Curate the model list: rank by privacy tier, price per completed task, retire duplicates

Problem: The model list is a pile, not a curation. 3 Mercury versions, 2 Gemma 4 variants, Aion 3.5 and Aion 3.5 Mini — near-duplicates serving the same function. No ranking, no rotation, no visible signal from user feedback. In the Private tier this costs real VRAM; a fixed budget gets spent serving the same role twice.

Ask:

1. Publish a ranking, split by privacy tier (Anonymous / Private / TEE / E2EE). Separate leaderboards, not one blended list. Bonus: it finally makes the tier difference legible to casual users who never read the privacy page.

2. Replace "cost per 1k tokens" with "cost per completed task" in the model card. A cheap model that fails three times is more expensive than a costly one that succeeds first try. Right now the UI hides that.

3. Rotation. Auto-retire models below a quality floor. Stop stacking near-duplicates.

4. Anonymous in-app feedback box. The official feedback channel is a third party and requires an identity. For a privacy company, that is telling.

Low-friction version: one-tap model switch mid-chat, plus thumbs up/down after each reply. The ranking builds itself from clicks, and an auto-router trained on them means the casual user never has to pick from 40 models. Anyone forced to choose between 40 models leaves for GPT.