# Enterprise AI Interface Kit

> An original, public-safe product design artifact by Kevin Astuhuaman for reasoning about uncertainty, authority, and recovery in enterprise AI interfaces.

## Canonical resources
- Live interface: https://kevinastuhuaman.github.io/enterprise-ai-interface-kit/
- Pattern specification: https://kevinastuhuaman.github.io/enterprise-ai-interface-kit/patterns.json
- Project metadata: https://kevinastuhuaman.github.io/enterprise-ai-interface-kit/project.json
- Source: https://github.com/kevinastuhuaman/enterprise-ai-interface-kit
- Portfolio: https://portfolio.kevinastuhuaman.com

## What this demonstrates
The reference workflow shows how an AI product can expose evidence provenance, calibrated confidence, role permissions, state-bound approval, observable events, scoped failure recovery, and honest no-data states. It is a coherent operator workflow, not a gallery of visual components.

## Pattern contracts
1. Provenance chain: What evidence produced this output? Contract: Source, observed time, transformation, and current availability
2. Calibrated confidence: How much uncertainty matters for this decision? Contract: Calibrated band, evidence coverage, threshold, and intended action
3. Permission boundary: What can this person and agent do now? Contract: Actor, allowed action, prohibited action, and scope
4. State-bound approval: Who owns the consequential decision? Contract: Owner, exact state binding, expiration, and invalidation rule
5. Observable run trace: What happened without exposing private reasoning? Contract: Timestamp, actor, observable event, input or output reference
6. Scoped failure: What stopped and what remains valid? Contract: Failed boundary, preserved work, retry scope, and owner
7. Honest empty state: What does the system know when evidence is absent? Contract: Absence reason, impact on decision, and bounded next action

## Reference states
- Evidence ready: Ready for operator review. Draft only. Account activation remains unavailable.
- Low confidence: Evidence gap. Read and request evidence. No draft or activation action.
- Approval required: Prepared action. Prepared by the agent. Only the named approver can activate this exact state.
- Scoped failure: Connector interrupted. Operator may retry the failed read. Completed sources remain immutable.
- No evidence: Empty result. Search and request correction only. The agent cannot invent or create a record.

## Role boundaries
- Viewer: can inspect evidence and trace; cannot draft changes or approve actions.
- Operator: can prepare drafts and scoped retries; cannot activate an account or change policy.
- Approver: can approve the exact prepared state; cannot change policy or approve a changed state.
- Admin: can change policy and role assignments; cannot approve a policy change they authored.

## Interpretation guidance
- Confidence is decision context, not authority.
- Approval binds to an exact state and expires after input changes.
- Traces show observable events and references, never private chain-of-thought.
- Failure recovery preserves completed work and retries the smallest safe boundary.
- Missing evidence produces an explicit absence state rather than invented certainty.

## Public boundary
All companies, people, policies, identifiers, evidence, and events in the reference product are fictional. This repository contains no employer assets, customer data, private prompts, credentials, production infrastructure details, or proprietary Trackly code.