Brain
CLIENT
Brain
YEAR
2026
INDUSTRY
Knowledge tooling
SERVICES
AI Systems
Key Highlights
One model's reading of a video is one opinion
Tutorials get details wrong, skip steps, or describe an older version of a tool. Trusting one at face value is risky, and a single AI reading inherits whatever that one model happened to notice or miss.
Disagreement is the signal
Two models watch independently and their readings are reconciled into one labelled spec: confirmed, single-source, conflict, or open question. A conflict-heavy tutorial ran to 281 lines; one where both readings agreed ran to 30.
Watch it once, get a fact-checked spec out
A pipeline that takes a tutorial, has two different models watch it independently, reconciles what they saw into one labelled specification, and can hand that spec on to be turned into a runnable skill.
In numbers
The project, counted.
281
Lines Of Conflict
On the tutorial the two readings most disagreed about.
30
Lines
On one where they agreed.
2
Independent Readings
Reconciled into one spec.
