Research Ideation

Research Ideation

From research goal to ranked ideas and a detailed proposal.

Step 0: Load evo-memory (M_I)
    ↓
Step 1: Define Scope & Goal
    ↓
Step 2: Literature Grounding (MUST use local-paper-navigator scripts)
    ↓
Step 3: Generate Ideas (3 Anchor Papers × Innovator persona)
    ↓
Step 4: Refine Ideas (3 tracks × N iterations)
    ↓
Step 5: ELO Tournament → Present Top-3 to User
    ↓
Step 6: Update evo-memory (IDE)
    ↓
User Selects
    ↓
Step 7: Expand into Proposal
    ↓
Step 8: Validate and Iterate

When to Use

• User wants to find a research direction or brainstorm research ideas within a specific quant scope stage
• User wants to evaluate whether an idea is novel or worth pursuing
• User wants to rank or compare multiple research ideas
• User wants to generate a research proposal from an idea anchored to an existing paper

Note: This pipeline is optimized for quantitative research where incremental, anchor-first contributions are preferred over architectural redesigns.

When NOT to Use

• Finding/reading papers → use local-paper-navigator
• Literature survey report → use research-survey
• Planning a paper (story design, experiment plan) → use paper-planning

Step 0: Load Prior Knowledge from evo-memory

Before any ideation begins, load Ideation Memory (M_I) from prior research cycles:

1. Read M_I at /memory/ideation-memory.md (refer to evo-memory skill)
2. Select the top-2 entries (k_I=2) most relevant to the user's current goal by comparing each entry's Summary and Retrieval Tags against the goal
3. Feasible directions from prior cycles → use as seeds in Step 3 (incorporate as candidate anchor directions alongside new ones, within the same scope stage)
4. Unsuccessful directions marked as fundamental failures → use during idea pruning in Step 4 (prune any idea that matches a fundamental failure pattern)
5. If M_I doesn't exist yet (first cycle), skip this step

This step prevents repeating known dead ends and builds on prior successes across research cycles.

Step 1: Define Research Scope & Goal

Research Scope

The long-term objective of this continual research program is to incrementally improve the quantitative research pipeline through publishable contributions in one of three core stages:

Each research session MUST focus on exactly one of the three stages above.

Hard constraints:

• The objective is not to redesign the entire pipeline, but to produce the smallest publishable improvement within a single stage.
• The proposed contribution should introduce one primary innovation, treating the remaining components as fixed background.
• Improvements should be incremental rather than architectural.

Research Goal

Within the chosen scope stage, define a concrete goal. Ask: "What is the smallest improvement that would be publishable in this stage?"

The goal should be narrow enough to complete in one research cycle, yet significant enough to advance the field.

Step 2: Literature Grounding (via local-paper-navigator)

Invoke local-paper-navigator to collect relevant papers from the local papers library. Do NOT skip this step or substitute with general knowledge — ideas must be grounded in real papers.

CRITICAL: All paper discovery in this step MUST use the local-paper-navigator skill and its scripts (local_search, xref_search, similar_papers, snippet_search, etc.). Using WebSearch, WebFetch, or any generic web search tool for finding papers is PROHIBITED. Generic web search returns blog posts, news articles, and low-quality results — only local-paper-navigator provides the local search, cross-reference, and keyword-similarity infrastructure needed for literature grounding.

Build Challenge-Insight Tree

From the collected papers, construct a challenge-insight tree — a many-to-many mapping between technical challenges and the insights/techniques that address them:

• Extract challenges: From each paper, what technical problem does it solve?
• Extract insights: What technique or key idea does it use?
• Map connections: Which insights address which challenges?

How this drives ideation:

• Challenges with few insights → unsolved problem (candidate for Step 3)
• Insights not yet applied to a challenge → cross-domain transfer opportunity (candidate for Step 4)
• Challenges with many insights → well-studied, avoid unless you have a fundamentally new angle

Also generate a condensed literature review synthesis as context for idea generation (for full surveys use research-survey).

See references/literature-tree.md for construction methodology.

Execution rule: Do NOT generate ideas without real paper grounding. The tree must reference actual papers with titles, sources, and findings. Paper search MUST go through local-paper-navigator — never use WebSearch/WebFetch as a shortcut.

Step 3: Generate Ideas

Generate 3 initial research ideas, each anchored to a specific paper from the literature grounding (Step 2), grounded in the literature.

Three Personas

Anchor-First Principle

Every proposal MUST be anchored to one Anchor Paper — a specific paper from the literature grounding (Step 2) that serves as the primary methodological foundation.

• ≥70% of the proposed method must be inherited from the Anchor Paper.
• The remaining ≤30% constitutes the innovation contribution.
• Prioritize extending an existing framework, not redesigning the entire system.
• The Anchor Paper's method is the baseline; the proposal's innovation is the delta above that baseline.

When generating ideas in Step 3, each idea must explicitly state:

• Anchor Paper: [title + paperId]
• Inherited components: [what is kept from the anchor, ≥70%]
• Innovation delta: [what is changed/added, ≤30%]

Single-Core Innovation

Each proposal may introduce at most 1 core innovation point (maximum 2 if tightly related — sharing the same mechanism or directly causally linked).

Innovation should come from refinement of existing methods — improvement, replacement, or extension — not from horizontal concatenation of unrelated methods or modules.

Disallowed: Combining technique A from paper X + technique B from paper Y where A and B address different problems and are not causally linked.

Allowed: Replacing paper X's optimization method with a more effective variant; extending paper X's factor mining pipeline with one additional module; adding one constraint to paper X's portfolio construction.

Process

1. Analyze literature + challenge-insight tree → select 3 candidate Anchor Papers (one per direction)
2. Generate one idea per Anchor Paper using Innovator persona
3. Each idea must follow Path 1 (Focused Contribution): single new component; clean hypothesis
   • Path 2 (System Contribution) is PROHIBITED under the single-core innovation constraint
4. Each idea must specify Anchor Paper, inherited components (≥70%), and innovation delta (≤30%)

Idea Format

# Research Idea: [Concise Title]

## Anchor Paper
- **Anchor Paper**: [title + paperId]
- **Inherited components**: [what is kept from the anchor, ≥70%]
- **Innovation delta**: [what is changed/added, ≤30%]

## Core Idea
[One paragraph: the proposal + which research direction it addresses + how the innovation delta extends the anchor]

## Validation Plan
[Concrete experiment outline. Datasets must be chosen from what is actually
available — run `quant-experiment-runtime`'s `discover_data.py --code-repo code-repo`
to list offline data packages, and use `local-paper-navigator` to recover the
paper's tested scope; plan around the intersection, scoped by the paper's test
range + budget + necessity (not the dataset's maximum coverage). Then: baselines,
metrics. See `references/proposal-extension.md` Section 4.]

## Baseline Feasibility
- **Anchor Paper source code**: [available at URL / ❌ no usable code]
- **Implementation mode (preliminary — for difficulty scoring)**: [Adapt / From-Scratch / Hybrid]
- **Difficulty correction**: [base score + adjustment = corrected score, e.g., 3+4=7 if From-Scratch]

Step 4: Refine Ideas

Run 3 parallel refinement tracks — one per initial idea. Each track uses all 3 personas.

For each track:
  For N=3 iterations:
    1. Evaluate current best idea (novelty, difficulty, relevance, clarity, anchor-coherence)
    2. All 3 personas generate refined versions based on evaluation
    3. Pick the best refinement as seed for next iteration
  Track champion = best idea across iterations

5 Evolution Strategies

1. Enhancement through Grounding: Strengthen with literature citations
2. Improving Coherence: Fix logical flaws in the mechanism
3. Inspiration and Combination: Combine with a different concept from literature
4. Simplification: Strip down to a clean, testable hypothesis
5. Literature-Driven Pivot: Abandon the mechanism; propose a new approach from literature

Critical rule: If evaluation says the approach is a dead-end, the persona MUST pivot — refinement is not restricted to patching.

Refinement Constraints

• Each refinement iteration MUST preserve the Anchor Paper as the methodological foundation. Pivoting to a different anchor paper is allowed, but adding new unrelated components is PROHIBITED.
• If refinement adds a second innovation point, it must be tightly related to the first (same mechanism or direct causal link).
• The 5 Evolution Strategies must operate within the anchor-first frame:
  • Enhancement through Grounding → strengthen the innovation delta with additional evidence
  • Improving Coherence → fix logical flaws within the inherited + innovation structure
  • Inspiration and Combination → combine with a concept from the Anchor Paper's domain, not an unrelated domain
  • Simplification → strip the innovation delta to its essential mechanism
  • Literature-Driven Pivot → replace the innovation delta with a better approach from literature, keeping the anchor foundation

Logical Cohesion Principles

• Too many variables → Focus via Subtraction: isolate the most promising variable
• Disconnected components → Justify via Strong Correlation: build explicit causal links

Step 5: ELO Tournament → Present Top-3

Rank all track champions through pairwise comparison, then present the top-3 to the user for selection.

Four Dimensions