Buffer week 13 · Mon 8 Feb – Sun 14 Feb 2027
Buffer 2 of 3 · catch up, consolidate, rest
16:15–17:00 after you wake properly. One untimed problem and revision – never new material.
Mon · 8 Feb
16:15–17:00
How to do it
0/3 steps2.Re-solve it untimed and AI-free
From a blank file, without looking at your old solution.
3.10 min of reviews/relearn.md
Tue · 9 Feb
16:15–17:00
How to do it
0/2 steps2.Re-solve it untimed and AI-free
From a blank file, without looking at your old solution.
Wed · 10 Feb
16:15–17:00
How to do it
0/3 steps2.Re-solve it untimed and AI-free
From a blank file, without looking at your old solution.
3.Or rest
Rest is a valid use of a buffer week.
Learn on Thursday, build Friday to Sunday. Friday flex is catch-up first, then stretch work.
Thu · 11 Feb
14:30–16:30
How to do it
0/2 steps1.Do
Finish the oldest unticked ‘done when’ check from Weeks 8–12 (P1, P2, P3).
2.Check you’re done
Done when: One more check ticked.
Short on sleep? Do the minimum version: One item. Let the next buffer week absorb the rest.
Fri · 12 Feb
09:00–10:30
How to do it
0/2 steps1.Do
Next unfinished item. If nothing is left, rerun P2’s eval and fix the worst failure.
2.Check you’re done
Done when: Eval rerun committed.
Short on sleep? Do the minimum version: 30 min. Let the next buffer week absorb the rest.
Fri flex · 12 Feb
10:45–12:15
How to do it
0/2 steps1.This block
If you’re caught up: polish P2’s README (diagram, eval table, limits, cost per answer).
2.Level up: Contextual retrieval
Before embedding and BM25-indexing each chunk, prepend a sentence saying where it sits in the document. Anthropic reports large drops in failed retrievals from this on its own test sets – your numbers will differ, which is the point of measuring. Add a one-line context prefix (page title + section heading is enough to start) to every P2 chunk, re-index, and add a fourth row to your configuration table. Done when: hit@5 for ‘hybrid + context prefix’ sits next to the other three configurations.
Sat · 13 Feb
09:30–12:30
How to do it
0/5 steps1.Start with 20 min revision
Explain retrieval failures vs generation failures in RAG.
2.Work through: Section 6.1 – P5 brief
3.Do
Message 5 people or organisations with a document-heavy question problem: a local charity, a community group, a small business, or a team at your work. Ask for 30 minutes and 20 real questions they get asked. Write projects/p5-capstone/brief.md: user, problem, 20 questions, success measure, data rules. Use public or shareable documents only, and no personal data.
4.Check you’re done
Done when: Brief committed, with at least one real reply or the fictional fallback chosen.
5.Commit
P5 brief.md
Short on sleep? Do the minimum version: Brief using the fictional fallback. Let the next buffer week absorb the rest.
Sun · 14 Feb
10:00–13:00
How to do it
0/3 steps1.Do
Write reviews/phase3.md: what you can explain about prompts, RAG, evals, agents and MCP. Update relearn.md.
2.Check you’re done
Done when: Review committed.
3.Commit
reviews/
$ git add reviews/$ git commit -m "Week 13: Phase review"$ git pushShort on sleep? Do the minimum version: Review. Let the next buffer week absorb the rest.
The last 30 minutes of Sunday. Saved with your progress.