Part I — Data, Memory, and Toquinho

There’s a moment — somewhere between experiment and ritual — where data begins to feel like memory.

On May 19th, 2025, I uploaded the first raw EEG dataset for Creativity in Vitro. It’s a fragile thing, both poetic and procedural: 20 repetitions of the same three-loop sequence of six lyrical phrases from Aquarela by Toquinho. Listened to with eyes closed, body still. Recorded continuously, without restarting the interface. Like a mantra looping in the cortex.

Each repetition is the same — and yet, not. Same phrases, same melody. But each loop is a different brain, a slightly different Lina. After each three-loop session, I blinked three times, deliberately. The blink acted as a biological marker, a ripple left in the EEG trace — not to be removed, but to be used.

These sessions were recorded with OpenBCI Cyton + Daisy, using 14 gel-based electrodes arranged according to the 10–20 system, skipping N9P and N10P for better spatial integrity. The headset doesn’t visualize Fz or Pz, but those points are present in the data — subtly pulsing in microvolts.

Here’s what I discovered:

• The OpenBCI GUI doesn’t create a new file each time you stop and start a session. It appends everything into a single .csv file.

• At first, this created confusion — signals seemed to “flatline” and restart. But those gaps were meaningful: each one marked the end of a session and the beginning of another.

• I wrote code to detect those gaps in the timestamp column and used them to split the full file into 20 distinct blocks.

• Each block was then trimmed: 7 seconds from the start, 3 seconds from the end, to remove noise and stabilization.

These 20 blocks — now cleaned and organized — were saved as .pkl files.

In a second notebook, I reopened them to identify the blink markers between loops. I didn’t insert any digital triggers — just pulses of the eyelid, visible in Fp1 and Fp2. I scanned the signal for two blinks per block, and used them to divide each session into three loops. That is supposed to give me 60 recordings in total — 3 per block, each tied to a listening of six phrases. Once I can confidently split each session into the 3 loops, the next challenge is to segment each loop into the six lyrical phrases — and that’s where I would really appreciate any insight now.

Next, I’ll divide each loop into six temporal slices, one per phrase — probably by second. It’s not perfect. It’s not clinical. But it’s radically personal, and methodologically transparent.

The data is raw, human, messy — not a simulation, not synthetic, not idealized. It’s my brain, listening to Aquarela, over and over again.

And now, it’s available

Part II — How to Version Raw EEG Data on GitHub with Git LFS

Uploading large EEG recordings to GitHub requires a few extra steps, since the platform has a 100MB file limit. To manage this, we used Git Large File Storage (LFS) to track and store the 122MB .csv file. Here’s a quick protocol for others working with physiological data:

# 1. Install Git LFS (only once per machine)
brew install git-lfs         # or: sudo apt install git-lfs / choco install git-lfs

# 2. Initialize Git LFS in your repo
git lfs install
git lfs track "*.csv"
git add .gitattributes

# 3. Add and commit your large file
git add your_data.csv
git commit -m "Track EEG dataset with Git LFS"

# 4. Push to GitHub (can take a few seconds for large files)
git push origin main

If you’ve already committed a file over 100MB, you’ll need to rewrite history using the BFG Repo-Cleaner:

# Optional cleanup for previous commits that exceed 100MB
git gc --aggressive --prune=now
bfg --strip-blobs-bigger-than 100M
git reflog expire --expire=now --all
git gc --prune=now --aggressive
git push --force