LEO local demonstration

Procurement & Accounting Anomaly Review Dashboard

LEO reviews procurement and accounting records, detects unusual patterns, explains why they may matter, and prepares evidence for a human reviewer. It does not decide guilt, block payments, mutate records, or replace institutional judgment.

What LEO analyzes Prices, quantities, supplier capacity, declared business activity, threshold-near fragments, and evidence consistency.

What LEO shows Risk signals, affected suppliers, supporting evidence IDs, reason text, review priority, and a human review memo.

What humans decide Whether a signal is justified, should be escalated, is a false positive, or requires more documents.

Institutional value LEO helps an organization notice weak procurement records before they become audit, governance, donor, or legal problems.

Practical use A reviewer can prioritize work, request missing documents, and preserve a transparent evidence trail.

Control model The system supports human judgment. It does not replace management, accounting, legal review, or formal audit authority.

Input

Demo procurement/accounting records, supplier data, benchmark context, and transaction facts.

Analysis

LEO compares values, detects mismatches, groups fragments, and identifies review-worthy anomalies.

Evidence

Each signal receives an evidence ID, severity, supplier reference, and reason text.

Human review

A reviewer accepts, escalates, marks false positive, or requests more evidence.

Report

The dashboard generates a local memo and optional export package without production mutation.

Institutional view mode

Choose how much detail to show. Executive View is designed for officials, auditors, board members, and non-technical institutional reviewers.

EXECUTIVE VIEW

Executive Summary

One-screen orientation for institutional reviewers. This is a review summary, not a legal or audit conclusion.

HUMAN REVIEW REQUIRED

Source Trust and Provenance

LEO analyzes provided data and shows how strongly that data should be trusted. Weak provenance means strong conclusions must remain restricted.

GUARDED

Priority Review Queue

What the reviewer should look at first. This queue prioritizes human review work; it does not issue verdicts.

REVIEW QUEUE

Review Completion

Package-level workflow status for the human reviewer. This shows review progress and blockers; it does not create approval, legal conclusion, or audit certification.

REVIEW_IN_PROGRESS

Export Readiness Gate

Pre-export warning layer. Export remains local and allowed with warnings, but unresolved findings, weak source provenance, and prototype limits must remain visible.

EXPORT_READY_WITH_WARNINGS

Reliability state

Deterministic reliability visibility for human review. This state supports interpretation boundaries only and does not create proof, legal conclusion, audit certification, fraud verdict, or production validation.

NOT_LOADED

Findings checked

Finding-level checks

Guarded

Usable with caution

Degraded

Context weakened

Limited / blocked / unsafe

Strong interpretation restricted

Input quality

NOT_LOADED

Input state affects confidence

Boundary

REVIEW_NOT_DECISION

Reliability is not authority

Donor / Grant Protocol Library

Local reviewer criteria and protocol references. These rules provide reviewer questions and documentation context only; they do not create donor compliance verdicts, legal conclusions, expense eligibility decisions, payment blocking, supplier sanctions, or autonomous enforcement.

REVIEW_CRITERIA_LIBRARY

Protocol ID

DONOR_GRANT_REVIEW

Local protocol reference

Version

v0.1

Reviewer criteria library

Rules

Concrete protocol rules

Linked findings

Display context only

Linked rule references

Unique rules mapped to loaded findings

Boundary

NO_VERDICT

Review criteria, not authority

Audit Event Timeline

Local review chronology for reconstructing how the package became visible, how reviewer actions were recorded, and how export context was generated. This is not legal certification, external audit submission, fraud verdict, enforcement, production mutation, or canonical registry mutation.

LOCAL_REVIEW_TIMELINE

Timeline events

Package + reviewer chronology

Reviewer actions

Browser-local annotations

Unresolved findings

Unresolved is not non-compliance

Phase-order events

Deterministic when timestamp unavailable

Export Package Integrity Manifest

Content and boundary manifest for the local human review package. v0.1 describes package contents only; it does not create hash, signature, legal certification, audit certification, donor approval, production mutation, or canonical registry mutation.

MANIFEST_V0_1

Manifest version

v0.1

Content + boundary manifest

Reviewed / unresolved

0 / 0

Local browser review state

Hash / signature

No cryptographic proof in v0.1

Archive guidance

GUIDE

Reviewer saves manually

Findings

Signals requiring review

Evidence objects

Parsed from demo report

Suppliers

Distinct supplier references

High+ risk

High or critical findings

Human-reviewed

Local review actions

Autonomous actions

Must remain zero

Runtime operational state

Local prototype state visibility. This panel exposes unresolved, degraded, recovery, export, and boundary status without creating production authority.

LOCAL PROTOTYPE

Report status

NO_REPORT_LOADED

Unresolved findings

Unresolved does not mean guilt or approval failure.

Degraded state

DEGRADED_NO_REPORT

Missing data is not treated as no risk.

Recovery state

RECOVERY_MANUAL_RELOAD_REQUIRED

Recovery is local reload, not state erasure.

Local persistence

UNKNOWN

Browser-local only, not institutional archive.

Export state

EXPORT_NOT_GENERATED

Export is not external submission.

Input quality

NOT_LOADED

Unavailable input quality does not validate input.

Autonomy boundary

ZERO_AUTONOMY_CONFIRMED

No execution, registry mutation, signing, or external action.

Input quality status

Read-only data readiness check for the CSV files used before LEO generates evidence signals. This section does not issue risk, fraud, legal, or enforcement conclusions.

Not loaded

CSV files

n/a

Procurement records

n/a

Supplier profiles

n/a

Benchmarks

n/a

Errors

n/a

Warnings

n/a

Input quality report not loaded yet.

How to read this demo

A first-time reviewer can follow this sequence without knowing the internal LEO architecture.

Step 1

Read the summary

Start with findings, suppliers, high-risk count, and autonomous actions.

Step 2

Check what LEO found

Look at signal families: price, supplier capacity, activity mismatch, quantity, and fragmentation.

Step 3

Review priorities

Open the strongest suppliers and findings before reviewing medium-risk items.

Step 4

Inspect evidence

Use evidence IDs to trace each signal back to the demo report.

Step 5

Record human judgment

Mark each finding as justified, escalated, false positive, or needing more evidence.

Step 6

Export memo

Export a local human review package without changing production or canonical records.

This is not a fraud verdict

Correct interpretation boundary for external viewers.

LEO does not state that fraud occurred. LEO states that a record or pattern differs from expected procurement/accounting behavior and deserves human review.

Legal conclusionNOT ISSUED

Fraud accusationNOT ISSUED

Payment blockNOT PERFORMED

Human reviewREQUIRED

What LEO found

Plain-language summary generated from the loaded evidence report.

Recommended review focus

Where a human reviewer should look first.

Top review priorities

The strongest individual findings to inspect first.

Pattern cluster summary

Repeated signals that should be understood as a group, not isolated rows.

Findings requiring review

Each card explains the signal, the risk meaning, the evidence reference, and the next human action.

Relationship graph view

Local read-only map of suppliers, findings, evidence, source records, signal families, and threshold clusters. Graph edges show review context, not legal causation.

GRAPH_VIEW_FOR_HUMAN_REVIEW_ONLY GRAPH_EDGE_NOT_LEGAL_CAUSATION SUPPLIER_NODE_NOT_SUPPLIER_GUILT CLUSTER_NOT_FRAUD_VERDICT LOCAL_PROTOTYPE_ONLY

Supplier filter Signal family filter

Supplier risk overview

Suppliers grouped by evidence references and maximum detected severity.

Supplier	Finding count	Max severity	Evidence references	Review state

Evidence object index

Raw evidence object summary. Full raw payload remains in the loaded local browser session only.

Evidence ID	Detected signal	Severity	Supplier / subject	Reason

Reviewer memo preview

Generated locally from loaded evidence and human review choices.