Month 9-Week 3: Track final push (part 2) + writeup draft¶

Week summary¶

Goal: Finish track project to v0.5 (presentable v1-RC). Begin the substantial Q3-closing post.
Time: ~9 h over 3 sessions.
Output: Track project polished; long-form post drafted.

Why this week matters¶

Polish is what separates "another GitHub repo" from "a presentable artifact." The Q3-closing post is also where the year's specialty fully crystallizes.

Prerequisites¶

M09-W01 + W02 complete.

Recommended cadence¶

Session A-Tue/Wed evening (~3 h): feature freeze + tests + docs
Session B-Sat morning (~4 h): writeup outline + draft
Session C-Sun afternoon (~2 h): incumbent re-read + writeup polish

Session A-Feature freeze, tests, docs¶

Goal: Stop adding features. Add tests. Write docs.

Part 1-Feature freeze (15 min)¶

Make a decision: no more new features for the rest of the month. Move all open ideas to BACKLOG.md.

Part 2-Tests (90 min)¶

For each major surface in your project, add at least one test. CI should run them on push.

For evals projects: snapshot tests on a tiny golden set. For agent projects: a unit test that the tool-use loop terminates and returns expected shape. For inference projects: a benchmark test that runs in <60s and verifies throughput within a band.

Part 3-Docs (75 min)¶

Add or polish: - README quickstart that works on a fresh clone. - Examples directory with 1-2 runnable examples. - API reference (auto-generated is fine-mkdocs or just clear docstrings). - DESIGN.md updated with current state.

Output of Session A¶

Tests added; CI green.
Docs presentable.

Session B-Q3-closing post: outline + draft¶

Goal: Outline and draft a 3000-word substantive post.

Part 1-Outline (45 min)¶

1. Hook (250 w)
2. The problem and the niche (400 w)
3. The approach (700 w)-design choices, code snippets
4. Comparison vs incumbent (600 w)-table from M07-W03
5. What I learned (500 w)-about the specialty itself
6. Honest gaps (300 w)-what doesn't work yet
7. What's next (250 w)-Q4 capstone

Part 2-Draft (180 min)¶

Write. Use real numbers, code, and charts. The audience is practitioners in your specialty, not novices-pitch accordingly.

Part 3-Save + sleep on it (15 min)¶

Don't publish today. Sleep on it; edit Sunday.

Output of Session B¶

Drafted post.

Session C-Read incumbents' source + polish¶

Goal: Re-read source from a respected incumbent. Refine your post with insights.

Part 1-Source-reading (75 min)¶

Re-read: - Track A: Inspect AI's solver/scorer source. - Track B: AutoGen orchestration code or LangGraph state machines. - Track C: vLLM scheduler or SGLang's RadixAttention.

What did they do differently from your project? What's better in theirs? What's better in yours?

Add a "what these projects do better" honesty paragraph to your post.

Part 2-Edit pass (30 min)¶

Read aloud. Tighten.

Part 3-Push v0.5.0 (15 min)¶

Tag. Update CHANGELOG.

Output of Session C¶

Polished post.
v0.5.0 tagged.

End-of-week artifact¶

Track project at v0.5 with tests + docs
Q3-closing post drafted (~3000 words)
Source-reading notes from incumbents

End-of-week self-assessment¶

My track project would survive a code review by someone in the specialty.
My post is honest about what's working and what isn't.
I can articulate my niche in 30 seconds.

Common failure modes for this week¶

Continued feature creep. No. Freeze. Polish.
Defensive post tone. Honest is more credible.
Skipping the source re-reading. It's where the post's depth comes from.

What's next (preview of M09-W04)¶

Publish the Q3-closing post + Q4 capstone planning + Q3 retro + profile updates.