MIDAS Explorer: Understanding and Governing Autonomous Decisions

Introduction

The MIDAS Explorer is a browser-based interface that lets you interact with the MIDAS governance engine in real time.

It is designed for:

• developers integrating AI or autonomous systems
• architects evaluating governance approaches
• reviewers assessing how decisions are controlled and audited

Unlike a typical API console, the Explorer is not just about sending requests — it is about understanding:

• why a decision was allowed
• how authority was resolved
• what governance record was produced

What MIDAS Actually Does

At runtime, MIDAS answers a single question:

Does this actor have the authority to execute this decision right now?

This is fundamentally different from other systems:

• Rules engines decide what should happen
• Policy engines decide what is allowed
• MIDAS decides whether execution is authorised

It sits downstream of decision logic, governing whether an action can proceed.

Getting Started

Run MIDAS locally and open:

http://localhost:8080/explorer

When the Explorer loads:

• A demo scenario runs automatically
• You immediately see a governed outcome
• The interface shows how MIDAS reached that decision

If authentication is enabled:

• A login overlay appears
• The Explorer is blurred and non-interactive until login
• Once authenticated, the full interface is available

Explorer Layout

Left Panel — Build and Understand Requests

• Scenarios: pre-built governance examples
• Request Builder: form or JSON
• Resource Panel: the authority model behind the scenes

Right Panel — Inspect Decisions

• Outcome
• Explanation
• Threshold evaluation
• Authority chain
• Raw response
• curl commands
• Decision record (envelope)

Try This First: Scenarios

The fastest way to understand MIDAS is through scenarios.

Threshold Example — Below Confidence

• Confidence: 0.30
• Required: ≥ 0.85

Result: Escalate

Why? The agent is not confident enough — so MIDAS routes to human review.

Consequence Example — High Impact

• Amount: £2,500
• Limit: £1,000

Result: Escalate

Even with high confidence, the impact is too large for autonomous execution.

Context Example — Missing Required Data

Result: Request Clarification

MIDAS does not guess — it requires complete context before evaluating authority.

Evaluate vs Simulate

This is one of the most important concepts.

Evaluate

• Creates a governance record (envelope)
• Writes audit data
• Represents real execution

Simulate

• No persistence
• No audit
• Safe experimentation

You can take a real decision and click “Simulate from this” to test variations.

Understanding a Decision

Outcome

MIDAS produces four outcomes:

• Accept
• Escalate
• Reject
• Request Clarification

Explanation Panel

This is where MIDAS becomes powerful.

You see:

• confidence vs threshold
• consequence vs limit
• which factor drove the decision

Example:

Confidence: 0.30 ≥ 0.85 ✗  Consequence: £100 ≤ £1,000 ✓  Outcome driver: confidence threshold

Authority Chain

Every decision resolves a chain:

Surface → Profile → Grant → Agent

This shows:

• what decision boundary was used
• what rules applied
• who was authorised
• why execution was permitted or denied

Governance Envelope (Decision Record)

Every evaluated decision produces a governance envelope.

The envelope includes:

• outcome
• reason
• authority chain
• evaluation data
• timestamp

This is your audit trail.

MIDAS makes decisions explainable and traceable by design

The Authority Model (Why Decisions Behave This Way)

The resource panel shows the model behind everything:

• Surface → what type of decision
• Profile → thresholds and rules
• Agent → who is acting
• Grant → who is allowed to act

These directly drive the scenarios you run.

Example Walkthrough

Low Confidence Decision

Input

{

 "confidence": 0.30,

 "amount": 100

}

Evaluation

• Confidence below threshold
• Consequence acceptable

Outcome
→ Escalate

Why
The agent is uncertain — a human must decide.

Troubleshooting

401 Unauthorized
→ Token missing or expired
→ Re-authenticate

Invalid JSON
→ Syntax error
→ Use form mode

INSUFFICIENT_CONTEXT
→ Required fields missing
→ Add context

SURFACE_NOT_FOUND
→ Invalid surface ID
→ Use demo scenarios

Key Takeaways

• MIDAS governs execution authority, not decision logic
• Authority chains are explicit and inspectable
• Decisions are explainable through threshold evaluation
• Outcomes are captured as governance records
• MIDAS fits naturally into cloud-native control planes

Closing

The MIDAS Explorer is not just a testing tool — it is a window into governed autonomy.

It shows how decisions can be:

• controlled
• explained
• audited
• trusted

‍