CSE Roadmap

Common tier · about 4 hours · 12 parts

Part 2 of 12 · 2 min read

A scope sentence with a number and a baseline is the cheapest quality gate

A project you can't state as one sentence with a number in it is a project you haven't actually scoped yet, however much work has gone into it.

Think of a contractor who starts pouring concrete before anyone agrees what the building is for, how big it needs to be, or what "finished" even looks like. The work might be excellent. But there's no way to tell.

In plain words: a capstone's scope is the user, job, measured result, baseline and limits you commit to, plus what you explicitly leave out. A checker can mechanically flag the specific gaps in it — a vague user, no baseline, no target. That lets you see a scope is unfinished before a single line of the project gets built.

A worked example. A deliberately vague fictional scope ("for users, the agent helps with tickets") raises 5 flags: a vague user, a multi-part job, no measurable result, no simple baseline, no cost or latency limit. A partially-filled scope raises 4 flags. These include an evaluation set that's too small (60 items against roughly the 155 needed for the planned gain) and a plan that already exceeds the available hours at the median. A tight scope — one user, one job, a number, a baseline, limits — raises zero flags.

The key lines. From code/mini-scope/scope.py:

# from code/mini-scope/scope.py
def check(s: Scope):
    out = []
    u = s.user.lower()
    if any(w in u for w in VAGUE_USERS) and len(u.split()) < 6:
        out.append(("user is vague", "name one person with a role and a situation"))
    if " and " in s.job.lower() or s.job.count(",") >= 1:
        out.append(("job has several parts", "pick the single job the user hires the agent for"))

Read it as: the checker looks for specific textual patterns — a generic word like "users," an "and" joining two jobs. These are crude heuristics, not a guarantee of a good scope. But they catch the single most common failure mode: a scope so broad it commits to nothing measurable.

Scope, baseline

Scope is the user, job, result, evidence and limits you commit to, plus what you explicitly leave out. The baseline is the score of the current or simplest alternative — a human process, rules, a single prompt — that your agent has to beat.

Example: the "no simple baseline" flag above exists because a result is only meaningful relative to something. A 70 percent success rate means nothing on its own. You need to know whether the simplest rule-based approach already gets 68 percent or only 20 percent.