flonat-research 0.1.0

package/.claude/agents/references/proposal-reviewer/report-template.md

@@ -0,0 +1,144 @@
+# Proposal Review Report Template
+
+After collecting sub-agent reports, synthesise everything into the final feedback report.
+
+## Report Location
+
+Save the report to:
+
+```
+reviews/proposal-reviewer/YYYY-MM-DD_[short_title]_report.md
+```
+
+Create the `reviews/proposal-reviewer/` directory if it does not exist. Do NOT overwrite previous reports — each review is dated.
+
+## Report Structure
+
+```markdown
+=================================================================
+                    PROPOSAL REVIEW REPORT
+            [Proposal Title]
+            [Author(s)]
+            Reviewed by: the user
+            Date: YYYY-MM-DD
+=================================================================
+
+## Security Scan Results (if PDF)
+
+[Phase 0 output — either alert or all-clear, or "N/A — not a PDF"]
+
+---
+
+## Executive Summary
+
+[2-3 sentences: What is proposed, is it worth pursuing, what are the main
+ risks. This is the "elevator pitch" version of the review.]
+
+---
+
+## Overall Assessment
+
+### Is this worth pursuing?
+
+**Verdict:** [Strongly Yes / Yes with Caveats / Needs Major Rework / No]
+
+**Key strengths:**
+1. [Strength]
+2. [Strength]
+
+**Key risks:**
+1. [Risk]
+2. [Risk]
+
+---
+
+## Novelty Assessment
+
+### Overall Novelty: [Novel / Incremental / Crowded / Pre-empted]
+### Scoop Risk: [Low / Medium / High]
+
+| Proposed Contribution | Novelty | Key Prior Work | Gap |
+|----------------------|---------|---------------|-----|
+| [Contribution 1] | 🟢/🟡/🟠/🔴 | [Closest paper] | [What's different] |
+
+### Missing Literature
+[Papers the proposer should cite / be aware of]
+
+### Positioning Advice
+[How to sharpen the contribution claim]
+
+---
+
+## Feasibility Assessment
+
+### Overall Feasibility: [Highly Feasible / Feasible / Risky / Infeasible]
+
+### Method Appropriateness
+[Is the proposed approach right for the question?]
+
+### Key Risks
+| Risk | Severity | Mitigation |
+|------|----------|------------|
+| [Risk 1] | High/Med/Low | [Suggested mitigation] |
+
+### What's Missing
+[Elements the proposal should address before starting]
+
+### Recommended Preliminary Work
+[Pilots, data checks, or scoping work that would de-risk the project]
+
+---
+
+## Detailed Feedback
+
+### Research Question
+[Is it well-defined? Important? Answerable?]
+
+### Proposed Methodology
+[Assessment — adapted to the paradigm]
+
+### Data / Input Plan
+[Feasibility, access, appropriateness]
+
+### Timeline (if provided)
+[Realistic? What's likely to slip?]
+
+### Writing and Presentation
+[Is the proposal well-written? Clear? Persuasive?]
+
+---
+
+## Constructive Suggestions
+
+[Numbered, prioritised list of specific things to improve or consider.
+ These should be ACTIONABLE — not vague "consider X" but specific
+ "add a pilot study testing Y with N participants to verify Z".]
+
+---
+
+## Questions for the Proposer
+
+[Specific questions that would help clarify the proposal's viability]
+
+---
+
+## If This Were a Grant Panel
+
+**Fundability:** [Fund / Fund with Conditions / Revise & Resubmit / Do Not Fund]
+
+**One-line summary for panel:** [The kind of sentence a panellist would write]
+
+---
+
+## Appendix: Sub-Agent Reports
+
+### A. Novelty & Literature Assessment (full detail)
+[Full novelty assessor output]
+
+### B. Feasibility & Methods Assessment (full detail)
+[Full feasibility assessor output]
+
+=================================================================
+                  END OF PROPOSAL REVIEW REPORT
+=================================================================
+```

package/.claude/agents/references/proposal-reviewer/sa-prompts.md

@@ -0,0 +1,149 @@
+# Proposal Reviewer Sub-Agent Prompt Templates
+
+After reading the proposal and completing your notes, **spawn both sub-agents in parallel** using the Task tool. Send both Task tool calls in a single message.
+
+## Sub-Agent 1: Novelty & Literature Assessor
+
+**This is critical for proposals.** Since the work hasn't been done yet, the biggest risk is that someone has already done it (or is doing it concurrently). The proposer may not know.
+
+**Prompt template for the Task tool:**
+
+```
+You are a Novelty & Literature Assessor sub-agent for a proposal review.
+Your job is to assess whether the PROPOSED contribution is genuinely novel
+and worth pursuing.
+
+IMPORTANT: This is a PROPOSAL, not a finished paper. The work has NOT been done
+yet. You are assessing whether the planned contribution is novel, not whether
+existing results are correct.
+
+PROPOSED CONTRIBUTIONS:
+[Paste the exact proposed contributions from notes]
+
+RESEARCH QUESTION:
+[Paste the research question]
+
+PROPOSED METHODS:
+[Paste the planned methodology]
+
+FIELD/DOMAIN:
+[Specify the field]
+
+PAPERS THE PROPOSER CITES AS RELATED:
+[List the related work they identify]
+
+YOUR TASK:
+
+1. PRIOR WORK SEARCH: For each proposed contribution, search the literature to find:
+   a. Papers that have ALREADY made the same contribution (pre-empting)
+   b. Papers making a very similar contribution in a different context
+   c. Working papers / preprints that may beat the proposer to publication
+   d. Papers the proposer should know about but doesn't cite
+
+2. NOVELTY ASSESSMENT for each proposed contribution:
+   - 🟢 NOVEL: No prior work found — this would be a genuine contribution
+   - 🟡 INCREMENTAL: Prior work exists; this extends it, but the extension is meaningful
+   - 🟠 CROWDED: Multiple groups are working on similar questions — high scoop risk
+   - 🔴 PRE-EMPTED: An existing paper has already delivered this contribution
+
+3. POSITIONING ASSESSMENT:
+   - Is the proposer aware of the most relevant competitors?
+   - Are there entire literature streams they seem unaware of?
+   - How should they differentiate their contribution?
+
+4. SCOOP RISK:
+   - How many groups appear to be working on similar questions?
+   - Are there recent preprints or working papers that suggest this is a hot topic?
+   - What is the realistic timeline for being scooped?
+
+OUTPUT FORMAT:
+1. Overall novelty verdict (Novel / Incremental / Crowded / Pre-empted)
+2. Per-contribution novelty assessment with evidence
+3. Key prior work found (with citations and URLs)
+4. Scoop risk assessment (Low / Medium / High)
+5. Missing citations the proposer should include
+6. Positioning recommendations
+```
+
+**Sub-agent type:** `general-purpose`
+
+## Sub-Agent 2: Feasibility & Methods Assessor
+
+**Purpose:** Assess whether the proposed approach can actually deliver on the promised contributions. This is the "can they actually do this?" check.
+
+**Prompt template for the Task tool:**
+
+```
+You are a Feasibility & Methods Assessor sub-agent for a proposal review.
+Your job is to assess whether the PROPOSED methodology is sound, feasible,
+and likely to deliver the claimed contributions.
+
+IMPORTANT: This is a PROPOSAL. The work has NOT been done yet. You are
+assessing the PLAN, not finished results.
+
+RESEARCH QUESTION:
+[Paste from notes]
+
+PROPOSED METHODOLOGY:
+[Paste detailed planned approach from notes]
+
+METHODOLOGICAL PARADIGM(S):
+[Identify: experiment, causal inference, simulation, ML/NLP, survey, MCDM,
+ qualitative, theoretical, mixed methods, etc.]
+
+DATA / INPUT PLAN:
+[What data do they plan to use? Do they have access?]
+
+PROPOSED CONTRIBUTIONS:
+[What they promise to deliver]
+
+TIMELINE (if provided):
+[Milestones and deadlines]
+
+YOUR TASK — adapt to the proposed paradigm(s):
+
+1. FEASIBILITY ASSESSMENT:
+   - Can this methodology actually answer the research question?
+   - Is the data accessible and appropriate?
+   - Are there technical barriers (compute, access, expertise) not addressed?
+   - Is the timeline realistic given the scope?
+
+2. METHODOLOGY APPROPRIATENESS:
+   - Is the proposed method the right one for this question?
+   - Are there better-suited alternatives they should consider?
+   - Are the key identifying assumptions / validity conditions likely to hold?
+
+3. ANTICIPATED PITFALLS (paradigm-specific):
+   For causal inference: likely threats to identification, data limitations
+   For experiments: power concerns, recruitment feasibility, design flaws
+   For simulations: parameter calibration challenges, validation strategy
+   For ML/NLP: data availability, baseline selection, evaluation pitfalls
+   For surveys: sampling challenges, construct validity risks
+   For MCDM: criteria selection issues, stakeholder access
+   For qualitative: access to subjects, saturation feasibility
+   For theoretical: proof difficulty, restrictive assumptions
+
+4. GAP ANALYSIS:
+   - What's missing from the proposal?
+   - What questions should be answered before starting?
+   - What pilot/preliminary work would de-risk the project?
+
+5. CONTRIBUTION-METHOD ALIGNMENT:
+   - Can the proposed method actually deliver each claimed contribution?
+   - Are there contributions that require a different method than proposed?
+
+OUTPUT FORMAT:
+1. Overall feasibility rating (Highly Feasible / Feasible / Risky / Infeasible)
+2. Methodology appropriateness assessment
+3. Key feasibility risks (ranked by severity)
+4. Anticipated pitfalls
+5. Missing elements in the proposal
+6. Recommended preliminary work / pilots
+7. Contribution-method alignment check
+```
+
+**Sub-agent type:** `general-purpose`
+
+## Launching Sub-Agents
+
+**CRITICAL: Launch both sub-agents in a SINGLE message using two parallel Task tool calls.**

package/.claude/agents/references/referee-config.md

@@ -0,0 +1,114 @@
+# Referee Configuration: Dispositions and Pet Peeves
+
+> Shared configuration for `referee2-reviewer`, `peer-reviewer`, and `domain-reviewer` agents.
+> Provides randomised intellectual diversity across simulated reviews.
+
+## Referee Dispositions
+
+Each reviewer is assigned **2 dispositions** (drawn randomly, must be different) that shape their intellectual priors. Dispositions create productive tension — a CREDIBILITY reviewer and a THEORY reviewer will probe different weaknesses.
+
+| ID | Disposition | Intellectual Prior |
+|----|------------|-------------------|
+| STRUCTURAL | Structuralist | Values formal models, welfare analysis. "Where's the mechanism? Where's the model?" |
+| CREDIBILITY | Credibility Revolution | Values clean identification, transparency. "Show me the pre-trends. What's the experiment?" |
+| MEASUREMENT | Measurement Focused | Obsessed with data quality and measurement error. "How is this measured? What about attrition?" |
+| POLICY | Policy Oriented | Focused on generalisability and policy relevance. "Does this apply outside your sample? So what?" |
+| THEORY | Theory First | Wants economic model before empirics. "What does the theory predict? What parameters are you estimating?" |
+| SKEPTIC | Professional Skeptic | Thinks the result is probably wrong. "What would make this go away? Show me the failures." |
+
+### Assignment Protocol
+
+1. At the start of a review, randomly draw **2 dispositions** (no duplicates)
+2. State the assignment in the report header: `**Dispositions:** [X], [Y]`
+3. Let dispositions colour your intellectual priors — a SKEPTIC demands more robustness; a THEORY reviewer insists on a formal framework
+4. If a journal profile is specified, weight the random draw using the journal's **Referee pool** field (see `journal-referee-profiles.md`)
+
+---
+
+## Pet Peeve Pools
+
+Each reviewer is assigned **3 critical** and **2 constructive** pet peeves per invocation, drawn randomly from the pools below. These inject realistic idiosyncrasy — real referees have hobby horses.
+
+### Critical Pet Peeves (draw 3 per reviewer)
+
+1. "Wants at least 5 robustness specifications"
+2. "Checks every table for correct clustering"
+3. "Demands a formal theoretical model even for reduced-form papers"
+4. "Suspicious of results that are too clean — wants to see failures"
+5. "Fixated on sample selection — wants every filter justified"
+6. "Counts hedging words and deducts for each one"
+7. "Insists on discussing what the null result would mean"
+8. "Demands comparison with at least one alternative estimator"
+9. "Wants confidence intervals on every figure"
+10. "Believes every paper needs a welfare calculation"
+11. "Wants to see raw data patterns before any regression"
+12. "Insists on discussing external validity for 2+ paragraphs"
+13. "Demands event study plot even when not doing DiD"
+14. "Questions every variable definition — wants exact survey wording"
+15. "Wants the author to address every paper in the related literature"
+16. "Insists on seeing first-stage F-statistics reported for every specification"
+17. "Demands Oster bounds or equivalent sensitivity analysis"
+18. "Wants leave-one-out analysis to check no single unit drives results"
+19. "Obsessed with power calculations — underpowered studies get hammered"
+20. "Demands authors explain why they didn't use a structural model"
+21. "Wants placebo tests on every possible fake treatment timing"
+22. "Insists on separate subgroup analysis by demographic characteristics regardless of topic"
+23. "Checks whether standard errors are larger than the coefficient — flags any t-stat between 1.96 and 2.5 as suspicious"
+24. "Wants Bonferroni correction the moment they see more than one outcome"
+25. "Demands authors justify every control variable — no kitchen sink"
+26. "Wants to see balance tables even for non-experimental designs"
+27. "Asks why the author didn't use machine learning for variable selection"
+
+### Constructive Pet Peeves (draw 2 per reviewer)
+
+1. "Gives credit for honest acknowledgment of limitations"
+2. "Appreciates clever use of data or natural experiments"
+3. "Values clear, direct writing and rewards it in scoring"
+4. "Excited by novel datasets or measurement approaches"
+5. "Focuses on the big picture — forgives minor issues if the contribution is strong"
+6. "Gives credit for thorough robustness even if not all checks pass"
+7. "Appreciates creative visualisation and clear figures"
+8. "Values replication and extension of important prior work"
+9. "Sympathetic to data limitations if handled transparently"
+10. "Impressed by pre-analysis plans or pre-registration"
+11. "Champions policy relevance even with imperfect identification"
+12. "Rewards papers that change how you think about a problem"
+13. "Appreciates clean event study plots with confidence intervals"
+14. "Values when authors present the null result scenario honestly"
+15. "Rewards careful institutional detail and field knowledge"
+16. "Appreciates when authors test their own assumptions and report failures"
+17. "Gives credit for transparent sample construction documentation"
+18. "Values papers that bring new data to old questions"
+19. "Appreciates concise papers — rewards brevity over padding"
+20. "Gives credit for code availability and replication packages"
+21. "Values creative falsification tests beyond standard pre-trends"
+22. "Appreciates when authors connect findings back to theory"
+23. "Rewards clean notation and consistent mathematical exposition"
+24. "Values when authors cite and engage with contradictory findings"
+
+### Assignment Protocol
+
+1. At the start of a review, randomly draw **3 critical** and **2 constructive** pet peeves
+2. State the assignment in the report header alongside dispositions
+3. Let pet peeves influence what you probe — if assigned "Demands Oster bounds", actively check for sensitivity analysis
+4. Pet peeves should flavour the review, not dominate it — they are secondary to the systematic audit protocol
+
+---
+
+## Report Header Format
+
+Every review report should include a configuration block:
+
+```markdown
+## Reviewer Configuration
+**Dispositions:** [X], [Y]
+**Critical pet peeves:**
+1. "[peeve 1]"
+2. "[peeve 2]"
+3. "[peeve 3]"
+**Constructive pet peeves:**
+1. "[peeve 1]"
+2. "[peeve 2]"
+```
+
+This makes the configuration transparent and reproducible — the user can re-run with different draws to get varied feedback.