Treni

Scratch Experiments

Exploratory, non-canonical, or debugging-heavy work that should not be confused with the main proof lanes.

Purpose

This page exists to keep the main docs clean.

Use this bucket for:

  • random UI probes
  • provider launch attempts
  • one-off model experiments
  • failed or inconclusive reruns
  • debugging traces
  • temporary comparison branches

Rules

  • If an experiment is claim-safe and repeatable, move it into the canonical pages.
  • If it is noisy, incomplete, or only useful for debugging, keep it here.
  • Do not cite scratch work as proof unless it is promoted later.

Current Scratch Topics

  • Lambda 9B launch attempts while provider capacity is unstable.
  • UI/console iteration that is useful for product polish but not part of the benchmark proof.
  • exploratory model promotions (0.8B vs 4B vs blocked 9B) before a stable larger-GPU host exists.

Where To Look For Canonical Results Instead

On this page