I grew up with Kaskade. Before YouTube existed, before we could find everything on the internet – this magazine was one of the first regular sources of information I got my hands on about juggling, circus, and shows.
When the magazine ceased publication in 2013, it felt like the end of an era – at least for me.
For several years, the magazines were still available for download as PDFs. Around 2017, the website went offline.
I often remembered small workshops, tutorials, or articles that inspired me back then. Now, years later, when I wanted to re-read something – there was no way to access it anymore.

Fortunately, using the Wayback Machine (The Internet Archive), I found a checkpoint from 2017 with the complete downloads (this isn't always the case, especially since it was about 3 GB of PDFs) – complete with German, English, and French editions.

While reviewing them, I realized that although many great articles and tutorials were hidden within the magazines, in today's world, hardly anyone would sift through 112 magazines that are merely photocopies. Well, unless you have a nostalgic interest =P

However, since the knowledge is too valuable to lose, I wanted to try using current technology to digitize everything in a way that makes it useful even today.

First, I removed all pages from the PDFs that did not contain relevant text.

For the actual transcription (or OCR), I used a multimodal language model from Google.
I used Gemini 2.0 Pro Experimental 02-05 with the prompt:

The attached PDF is a photocopy of a magazine. Extract all text, keep the document structure intact as much as possible, also extract single images and have them correctly in context.

The output is then cleaned up with gemini-2.0-flash-exp and the following prompt (along with the attached text we extracted):

The following text is extracted with OCR from an old magazine. Your task is to clean this up. Remove artifacts (like page-numbering, unnecessary linebreaks) or unneeded parts but keep the structure, articles etc. intact. Use a proper Markdown formatting to structure the text correctly.

Text:

The result is then reviewed and corrected manually one more time.

[!Danger]+ Important:
The focus was on extracting articles, workshops, interviews, etc.
Contributions like classified ads, etc., were discarded.
The extraction and cleanup of the texts were performed using LLMs, so there's always a possibility that texts were not transcribed 1:1 or that the content slightly deviates from the original. I tried to keep the error rate as low as possible by comparing text passages on a random basis.

Cascade 002:
In the original German PDF, issues 2+3 are combined.

Cascade 003:
Issue 004 can be found here in the original German PDF.

Cascade 004:
The title page is missing here; I haven't been able to figure out what it belongs to yet...
It looks like a copy of issue 009, without a title page.

Fix:
I have split the PDF of Cascade 002 (original) into 002 and 003.
Renamed Cascade 003 (original) to Cascade 004.
And deleted Cascade 004 (original).