Exercises¶
One file per exercise. Every exercise that touches nanorag/ must ship with a
measurement and an interval โ a point estimate is not a result.
Stuck on one? Each exercise has a long-running clinic thread in Q&A โ ask there rather than opening a new thread, so the answers accumulate where the next person will look for them.
| Exercise | Clinic | ||
|---|---|---|---|
| ๐ข | EX-01 |
Attribute ten failures | ask |
| ๐ก | EX-02 |
Find the N where the ceiling stops moving | ask |
| ๐ด | EX-03 |
Break the funnel on purpose | ask |
| ๐ก | EX-04 |
Manufacture an eval set for a new domain | ask |
| ๐ด | EX-05 |
Make starvation reproduce | ask |
| ๐ก | EX-06 |
Null questions for a real corpus | ask |
| ๐ก | EX-07 |
Chunking bake-off on your own corpus | ask |
| ๐ด | EX-08 |
Implement a new chunking strategy | ask |
| ๐ด | EX-09 |
Survive an encoder upgrade | ask |
| ๐ด | EX-10 |
Prove permission isolation | ask |
| ๐ข | EX-11 |
BM25 by hand | ask |
| ๐ก | EX-12 |
The analyzer audit | ask |
| ๐ด | EX-13 |
Route ฮฑ by query class | ask |
| ๐ด | EX-14 |
Beat the reranker | ask |
| โซ | EX-15 |
Make the dense leg earn its keep | ask |
| ๐ก | EX-16 |
Price your own k | ask |
| ๐ด | EX-17 |
Measure position sensitivity for real | ask |
| โซ | EX-18 |
Solve abstention | ask |
| ๐ก | EX-19 |
Calibrate a judge against humans | ask |
| ๐ด | EX-20 |
Build a third bias probe | ask |
| ๐ก | EX-21 |
Cost model for a real client shape | ask |
| โซ | EX-22 |
Fix evidence retention | ask |
Difficulty: ๐ข first exercise ยท ๐ก needs a measurement ยท ๐ด a cohort project ยท โซ open research
How to submit¶
- Post it in Show and Tell
titled
[solution ยท EX-NN] ..., with your numbers, your interval, and what you tried that did not work. - Reviewers reply with a measurement, not an opinion.
- If it changes code, open a PR that links the thread.