WEEK 07 · THE EXTRACTION AGENT, MEASURED

Do one thing.
Then you are ahead.

Now the junior analyst meets the answer key. You build extract/agent.py with two brains, a frontier model through the API and a local Qwen on your 3050, plus a fake brain for tests. Then you do the thing that distinguishes an AI scientist from an AI enthusiast: you measure it per field, not as one flattering average. It will be very good at "organism" and bad at "working volume", and that unevenness is the finding. …

If you only do one thing this week: the F1 table, whatever it says

▶

THE NEXT BLOCK

How long have you got?

10 min 25 min 50 min 90 min 12 of 13 cards still open in week 07 · 11 finished overall

1/12 Read 30 min? W07 Monday Craft

Zheng et al. (2023), "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena"

Monday build block · Read it inside Monday's build block

Build-block reading · Zheng et al. (2023), "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena", NeurIPS Datasets & Benchmarks, sections 1-3 (judge biases you must avoid in week 14)

Find it ↗ search

In: LABLOOP-W07 Notebook

↻ Something else See everything in week 07 →

By Sunday you should be able to say: “I can read my own F1 table aloud, say why the worst field fails, and explain why tuning on the gold set is dangerous.”

🔥

MOMENTUM

What you have actually done

Focus today0 min0 blocks on the timer

Last 7 days0 min0.0 hours measured, not estimated

Streak0days in a row

Cards finished11across the programme

This week1/13

Time on the timer, day by day last 14 days

2026-09-29 18:29 Finished: SciPy docs, scipy.integrate.solve_ivp
2026-09-29 17:44 Finished: LearnChemE (CU Boulder) · "Monod Cell Growth Example (Bio)"
2026-09-28 22:11 Finished: Crash Course Biology #3 · "Biological Molecules
2026-09-28 21:50 Finished: OpenStax Biology 2e
2026-09-25 11:16 Finished: OpenStax Biology 2e