe View: Gates at Every Boundary

Evidence Package: Submit-Ready from Day One

Judgment & Responsibility OS Structure

Judgment does not scale. Execution does.

Condense judgment. Make it executable.

Design judgment as executable code

Observe, debug, and control

Threshold Stop: Amount exceeds limit

Conflict Resolution as a Common Language

Conflict Classification

Transform Conflicts into Structured Decisions

Scheduled Knowledge Cleansing

Each Loop Ascends to Higher Dimensions

Ascend Through Scales of Impact

Tick-Driven Knowledge Loop

Phase Alignment Loop

Impact Rescoring

Human-in-Loop Feedback

Seven Pillars of Reliability

Decision Maturity Levels

Agentic Skills in the Real World

Autonomous execution, governed by design.

Universe Builder: from judgment to organization

Configure governance before deployment

Human-in-the-Loop Points

Scale Without Losing Control

You don't add agents. You grow universes.

What MARIA Does Differently

Decisions are first-class objects

What you see in MARIA OS

Human-in-the-Loop Is Not an Exception

Governance That Matches Real Organizations

Evidence, Audit, and Trust

DECISION
INTELLIGENCE
OS

What Does Your Organization
Actually Value?

Values are never mixed.
Processed in order.

Autonomy scales.
Responsibility does not.

Universe Builder creates
replicable AI orgs.

HITL is not a fallback.
A collaboration point.