scope-control

Scope Control (Meta-Control Layer)

Overview

A focused paper with 1–2 strong contributions beats a scattered paper with 5 weak ones. This skill can trigger at ANY phase when scope expansion is detected.

Core principle: Focus is a feature. Scope creep is a bug.

Violating the letter of this rule is violating the spirit of this rule.

Trigger Conditions

Any ONE of these activates this skill:

• Contribution points or innovation claims exceed 2
• Experiment matrix is exploding (too many method × dataset × setting combinations)
• Story line is splitting into unrelated threads
• Time or resource budget is being exceeded
• User adds "one more thing" that substantially changes project scope

The Mandatory Response

When triggered, the agent MUST execute all four steps in order:

1. WARN:     "Current scope may be too large for [target venue] as a single paper."
2. QUANTIFY: "Current experiment plan requires ~X GPU-hours. Budget is Y. Excess: Z%."
             or "There are N contribution points. A focused paper typically has 1-2."
3. PROPOSE:  Reduction options, prioritized:
             a) Core contribution to keep
             b) Items to defer to future work
             c) Items to cut entirely
4. PRESENT:  User decides. Agent cannot decide scope reduction on its own.

Skip any step = scope creep enabled = paper quality at risk

Rules

• The agent proposes reductions but never executes them without user approval.
• Deferred items go into an explicit "future work" log — they are not forgotten, just prioritized out.
• After user decides, update the plan and re-verify scope is within bounds.

Red Flags — STOP

• Adding a third major contribution
• Experiment table exceeding one page
• Story line requiring two separate introductions
• "Just one more baseline" appearing more than twice
• Resource estimate exceeding budget by more than 30%

Rationalization Prevention

Excuse	Reality
"We can fit it all in"	Page limits exist. Reviewers penalize unfocused papers.
"Each piece is small"	Small pieces compound into confusion.
"We already did the work"	Sunk cost. Extra work that dilutes the paper hurts more than helps.
"The reviewer might ask for it"	Anticipate reviewers, but don't preemptively answer every possible question.