Read the whole story

This tutorial follows the repository recipe Read the whole story.

What you will learn

• How the stages compose: each one consumes the previous run’s id
• What kno report composes and what it deliberately caveats

The chain

# 1. The reference point
kno baseline --evals cases.jsonl

# 2. Per-asset measurement
kno value --evals cases.jsonl --pool pool.jsonl --baseline-run-id <baseline run id>

# 3. The keepers, under budget
kno select --value-run-id <value run id> --pool pool.jsonl

# 4. The artifacts
kno export --select-run-id <select run id> --pool pool.jsonl --destination knowledge_base

# 5. The whole story
kno report

What the report says, and does not say

The report composes the recorded stages into one page: verdicts per asset, the selected portfolio, the gaps, and the caveat that nothing is validated on holdout yet. Until validate ships, the report is honest about exactly that: selection-time gains are inflated by the winner’s curse, and the holdout number is the one that belongs in a slide.

Full recipe: the cookbook entry.