TRIBE v2 Long-Form Brain-Response Breakdown

Cortical-response model output for How I Turned Pi Into the Ultimate Coding Agent, a 21:26 YouTube video processed on the RTX 5090 rig.

June 2, 2026· Prepared by Codex· Research sandbox only
How to read this

TRIBE v2 predicts cortical-response maps from video, audio, and language features. The timeline is useful for comparing where the modeled response rises, falls, or shifts. It is not a virality score, preference score, clinical interpretation, or proof that a specific edit caused retention behavior.

i
YouTube retention context

The supplied YouTube Studio screenshot shows an average view duration of 6:24, an average percentage viewed of 29.9%, and 62% of viewers still watching around 0:30, which Studio labels below typical. The screenshot is contextual evidence, not a machine-readable retention series, so this report does not compute a numerical correlation coefficient yet.

Video duration
21:26
1,287 retained one-second segments
Prediction array
1,287 × 20,484
seconds × fsaverage5 vertices
Words transcribed
5,181
341 sentence events
Total runtime
103.0 min
480p analysis input

Timeline

The first panel shows overall cortical-response magnitude and its 95th-percentile envelope. The second shows second-to-second map change. The third compares left and right hemisphere magnitude. Vertical guides mark the strongest response-magnitude moments.

TRIBE v2 long-form timeline breakdown chart

Sampled Cortical Maps

Twelve evenly spaced cortical maps summarize how the predicted surface response develops across the video.

Sampled TRIBE v2 cortical response maps

Strongest Response Moments

These are the largest per-second mean absolute cortical-response values. Nearby transcript words are editing landmarks, not causal explanations.

Peak TRIBE v2 cortical response maps
TimeMean absolute responseNearby transcript words
0:310.2045Pi coding agent. I know you've probably heard of a lot of different coding agents and seen plenty of videos
0:370.1921me and hear me out on this one. This one is different. The way it works is not like any of the other
0:060.1523impressions video where I talked about the SDK a ton, some brief thoughts on the
10:340.1514It's fully cleared. It all just kind of works. I can do slash resume whenever I want to, to go back to an existing
6:180.1505our CI either in GitHub Actions or in some solution that tries to fix GitHub Actions. But the
11:130.1435is fully open source. I will link it down below. If you want to go clone it onto your machine, you can have the exact
10:160.1427the last message and as context, so I can paste into a new session to clear out the context
6:220.1413Actions. But the problem is GitHub Actions is fundamentally broken. And I think at this point it is much better to

Largest Map Transitions

These seconds have the largest change in the predicted cortical surface map versus the prior second. They are useful places to inspect cuts, topic changes, visual reveals, and pacing shifts.

TimeChange RMSNearby transcript words
5:000.1357That's it. I now have a fully custom extension added to my PI instance. That's literally all
8:200.1099for my terminal. This is a project that was released a couple months ago, I'm pretty sure. I was a
6:400.1078which means that it is fully compatible with GitHub Actions syntax. The only thing you need to do to migrate
0:280.0938with GPT 5.5 on low reasoning and the Pi coding agent. I know
0:010.0926A couple of weeks ago, I put out a video about Pi. This was kind of a first
21:260.0895And until next time, enjoy your new pie.
11:400.0838to hyper customize your pie into exactly what you want it to be. Mine should
10:290.0829custom extension that I made. We'll talk about that in a second. And then I can just do slash new, make a new session.

Aggregate Diagnostics

MetricValueMeaning
Mean absolute response, average0.0874Average magnitude across time and cortical vertices.
Mean absolute response, maximum0.2045Largest one-second response-magnitude moment.
Change RMS, average0.0301Typical second-to-second shift in the cortical surface map.
Change RMS, maximum0.1357Largest second-to-second map transition.
Left hemisphere magnitude, average0.0842Average model-output magnitude on left-hemisphere vertices.
Right hemisphere magnitude, average0.0906Average model-output magnitude on right-hemisphere vertices.

Run Details

Source
Public YouTube watch page
Input SHA-256
fec1c4506028d2be99a62026c6a7c56069c0ef2ae709065fc919e0a3a5609a6c
GPU
NVIDIA GeForce RTX 5090
CUDA
PyTorch 2.7.0+cu128 · compiled CUDA 12.8 · capability [12, 0]
Numerical sanity
Finite predictions float32 from -1.0531 to 0.9615
Events
{"Audio": 21, "Sentence": 341, "Text": 1, "Video": 21, "Word": 5181}
Boundary handling
Report-facing outputs are constrained to the original 1286.27-second MP4 duration. Preserved raw artifacts retain the untrimmed model output for audit.
!
Research-use boundary

The released TRIBE v2 source and weights are licensed under CC-BY-NC-4.0. Keep this analysis in the research sandbox unless commercial permission or an acceptable legal path is confirmed.