---
title: 'Top Problem Nodes'
description: Fleet operational health summary — a 0-100 score across battery, comms, data quality and passport completeness, ranking the nodes that need attention first.
section: AI Analytics
weight: 2
related:
  - ai-analytics/node-reports/consumption-analytics
  - ai-analytics/fleet-reports/suspicious-nodes
  - ai-analytics/node-reports/battery-forecast
---

import Alert from '@/components/docs/Alert.astro';

Fleet nodes ranked by **operational health**: whose battery is dying, who dropped off the network, who has bad data or an empty device passport. It is built for operations and dispatch — open it in the morning, see what to fix first, then drill into a node's full analytics.

<Alert type="note">
Metering-bypass signs (under-metering, tampering events) live in the dedicated [Suspicious Nodes](/en/platform/v3/ai-analytics/fleet-reports/suspicious-nodes) report — that is where violators are hunted. This report covers only what operations can fix. That is why "dead" nodes do appear in this ranking: a silent node is a prime repair candidate, even though it is useless for bypass detection.
</Alert>

## The four dimensions

| Dimension | Weight | What it evaluates |
|---|---|---|
| Battery | 30% | last charge (mV) and drop rate over the window; below 3000 mV — critical |
| Comms | 30% | days since the last session + share of gaps in the hourly archive |
| Data quality | 20% | out-of-range values, stuck sensors, pressure/temperature spikes |
| Passport | 20% | device passport completeness (max flow, serial number, gas composition…); an empty passport means an inventory visit is needed |

The overall 0-100 score is a weighted sum: ≥60 is critical, 30-60 is watch, and below 30 is normal.

## How it works

A two-stage scheme keeps the load light. First a **pre-selection** of the whole fleet by metadata (session age, device status) — no heavy requests. Then a **deep analysis** of the top-200 candidates (hourly archive and link sessions) in several parallel workers. The final table is built from that result.

## Run parameters

| Parameter | Default | Meaning |
|---|---|---|
| Period from / to | last 30 days | analysis window |
| How many nodes in the table | 100 | final table size |
| Nodes for deep analysis | 200 | candidates to analyze after pre-selection |
| Parallel download workers | 8 | lower to 2-4 on HTTP 429 |
| Utility company | from context | fleet filter |

## What's inside

A level distribution, an AI summary and conclusions, and a table with four dimension badges per node. Each node carries an **Analytics** button that opens its full breakdown — events, pressure, temperature and consumption.