[clinic · EX-22] Fix evidence retention

[akash-coded]

Exercise clinic. One long-running thread per exercise. Ask here rather than opening a
new thread — the answers accumulate where the next person will look for them.

Exercise: EX-22 — Fix evidence retention · difficulty ⚫

What this exercise asks

Read the exercise first; this thread is for when you are stuck, not instead of it.

Three ways people get stuck on this one

1. Raising k and calling it fixed. It works, at 93% more tokens and a collapse in context_precision — see #28.

2. Applying a coverage constraint globally. It helps comparison questions and hurts single-hop ones; it has to be conditional and that has to be shown.

3. Measuring only full_chain_recall. The constraint can improve it while damaging precision, and both belong in the report.

How to ask here

Post the numbers you got, the config that produced them, and what you already tried.
A question with a measurement in it gets answered in one round trip; a question without one
takes three.

If you have finished it, post the write-up in Show and Tell with the
[solution · EX-22] prefix rather than here — this thread stays for people who are stuck.