Make AI Show Its Receipts

Create a Word report that pairs AI claims with highlighted source passages.

Introduction

30 Second Summary

An AI answer about a long document can sound convincing while hiding the lines that support it. Checking every claim by hand means searching through the source again.

In this project, you will ask GitHub Copilot CLI the same question with or without the eyeball plugin. You will use a PDF plus a web page to generate a Word file that pairs every claim with a highlighted screenshot of its source.

What You'll Build

You will open a Word report where each claim sits beside a highlighted screenshot from the exact passage behind it.

By the end of this project, you'll have:

An auditable document report that lets you check every claim against its source passage.
A side-by-side comparison showing how the same question changes when the answer must provide evidence.
A PDF and web workflow you can run against a document you own or a live page.
Secret Mission: An optional challenge to push your evidence-checking skills further.

Are there any prerequisites?

You need a GitHub account on the Copilot Free plan.

Your machine also needs Python 3.8 or newer plus Microsoft Word or LibreOffice for the generated report.

Before We Start

Document checks move quickly when each prerequisite is within reach. A missing login can interrupt the comparison later.

This step gathers four things the workflow depends on: GitHub access, a terminal, a PDF you own, plus a web page URL. The workflow setup begins after this readiness check.

In this step, get ready to:

Confirm a GitHub account with Copilot Free access.
Confirm that a terminal is available on your computer.
Set aside a PDF plus a web page for the evidence comparison.

Confirm your account and terminal

Copilot CLI uses your GitHub account for access. Your terminal provides the workspace where you will run it.

Your password stays inside GitHub's sign-in flow. You never paste it into this project.

Visit GitHub in your browser.
Create an account through GitHub's sign-up flow if you do not already have one.
Sign in to your GitHub account.
Open the Copilot plan area in your GitHub account settings.
Activate Copilot Free if your account does not already have access.
Press Cmd+Space on macOS or the Windows key on Windows to open app search.
Search for the terminal application installed on your computer.
Select the terminal application from the search results.

You'll see a terminal window with a prompt ready for input. GitHub also shows that your signed-in account can use Copilot Free.

Can't Confirm Both Prerequisites?

If GitHub does not show Copilot Free access, confirm that you completed the account sign-up flow. Return to the Copilot plan area after signing in again.

If your terminal application does not appear in search, check which terminal comes with your operating system. Help me find the terminal on my computer.

Choose your PDF

The PDF is the fixed source for your first comparison. Choose one you own so you can inspect the source beside the resulting report.

Use Finder on macOS or File Explorer on Windows to locate a PDF you own.
Select a PDF that contains several paragraphs of readable text.
Open the PDF in your usual document viewer.
Inspect one page to confirm that the document contains enough detail for a question.
Save the file name here: your PDF file name.

You now have a specific document that opens correctly. Its readable passages give the workflow evidence to capture later.

Save a web page for later

The web page provides a live source for the second comparison. Choose a page you can revisit without signing in.

Switch to your browser.
Visit a web page that contains several paragraphs of useful text.
Confirm that the page loads without requesting an account.
Copy the complete URL from the browser address bar.
Save the copied URL here: your web page URL.

Before the final check, do you expect all four prerequisites to pass on the first try?

Switch back to your GitHub tab.
Confirm that your signed-in account can use Copilot Free.
Switch back to your terminal window.
Confirm that the terminal prompt is visible.
Return to the PDF viewer.
Confirm that the chosen PDF opens correctly.
Reload the selected web page in your browser.

That clears the runway. Every prerequisite is ready for the document analysis workflow.

Your account, terminal, PDF, plus web page are ready. Next, you will install Copilot CLI using the GitHub account you confirmed.

Install and Sign In to Copilot CLI

The document checks later depend on an authenticated command-line session. Right now, your terminal has no active Copilot session.

GitHub Copilot CLI brings Copilot into your terminal. Copilot Free includes this tool. This step installs it before linking your GitHub account.

In this step, get ready to:

Check whether Copilot CLI is already installed.
Install Copilot CLI with the official method for your operating system.
Authenticate the prompt with your GitHub account.

Check and install Copilot CLI

A version check tells you whether the Copilot CLI executable is already available to your terminal.

Press Cmd+Space on macOS. Press the Windows key on Windows.
Type Terminal on macOS. Type PowerShell on Windows.
Press Enter to open the terminal.

You now have a terminal ready for the installation check.

Check for an existing Copilot CLI installation by running this command:

copilot version

What does this check do?

The copilot version command asks the installed executable for its version information.
A version response confirms that your terminal can find Copilot CLI.

✔️ I see version information

Good news. Copilot CLI is already available in your terminal.

ⓧ The command is unavailable on macOS or Linux

The official install script downloads Copilot CLI for macOS or Linux.

Install Copilot CLI by running this command:

curl -fsSL https://gh.io/copilot-install | bash

What does this command do?

The curl command downloads GitHub's official installer.
The bash command runs the downloaded installation script.

Confirm the installation by running this command:

copilot version

What should I see?

You'll see Copilot CLI version information. This confirms that the new executable is available in your terminal.

Terminal cannot find Copilot CLI?

Restart your terminal session so its command path reloads.
Run the installation command again if the download was interrupted.
Help me troubleshoot this installation.

ⓧ The command is unavailable on Windows

The Windows package manager may ask you to approve the installation. This approval allows it to add Copilot CLI to your computer.

Install Copilot CLI by running this command:

winget install GitHub.Copilot

What does this command do?

The winget command retrieves the official Copilot CLI package.
The package manager installs the executable for your Windows account.

Approve the installation if your package manager asks for confirmation.
Confirm the installation by running this command:

copilot version

What should I see?

You'll see Copilot CLI version information. This confirms that the new executable is available in PowerShell.

Installation not completing on Windows?

Restart your terminal before repeating the version check.
Use an administrator-approved terminal if package installation is blocked on your computer.
Help me troubleshoot this installation.

Sign in with GitHub

OAuth keeps your credential entry inside GitHub's browser flow. Copilot CLI stores the resulting authorization in your operating system's credential store.

Your browser may open automatically after the command starts. This is the expected sign-in flow.

Start the GitHub sign-in flow by running this command:

copilot login

What does this command do?

The copilot login command starts an OAuth authorization flow.
A local terminal uses the browser flow by default.
Copilot CLI stores the authorization after GitHub approves the request.

Open the URL shown in your terminal if your browser does not open automatically.
Sign in to the GitHub account that has Copilot Free.
Review the requested access on GitHub.
Click Authorize GitHub Copilot CLI.
Return to your terminal after GitHub completes the authorization.

You'll see a success message in the terminal. That is the account link done. Copilot CLI can now use your Copilot Free access.

Sign-in not completing?

Confirm that the browser uses the GitHub account with Copilot Free access.
Repeat copilot login if the browser authorization expired.
Help me troubleshoot this sign-in.

Verify the authenticated prompt

The interactive prompt proves that the installed executable can use its stored GitHub authorization. The current-user check identifies the account behind that session.

Before you launch it, which GitHub account do you expect Copilot CLI to use?

Launch the interactive Copilot CLI prompt by running this command:

copilot

What does this command do?

The copilot command launches the interactive user interface in your terminal.

Verify the current account from the Copilot CLI prompt by entering this slash command:

/user

What does this check prove?

The /user command shows the current GitHub user. Seeing your account confirms that authentication is active.

Copilot CLI is installed locally. The prompt is authenticated with your GitHub account on the free Copilot plan.

Prompt showing the wrong account?

Exit the interactive prompt before starting the sign-in flow again.
Authorize the GitHub account that has Copilot Free access.
Help me correct the active account.

Your authenticated terminal is ready. Next, you'll bring the eyeball plugin onto your computer.

Set Up the Eyeball Plugin

Your authenticated Copilot CLI can answer questions about files. It still needs Eyeball before it can attach visible source evidence to each claim.

The Eyeball setup creates a Python environment for the plugin. It downloads Chromium so Eyeball can capture passages from documents or web pages.

In this step, get ready to:

Confirm that Python meets Eyeball's minimum version.
Clone the Eyeball repository to your Desktop.
Run the setup script to make the plugin ready for Copilot CLI.

Check your Python version

Eyeball requires Python 3.8 or later. The setup script uses this runtime to create the plugin environment.

Check your installed Python version using the tab for your operating system:

macOS

Print your Python 3 version from the terminal you used earlier by running this command:

python3 --version

What does this command do?

The --version option prints the installed Python release. The first two numbers show whether it meets Eyeball's requirement.

Windows

Print your Python 3 version from PowerShell by running this command:

py -3 --version

What does this command do?

The Python launcher selects an installed Python 3 release. The --version option prints that release's version.

Linux

Print your Python 3 version from the terminal by running this command:

python3 --version

What does this command do?

The --version option prints the installed Python release. The first two numbers show whether it meets Eyeball's requirement.

✔️ I see version 3.8 or higher

Your Python runtime is ready. Eyeball can use it to create the environment for its document tools.

ⓧ I see an older version

Your current Python release is below Eyeball's minimum requirement of 3.8.

On macOS, an older system Python may remain present after you install a current release. The official installer's defaults place the newer python3 command first.

Visit the official Python downloads page.
Select the installer or platform guide for your operating system.
Complete the installation using the documented defaults.
Repeat the version check above.

ⓧ Command not found

Your terminal cannot find a Python 3 installation. Install a current Python 3 release before continuing.

Visit the official Python downloads page.
Select the installer or platform guide for your operating system.
Complete the installation using the documented defaults.
Return to the terminal from earlier.
Repeat the version check above.

Prepare the local repository

A local repository contains Eyeball's setup scripts plus its plugin files. Git copies that repository into a folder you can use from your terminal.

Move to your Desktop by running this command:

cd ~/Desktop

What does this command do?

The command changes the terminal's current location to your Desktop. This keeps the cloned repository easy to find.

Clone the Eyeball repository by running this command:

git clone https://github.com/dvelton/eyeball.git

What does this command do?

The git clone command downloads the repository files from GitHub.
Git stores the files in a new eyeball folder on your Desktop.

Good progress. The eyeball folder now contains the setup scripts needed for the plugin.

Repository clone not completing?

Check that your internet connection is active. Remove any incomplete eyeball folder before retrying the clone.

If the clone still stops, help me troubleshoot the repository download.

Run the setup script and install the plugin

The setup script creates Eyeball's Python environment. Playwright downloads a Chromium browser build for evidence capture.

Expect this command to take longer than the earlier checks while Chromium downloads. The terminal may pause between progress updates during this first setup.

Start the repository setup using the tab for your operating system:

macOS

Start the setup from the cloned repository by running these commands:

cd eyeball
bash setup.sh

What do these commands do?

The first command moves the terminal into the cloned eyeball folder.
The setup script creates the Python environment for Eyeball.
The setup script installs the required Python packages.
The setup script downloads Chromium for source screenshots.

Windows

Start the setup from the cloned repository by running these commands:

cd eyeball
.\setup.ps1

What do these commands do?

The first command moves PowerShell into the cloned eyeball folder.
The PowerShell script creates the Python environment for Eyeball.
The setup script installs the required Python packages.
The setup script downloads Chromium for source screenshots.

Linux

Start the setup from the cloned repository by running these commands:

cd eyeball
bash setup.sh

What do these commands do?

The first command moves the terminal into the cloned eyeball folder.
The setup script creates the Python environment for Eyeball.
The setup script installs the required Python packages.
The setup script downloads Chromium for source screenshots.

You will see the Python dependency installation complete. You will also see the Chromium download finish before the terminal prompt returns.

That is the heavy setup done. Eyeball now has the runtime it needs to capture evidence.

Setup script not completing?

Confirm that the terminal is inside the Desktop's eyeball folder. Check that the Python version from your earlier result is 3.8 or later.

Retry the script with an active internet connection. If it still stops, help me diagnose the Eyeball setup.

The completed environment makes Eyeball runnable. Copilot CLI now needs to install the plugin so it can call those tools.

Before you send the installation prompt, do you expect Copilot CLI to recognize Eyeball as a plugin?

Switch back to the authenticated Copilot CLI prompt from earlier.
Ask Copilot CLI to install Eyeball by entering this prompt:

Install the plugin at github.com/dvelton/eyeball for me.

What does this prompt do?

Copilot CLI uses the repository as an installable plugin source. It registers Eyeball so future prompts can call its document-analysis workflow.

Copilot CLI confirms that the plugin is installed. Eyeball is now ready to call with a document or web page.

Plugin installation not confirmed?

Confirm that you entered the installation prompt inside Copilot CLI. A regular terminal shell treats the text differently.

If Copilot CLI cannot complete the installation, help me troubleshoot the Eyeball plugin setup.

Your evidence plugin is ready. Next, you will ask Copilot CLI about your PDF without Eyeball to create an unsupported baseline.

Capture an Uncited PDF Answer

Your Eyeball environment is ready for document analysis. The fastest way to understand what it adds is to create a baseline first.

A fair comparison uses the same PDF and question in both runs. This step captures an answer from Copilot CLI while Eyeball stays out of the request.

In this step, get ready to:

Choose a PDF that you own.
Record one focused question about its contents.
Save Copilot CLI's answer as your baseline.

Set a fair comparison

The document and question are your fixed inputs. Reusing them later keeps the comparison fair.

Choose a PDF you own that is safe to process through Copilot CLI.
Confirm the file name ends in .pdf.
Record the PDF's path here: path/to/your/document.pdf.
Record one factual question about the document here: your question about the PDF.

Run the unassisted analysis

Copilot CLI can attach your PDF directly to an interactive prompt. This request explicitly excludes Eyeball to isolate the default answer.

Before you send the prompt, do you expect the response to place a highlighted source screenshot beside every claim?

Switch back to the authenticated Copilot CLI prompt from earlier.
Attach your PDF and submit your recorded question by entering this prompt:

Do not use the eyeball plugin. Answer [[PDF_QUESTION="your question about the PDF"]] using @[[PDF_PATH="path/to/your/document.pdf"]].

What does this prompt do?

The @ prefix attaches the PDF at your recorded path.
The opening sentence excludes Eyeball from this run.
The final sentence reuses your recorded question.

Approve access to the PDF's folder if Copilot asks.
Wait for Copilot to finish the response.
Check that the response addresses your question about the PDF.

You should see a completed written answer in the terminal based on your selected PDF.

PDF not attaching?

Check that @ sits immediately before the PDF path.
Use the arrow keys to choose the matching path suggestion.
Press Tab to complete the selected path.
Approve access to the PDF's folder if Copilot requests permission.
Help me troubleshoot why Copilot CLI cannot attach my PDF..

Save and inspect the baseline

A useful baseline preserves the model's original wording. Saving the complete response gives you an exact result to compare later.

Select the entire completed Copilot response in your terminal.
Copy the selected response with Cmd+C on macOS or Ctrl+C on Windows.
Store the complete response here: paste the baseline answer here.

Before you inspect the saved answer, predict whether every claim has visible evidence from the original page beside it.

Review your saved baseline answer from beginning to end.
Look beside each claim for a highlighted screenshot from the source.
Check the response for an attached Word file containing those screenshots.

The answer gives you claims about the PDF. No highlighted source screenshot appears beside them.

You now have a clear comparison point. The saved answer preserves the exact evidence gap you observed.

What is grounding?

Grounding ties an AI claim to source evidence you can inspect. Your baseline shows what the answer looks like when that visible link is missing.

Eyeball appeared in the response?

Return to the Copilot CLI prompt.
Submit the baseline request again with its opening instruction unchanged.
Replace the saved baseline answer with the new response.
Help me keep Eyeball out of this baseline run..

Your uncited baseline is saved. Next, you will run the same question through Eyeball to compare the evidence it captures.

Generate Cited PDF Evidence

Your baseline answer showed how quickly a useful response becomes difficult to audit. Its claims have no attached view of the PDF text behind them.

In this step, you will route the same question through Eyeball inside Copilot CLI. The plugin uses grounding to place each claim beside a highlighted screenshot from your PDF.

In this step, get ready to:

Run the same PDF question through Eyeball.
Watch Eyeball highlight the cited PDF passages.
Confirm the Word evidence document reaches your Desktop.

Keep the comparison fair

A fair comparison keeps the source document unchanged. It also keeps the question unchanged.

Switch back to the authenticated Copilot CLI conversation from earlier.
Fill the Eyeball prompt with your recorded source details by entering:

use eyeball on [[PDF_PATH="path to your PDF"]] -- [[PDF_QUESTION="your PDF question"]]

What does this prompt do?

The use eyeball on phrase tells Copilot CLI to use the plugin for this request.
The first value points Eyeball to your selected PDF.
The -- divider separates the source from the analysis request.
The final value reuses the question that produced your baseline answer.

Press Enter to submit the prompt.
Confirm the terminal begins processing your selected PDF through Eyeball.

Eyeball not starting?

Confirm path to your PDF still points to the local PDF you selected.
Confirm the Copilot CLI conversation from earlier is still authenticated.
Review the eyeball repository setup notes if the plugin does not activate.
Help me troubleshoot this Eyeball run.

Watch the evidence capture

Eyeball renders the PDF as a source it can inspect. Each matched passage becomes a highlighted screenshot for the report.

Keep the Copilot CLI run active until evidence capture finishes.
Observe Eyeball open your selected PDF.
Watch Eyeball locate the passages used for its claims.
Confirm the cited text receives a yellow highlight.

That is the capture stage complete. Your run has produced evidence for the same question that created the unsupported baseline.

Evidence capture stopping?

Confirm the PDF opens normally in your usual document reader.
Keep the Copilot CLI session active until the evidence run finishes.
Help me diagnose the evidence capture.

Find the finished artifact

A generated report becomes reusable once you can identify its saved file. The file gives your next audit a stable artifact.

Before you check, do you expect the evidence to remain inside the Copilot CLI conversation or appear as a separate file?

Open the Desktop folder in Finder on macOS or File Explorer on Windows.
Find the newest .docx file created by Eyeball.
Record its file name here: your generated Word evidence file name.

You will see a new Word evidence document on your Desktop. Inside it, each generated claim sits beside a screenshot of its highlighted PDF passage.

Your PDF evidence report is ready. Next, you will open it to test whether each screenshot supports the claim beside it.

Review the PDF Evidence

Your eyeball run transformed the PDF question into an evidence document with a highlighted source screenshot beside every claim.

Those screenshots make grounding visible. A plausible claim can still include details that its source never states.

In this step, get ready to:

Open the evidence document from your Desktop.
Review every claim against its highlighted source screenshot.
Record at least one claim that its screenshot does not support.

Open the evidence document

The document is your audit surface. Its claim and screenshot pairs let you check each answer without searching through the original PDF again.

Locate the evidence document that eyeball wrote to your Desktop in the previous step.
Double-click the document to open it in Microsoft Word or LibreOffice.
Scroll to the first generated claim.

Good, your audit trail is open with the first claim positioned beside its highlighted source screenshot.

How is the evidence arranged?

Each written claim is the statement you need to test. The adjacent screenshot shows the passage eyeball selected as support.

The highlight narrows your review to the exact source wording. This makes unsupported details easier to spot.

Document not displaying correctly?

Confirm that you opened the evidence document from your Desktop. The original PDF does not contain the generated claim and screenshot pairs.

Try the other supported document application if the layout appears blank or incomplete. Help me troubleshoot the evidence document layout.

Audit each claim against its source

Evidence supports a claim when the highlighted wording justifies the complete statement. Even one unsupported detail creates an evidence mismatch.

What counts as an evidence mismatch?

A mismatch exists when the claim contains meaning that the highlighted screenshot does not support.

The claim adds a cause that the passage never gives.
The claim includes a number that the passage never states.
The claim sounds more certain than the passage.
The claim combines separate ideas into one unsupported conclusion.

Predict whether every highlighted passage will support the complete claim beside it.
Read the first generated claim.
Read the highlighted wording in its adjacent screenshot.
Compare the meaning of the claim with the highlighted wording.
Repeat the comparison for every claim in the document.
Select the clearest claim containing a detail that its screenshot does not support.
Record the unsupported claim plus the evidence gap here: your unsupported claim and evidence gap.

That is the core audit complete. You have identified a generated claim that goes beyond its highlighted source.

Verify the evidence mismatch

A useful mismatch record names the claim plus the unsupported detail. A final comparison confirms that another reader can see the same gap.

Predict whether the highlighted wording will support every part of your recorded claim.
Return to the pair captured as your unsupported claim and evidence gap.
Read the complete claim once more.
Read the highlighted passage once more.
Identify the specific detail that appears only in the claim.

You should see at least one claim detail that is absent from the highlighted passage. That visible gap confirms your evidence mismatch.

Your PDF report now shows both the generated claims and the evidence gap you found. Next, you will test the workflow against the live web page you saved earlier.

Compare Evidence from a Web Page

Your PDF review gave you a concrete test for eyeball: every claim has to survive comparison with its own screenshot. You already found EVIDENCE_MISMATCH in the first report.

A live page gives the workflow a second source type. In this step, you'll run that page through Copilot CLI before comparing its evidence with the fixed PDF screenshots.

In this step, get ready to:

Define a focused test for a live web page.
Generate a Word evidence document from the page with eyeball.
Compare web-page screenshots with PDF screenshots.

Define the web test

Web pages can change after publication. A focused question makes the captured evidence easier to compare with your PDF report.

Select the live web page you kept for this test.
Confirm that its main text is visible without signing in.
Record its URL here: paste the live web page URL.
Write one focused question that the page can answer.
Record your question here: your question about the web page.

Generate web evidence

Eyeball uses Chromium to render the web page. It captures the relevant page regions before assembling them into a Word evidence document.

Return to the authenticated Copilot CLI conversation from earlier.

Before you send the prompt, do you expect the live-page screenshots to preserve as much context as the PDF screenshots?

Send your saved URL and question to eyeball by entering this prompt:

run eyeball on [[WEB_URL="paste the live web page URL"]] -- [[WEB_QUESTION="your question about the web page"]]

What does this prompt do?

The run eyeball on instruction asks Copilot CLI to activate the plugin for your source.
The WEB_URL value supplies the live page to render.
The WEB_QUESTION value focuses the analysis on the information you want to verify.

Wait for Copilot CLI to finish the eyeball run.
Return to your Desktop in Finder on macOS or File Explorer on Windows.
Identify the newest Word document created by eyeball.
Record its exact filename here: web evidence document filename.
Open web evidence document filename using the document app from earlier.

You should see generated claims beside screenshots captured from the live web page. That's the second evidence source captured.

No web evidence document?

Check that your saved URL opens without a sign-in in a regular browser window. Pages that load their main text later can produce incomplete captures.

Try a page whose main text is visible as soon as it loads. Help me troubleshoot why eyeball did not create a Word evidence document from my web page.

Compare both evidence documents

Both reports pair claims with visual evidence. The source format changes how much layout or surrounding text each screenshot preserves.

Before you compare them, which source do you expect to show clearer surrounding context?

Switch back to PDF_EVIDENCE_DOCX from earlier.
Use your window controls to place web evidence document filename beside PDF_EVIDENCE_DOCX.
Select one claim in the web evidence document.
Check whether its highlighted screenshot supports the claim.
Compare the surrounding context in that web screenshot with one PDF screenshot.
Compare the readability of the two highlighted passages.

You should now be able to point to a visible difference between the live-page capture and the fixed PDF capture. The same evidence check now works across both source types.

Strong finish. You now have two evidence documents that show how eyeball captures receipts from a PDF and a live web page.

Secret mission

Audit the Evidence Behind Each Claim

Audit a sample of claims from your PDF and live-page evidence reports. You will score how closely each highlighted screenshot supports the claim beside it.

Clean Up Your Resources

This workflow runs entirely on your machine, so leaving its files in place creates no ongoing service cost. Choose whether to keep the workflow ready, pause your work, or remove the local resources.

Resources you used:

The PDF_EVIDENCE_DOCX evidence document on your Desktop.
The WEB_EVIDENCE_DOCX evidence document on your Desktop.
The BASELINE_ANSWER saved without source screenshots.
The local eyeball repository with its Python environment.
The Chromium download used to capture the evidence screenshots.

Keep everything running

No action is needed while you want the workflow ready for more document checks. The local setup uses disk space without creating an ongoing service cost.

Keep PDF_EVIDENCE_DOCX on your Desktop as the audit artifact from your PDF comparison.
Keep WEB_EVIDENCE_DOCX on your Desktop for the live-page comparison.
Retain the local eyeball repository for future evidence checks.
Leave the Python environment with the Chromium download in place for future runs.
Leave GitHub Copilot CLI installed if you plan to repeat the workflow.

Pause - I'll come back to this later

There is no cloud service or local server to pause. Closing your active files ends the session while preserving the complete workflow.

Close PDF_EVIDENCE_DOCX after saving your review notes.
Close WEB_EVIDENCE_DOCX after saving your comparison.
Close the terminal session that you used for Copilot CLI.
Leave the local eyeball repository in its current location.

That gives you a clean stopping point. Your reports and local setup remain ready for your return.

Delete - I don't want to use this again

Deleting these local files is permanent after your operating system clears them. Your original PDF remains untouched because this project did not create it.

macOS

Remove the local setup:

Press Cmd+Space to open your search bar.
Type Finder and press Enter to open it.
Return to the eyeball Python environment from earlier.
Delete the eyeball Python environment.
Return to the eyeball Chromium download from earlier.
Delete the eyeball Chromium download.
Return to the local eyeball repository from earlier.
Delete the local eyeball repository.

The repository, Python environment, and Chromium download are now removed from their original locations.

Remove the generated reports:

Return to your Desktop in Finder.
Delete PDF_EVIDENCE_DOCX.
Delete WEB_EVIDENCE_DOCX.
Return to the location where you saved BASELINE_ANSWER.
Delete BASELINE_ANSWER.
Clear your operating system's deleted items to reclaim the disk space.

You should no longer see the eyeball setup or its generated reports in their previous locations.

Windows

Remove the local setup:

Press the Windows key to open your search bar.
Type File Explorer and press Enter to open it.
Return to the eyeball Python environment from earlier.
Delete the eyeball Python environment.
Return to the eyeball Chromium download from earlier.
Delete the eyeball Chromium download.
Return to the local eyeball repository from earlier.
Delete the local eyeball repository.

The repository, Python environment, and Chromium download are now removed from their original locations.

Remove the generated reports:

Return to your Desktop in File Explorer.
Delete PDF_EVIDENCE_DOCX.
Delete WEB_EVIDENCE_DOCX.
Return to the location where you saved BASELINE_ANSWER.
Delete BASELINE_ANSWER.
Clear your operating system's deleted items to reclaim the disk space.

You should no longer see the eyeball setup or its generated reports in their previous locations.

Nice Work!

You did it! You built an auditable document analysis workflow with Copilot CLI and eyeball. Your evidence checks even exposed a claim whose highlighted screenshot did not support it.

You've learned how to:

Capture a baseline answer with no attached source screenshots.
Generate Microsoft Word evidence reports that place each claim beside a highlighted source screenshot.
Audit PDF evidence to identify a claim that its screenshot does not support. Compare those results with evidence captured from a live web page.
Secret Mission: Complete the optional challenge to push your evidence-checking workflow further.

Ready to quiz yourself?