Skip to main content

​
Overview

Exploratory Data Analysis (EDA) is how Dexter works to understand a data source before anything is built on it. Dexter profiles the data, raises quality problems and codes it can’t interpret, and confirms its reading with you. Then it produces whatever you asked for. EDA is not a saved object. There is no “EDA record” in the platform. What lasts is what the analysis produces, and that depends on your request. Dexter’s approach is opinionated for a reason. Manufacturing data is often dirty, misleadingly labeled, and full of codes whose meaning lives in someone’s head rather than in the file. So Dexter profiles each source before extracting from it, asks you about ambiguities instead of assuming them away, and doesn’t present a number without knowing where it came from.

​
Starting an Analysis

  • Analyze your new data? When an upload you started on the Data page finishes, this prompt appears in the corner of the screen. Analyze it asks Dexter for a full exploratory analysis of the new files.
  • Create with Dexter. On the Data page, pick the uploads and integrations to analyze and describe what you want out of them.
  • Chat. Attach files or name existing uploads and ask Dexter to explore them. On a new chat, Explore uploaded data under Suggestions pre-fills a prompt for you.
The depth follows your request. Dexter runs a full exploratory analysis when you ask for one, or when your request builds on data it doesn’t yet understand. A quick question about a file gets a quick answer. The Data page’s Analyze your new data? prompt appears in every mode. See Response modes.

​
When to Use EDA vs. a Quick Question

The line isn’t rigid. A quick question can turn into an exploratory analysis as soon as it’s clear the source needs real understanding before the answer can be trusted.

​
Step by Step

1

Orient

Dexter scans the files and confirms anything you haven’t already said about where the data comes from and what you want from it. This sets what the analysis should show.
2

Profile each source

Dexter profiles every source before anything builds on it: what one row represents, what the columns mean, quality problems, and how the sources connect. For larger jobs, Dexter can split the investigation across sub-agents.
3

Fill gaps with you

Dexter raises each data-quality issue, and each code or term it can’t confirm from the data, as a question before any work builds on the affected data. If your answers show it misread a source, it goes back and profiles again.
4

Present the findings

Dexter presents the results in the form your request needs, along with the evidence and the limitations.
5

Put them to work

If the request involves further work in ProDex, Dexter uses the findings directly, for example to build a simulation model or to set up a pipeline.

​
What You Get

What EDA produces is shaped by what you asked for:
  • An answer to your question, in chat, with its evidence and limitations.
  • Parameters for a model, such as service times, yields or rates for a simulation model built from your data.
  • A report documenting a targeted analysis, saved in the Reports folder under Files. See Reports.
  • A dashboard, when exploring or monitoring calls for one. Where custom modules are enabled, Dexter builds it as a module, often from the Data insights template, with the query or code behind each chart shown beside it. See Custom modules.
  • A pipeline, only when the same data will arrive again and should be prepared the same way each time. See Pipelines.
Every analysis also leaves you with a confirmed reading of the source’s codes, units and quirks, agreed with you during the analysis.

​
Limits and Caveats

  • EDA runs through Dexter. The Data page prompts and the chat suggestion all start a conversation. There is no separate EDA screen.
  • Source quality is a hard floor. If a source is missing the columns a figure needs, or a critical code can’t be resolved, Dexter surfaces the gap instead of making up a number.
  • Rows aren’t dropped silently. If rows fail to parse or are excluded, Dexter counts them, inspects them, and tells you why the loss doesn’t matter.
  • Unchecked results are called unchecked. When Dexter reproduces a calculation you normally run elsewhere, it checks the result against a real output of yours. If there’s nothing to check against, it says so.
  • Joins across sources need your confirmation. Joining MES to ERP, or a cycle log to work orders, needs you to confirm what the join key means. Dexter won’t silently guess.
  • Macros are read before they run. Dexter reads an uploaded workbook’s macro code before running it. It shows you any macro that reaches outside the workbook or runs when the file opens, instead of running it.