# Meeting notes w/Joachim 19/05

Plan:
Re-design the original paper slightly, keeping in mind what we learnt with the claude experiments

**Things to wary:**
- Mass in the neutrals/passives segment (by defining the bin sizes - moving away from standard NPS definitions)
- Amplitude to increase the queue load over time
- NPS prediction model; relative strength of the topic features (keep topic features in study 2 => improve predictive power in study 3)
- Alternative to NPS-bins: instead of nps-predicted, without error-term, add an error distribution (and change the variance of this)

Practical questions:
- No overleaf subscription? How to collab on papers?