Project 2 — Research Opportunity Report

“What has been is what will be, and what has been done is what will be done, and there is nothing new under the sun.” — Ecclesiastes 1:9 (ESV)

Chapter: 2 — Finding a Problem Worth Solving Due: End of Week 2 Weight: 5% of the course grade (first graded deliverable) Submit: A week-02/ folder in your portfolio Git repo, pushed, containing research-opportunity-report.txt (or PDF), the completed question-scorer.csv, and ai-use.txt. Submit the repo link on Canvas. Allowed tools: Anything — Google Scholar, ACM DL, IEEE Xplore, arXiv, Semantic Scholar, dblp, OpenReview, Connected Papers, Zotero, and AI assistants. AI policy: Open. AI may brainstorm, generalize, and rephrase. AI may not supply an unverified gap or an unverified citation — you verify every reference in dblp/Semantic Scholar/publisher, and you own every claim. Log all generative AI use in ai-use.txt (Appendix C).

The Setup

You are a first-semester master’s researcher. Your advisor gives you the standard, terrifying instruction: “Find something to work on.” You have a domain you’re drawn to and a vague sense that AI is changing it — but a vague sense is not a thesis. Your advisor doesn’t want a topic. She wants to see that you can find a specific, defensible opening in a crowded field and tell the truth about how sure you are that it’s open.

This report is that conversation, on paper. It is deliberately short and deliberately sharp. Get it right and the next fourteen weeks have a spine. Get it vague and you’ll spend the semester re-deciding what you’re doing.

Learning Targets

You will demonstrate that you can:

• Commit to one CS domain and articulate how modern AI advances it.
• Distinguish a topic from a problem from a falsifiable research question.
• Perform gap analysis: back a claimed opening with a cited absence, not a hunch.
• Score and triage candidate questions with FINER and the Heilmeier Catechism.
• Use AI as a brainstorming tool while verifying every citation and being honest about what’s confirmed.
• Tell the truth about uncertainty (confirmed vs. unconfirmed) and disclose AI use.

Normal Tier

Meets the deliverable bar for the week: a clear, honest report your advisor can act on.

Required features

1. Three candidate domains from the course menu, each with one sentence on why AI is changing it now, and an Access note (data/compute/baseline you’d have).
2. Ten candidate research questions — at least three per domain across the first two domains — each measurable and comparative (names a metric and a baseline).
3. A completed question-scorer.csv (code/question-scorer.csv) scoring all ten on FINER (1–5 each) with a total and a one-line note.
4. A gap-analysis writeup for your top one-or-two questions using the three CARS moves (territory → niche → occupy) from code/gap-analysis-template.txt, with real citations.
5. confirmed/unconfirmed tags on every cited gap — honest about what you’ve verified versus what’s still a hunch for Chapter 3.
6. An ai-use.txt classifying each use assistive vs. generative.

Normal-tier rubric (out of 100)

Medium Tier (+ up to 25% extra credit)

Stronger rigor and novelty:

• Run a Heilmeier Catechism (code/heilmeier-catechism.txt) for your top two questions, answering #2 (limits of current practice) and #4 (who cares) with citations, not opinions. (+10%)
• Mine an OpenReview thread for your domain and tabulate at least four reviewer “the authors did not…” objections, promoting the ones that match your gaps to confirmed. (+8%)
• Run and report the AI-citation drill: ask for five references, verify each in dblp, and report your fabrication/metadata-error rate against the Walters & Wilder 2023 baseline (55%/18%). (+7%)

Hard Tier (+ up to 25% additional extra credit)

Publication-track ambition — and a judgment call no tool can make for you.

Write a one-page Domain Commitment Memo that an AI cannot write, because it requires you to weigh your own situation:

• Pick the one domain and the one question you will carry to the symposium, and defend the choice against the other two you’re rejecting — on Access, interest, and live edge (§2.2), not on which sounds most impressive.
• State, in plain Heilmeier-#1 language, the single sentence your eventual paper will be able to claim — and the honest probability you assign to actually being able to run it this semester given your real compute and data.
• Name the strongest reason your question might not be novel (the gap you most fear is already closed), and your concrete plan to confirm or kill it in Chapter 3.

The memo is graded on honesty and judgment, not optimism. A memo that talks you out of an impossible question and into a feasible-and-good one scores higher than one that promises the moon.

Submission

• Push week-02/ to your portfolio repo: research-opportunity-report.txt (or PDF), question-scorer.csv, ai-use.txt, and (if attempted) domain-commitment-memo.txt.
• Drop your gap-analysis paragraph, in rough, into your Week-1 acmart/IEEEtran skeleton’s introduction. Note the commit in your report.
• Submit the repo link on Canvas.

Hints

• Start from Future Work sections and the OpenReview record — that’s where reviewers and authors hand you certified gaps (§2.3).
• A question that can’t come back “no” isn’t a question. Re-read §2.1 and §2.5.
• Apply the Access filter first (§2.2). Reshape un-runnable questions toward API-based, public-benchmark, CPU-scale work before you fall in love with them.
• Use the Research Question Workshop widget (§2.x) to triage all ten fast before you write them up.
• If a citation came from an AI, assume it’s wrong until dblp says otherwise.

What Mastery Looks Like

A reader finishes your report and knows exactly what you’re going to study, why it’s open, how sure you are that it’s open, and what you’ll do next to find out. There is no jargon hiding a foggy idea. Every gap has a citation or an honest unconfirmed flag. The question is comparative and falsifiable. And the domain choice reads like a commitment, not a maybe.

Coach’s Note — The grade on this one is almost all in the gap analysis and the honesty of the tags. A confident “this gap is confirmed by Smith 2025 §6” beats ten vague questions. A humble “I believe this is open but haven’t verified it — Chapter 3 to-do” beats a fabricated certainty every time.

When You’re Done

• Three domains, each with “why AI now” + Access note.
• Ten measurable, comparative questions.
• question-scorer.csv complete with FINER scores.
• Gap-analysis writeup with verified citations and CARS moves.
• Every gap tagged confirmed/unconfirmed.
• ai-use.txt present and classified.
• Every AI-supplied citation verified in dblp/Semantic Scholar.
• Gap paragraph dropped into the paper skeleton; repo pushed; link submitted.

A theological footnote. This week’s verse — “there is nothing new under the sun” (Eccl. 1:9, ESV) — is exactly the despair that this report is built to defeat. Notice the report never asks you to invent something from nothing; it asks you to search a matter out (Prov. 25:2) — to find a specific, cited absence in an order that already exists and fill it honestly. That is the only kind of novelty available to creatures, and it is enough. The unconfirmed tag is theology, not just bookkeeping: it is the humility to say “I have not yet seen this clearly,” which is the posture of everyone who searches for a truth they did not make. Do the work as for the Lord, tell the truth about what you’ve checked, and let the smallness of your paragraph be a relief rather than a burden. The conversation was here before you and will outlast you; your honest contribution to it is the whole assignment.