Figure Friday 2025-w22

Marvel films, compare worldwide gross revenue against budget, an AI Claude-experiment.

The Figure Friday dataset primarily contains financial data for all Marvel films. After reviewing the data and converting several columns from string values (e.g., “38%”) to numerical values, I decided to experiment with a few combinations:

• Vibe Coding with Claude via OpenRouter.ai
• A Dash Mantine app instead of a Dash Bootstrap app
• Applying a UI/UX/Mantine style guide created by a member of the Plotly community (see video for explanation and download link)

The Prompt

The app needed to be simple: dark/light mode, a category filter (since Marvel films are often part of a series), and a scatter plot clearly showing how global earnings relate to the film’s budget.

After a somewhat failed experiment—due to version mismatches between various Python libraries on my system—I decided to use this prompt, followed by the pasted style guidelines:

#the data loaded is information about marvel movies. 

#assignment:
#create a dash app, use only dash mantine components.
#create the app based on the uploaded best practices for UI/UX/Mantine.
#add a light/dark mode switch with callback, make dark mode default
#create a filter dropdown for category, add "all categories" and make "all categories" default
#on appload, load the plots for "all categories"

#create a scatterplot
# x = year, y="% budget recovered"
# colors for traces: red if y < 200, blue if y > 250 else purple
# create a custom hover with film, category, worldwidegross and budget
# create a callback which makes it possible to filter on category

Result

The result didn’t work right away—there was an error in the component structure, and the hover data referenced a few columns that didn’t exist.

What bothered me more, though, was a horizontal line at 100% in the visualization labeled “Break Even.” That’s not necessarily accurate; I actually wanted the data points to change color based on the percentage. In the film industry, a movie typically doesn’t break even when ticket sales match the budget. It usually happens only when earnings reach 2.5 to 3 times the budget. Some argue for 2 to 3 times, but either way, it’s incorrect to assume that ticket sales equal to the budget means the film has broken even.

In the final visual, this is reflected by:

• Orange dots if earnings are below 200%,
• Gray for the in-between range,
• Green if ticket sales are 3 times or more than the budget.

Although the app includes a dark/light mode switch, text colors in the visualization don’t adjust when switching to Dark Mode, making some text unreadable. Additionally, the legend disappears when selecting a category. These are fixable details, either by AI or by myself (I didn’t fix them and didn’t ask AI to help). Still, they annoyed me.

Model and Costs

Generating what I asked for in the prompt plus one small tweak initiated by Claude, ended up costing $1.82. We’re talking about a fairly simple task here. Compared to models generally considered “less advanced,” that’s a relatively high amount. So naturally, expectations are higher as well. Next time, I’ll try using a “lighter” model first.

The $1.82 was what remained of $10 in credits I purchased two months ago. I’m not a heavy-paying AI user, but I’ve done a few things via OpenRouter that I considered more complex, involving much more interaction. And each time I thought, “Well, that basically costs nothing.” 🙂 So, I’ve learned something from this experience.

The Prompt, One More Time

On Figure Friday, people post a link to their solution along with some explanation of how and why they built their app or chart. In response to my submission, someone commented: “It also depends on your prompt.” 🙂
This person had sent the data to Claude in free format, accompanied by a brief description. The result? A technically impressive app with KPI cards, lots of them, along with many filters and a variety of other chart types.

Regardless of the result, it was a completely different approach. My angle was: “I have this data and want to communicate it in this specific way.” His angle was more like: “Go ahead, make something cool with it.”
Function-driven versus more technically exploratory. One method isn’t necessarily better or worse than the other and the same goes for the outcomes.

My preliminary conclusion? At this point, I could just as easily stick with free or more affordable AI tools, as they would likely deliver similar results to Claude via OpenRouter for what I’m doing.

Figure Friday is an initiative by the Dash/Plotly community, where each Friday a new dataset is shared and participants create a visualization or small app to extract insights from it. The following Friday at 18:00, there’s a Zoom session where some participants explain the thought process behind their work. In the community thread, people also share their code — and when possible, a live demo — to learn from each other.