utd24-stat4-methods

name

utd24-stat4-methods

description

Use this skill for developing literature-based research ideas into reproducible statistical, computational, and decision-analytic methods for operational data, especially projects targeting UTD24-style management journals and top statistics journals. Use it for project intake, anchor paper analysis, contribution design, baseline reproduction, numerical experiments, dataset experiments, and claim-evidence audit. Do not use it for generic paper monitoring, finance/accounting analytics, humanities-style literature synthesis, or pure proof development unless explicitly requested.

UTD24-Stat4 Methods

Purpose

This skill supports automated research development for data-driven analytical method papers in management science, operations, marketing, information systems, statistics, econometrics, and adjacent fields.

It is designed for projects that start from:

• an initial research idea;
• one or more anchor papers;
• available or planned datasets;
• existing or planned code;
• numerical, simulation, benchmark, or dataset experiments;
• a target paper positioning such as UTD24-oriented, Stat4-oriented, hybrid, field-journal-oriented, or working-paper-stage research.

The skill helps transform an initial idea into an auditable research project by moving through:

1. project intake;
2. anchor paper analysis;
3. contribution design;
4. go / no-go evaluation;
5. reproduction planning and reporting;
6. method specification;
7. experiment design;
8. result interpretation;
9. claim-evidence audit;
10. paper-ready packaging;
11. anchor writing style analysis;
12. LaTeX manuscript planning and section drafting.

The skill is not intended to replace formal proof development. Formal theoretical proof construction and proof verification should be handed off to a separate MineProof-style workflow when needed.

---

Core Philosophy

The central unit of a research paper is not a list of components. It is a unified core Q&A.

A strong paper should answer:

• What question does the paper ask?
• Why is the question important?
• Why is the question challenging?
• What does the paper answer?
• What knowledge boundary does the answer expand?
• What evidence supports the answer?
• What claims must be qualified, downgraded, or removed?

The assistant must evaluate and develop research projects through three contribution dimensions:

• Significancy: whether the problem and result matter in terms of profit, cost, efficiency, social value, statistical validity, decision quality, or scientific understanding.
• Theorization: whether the problem is formalized, whether the gap is challenging, whether the method or result generates analytical or structural insight, and whether formal claims are properly supported.
• Generalization: whether the result survives robustness, sensitivity, extension, external validity, multiple settings, or clearly stated boundary conditions.

The skill should avoid treating a project as valuable merely because it has:

• a new dataset;
• a new pipeline;
• a reproduction of an existing paper;
• an implementation tweak;
• an extra metric;
• a private dataset;
• a benchmark improvement without mechanism;
• a theorem statement without proof;
• experiments without claim-evidence discipline.

---

Global Rules

1. Claims Must Follow Evidence

Do not state claims stronger than the available evidence.

Use the Claim-Evidence Audit as the controlling source once it exists.

If a claim is:

• Supported: it may be stated directly.
• Partially Supported: it must be qualified.
• Evidence Pending: it must not be stated as established.
• MineProof Pending: it must not be written as a proven theorem.
• Overclaimed: it must be downgraded before use.
• Unsupported: it must not be used as a paper claim.
• Contradicted: it must be removed or stated only as a limitation / boundary condition.
• Remove: it must not appear in the manuscript.

2. Do Not Force a Decision-Making Frame

If the project has no explicit action, treatment, policy, allocation, intervention, or decision variable, do not force it into a decision-making frame.

Such projects may still be valuable as:

• statistical inference;
• hypothesis testing;
• measurement;
• latent structure recovery;
• representation learning;
• data construction;
• platform analytics;
• evaluation methodology;
• empirical or operational analytics.

3. Reproduction Is a Means, Not a Contribution by Itself

Reproduction should clarify:

• whether anchor methods are understood;
• whether baselines are credible;
• whether code and data are usable;
• whether the current project can build on or depart from the anchor paper.

Do not treat reproduction alone as a paper contribution.

4. Experiments Are Evidence, Not Decoration

Every experiment must support, weaken, test, or falsify a claim.

Do not add experiments merely to increase the number of tables or figures.

5. Negative Results and Failure Cases Must Be Interpreted

Failed, weak, unstable, or contradictory results should not be hidden.

They may imply:

• claim downgrade;
• method revision;
• additional experiment;
• boundary condition;
• theory patch;
• return to Go / No-Go;
• project stopping.

6. Anchor Papers Are Structural References, Not Text Sources

Use anchor papers to learn:

• section structure;
• paragraph logic;
• transition patterns;
• method exposition;
• theorem presentation;
• experiment interpretation.

Do not copy anchor-paper wording.

7. User-Provided LaTeX Templates Are Authoritative

When writing into LaTeX, follow the user-provided LaTeX folder or journal template.

Do not assume or impose a default file tree.

Do not create new manuscript structure unless the user explicitly approves it.

---

Workflow Overview

The full workflow is:

Project Intake
→ Anchor Paper Analysis
→ Contribution Design
→ Go / No-Go Evaluation
→ Reproduction
→ Method Specification
→ Experiment Design
→ Result Interpretation
→ Claim-Evidence Audit
→ Paper-Ready Packaging
→ Anchor Writing Style Analysis
→ LaTeX Manuscript Writing

The assistant should not always run the full workflow. Use the workflow that matches the user's current stage.

---

Routing Guide

If the user provides a new research idea

Use:

• workflows/project-intake.md
• templates/project-card.md

Generate:

• reports/00_project-card.md

Goal:

• clarify project type;
• identify research object;
• identify data, code, anchor papers, target positioning, and open questions.

---

If the user provides anchor papers

Use:

• workflows/anchor-paper-analysis.md
• templates/anchor-paper-map.md

Generate:

• reports/01_anchor-paper-map.md
• optionally reports/anchor-notes/*.md

Goal:

• identify anchor paper Q&A;
• extract reusable methods, data structures, assumptions, experiments, baselines, and limitations;
• avoid superficial paper summaries.

---

If the user wants to clarify the paper contribution

Use:

• workflows/contribution-design.md
• templates/contribution-design.md
• references/contribution-rubric.md

Generate:

• reports/02_contribution-design.md

Goal:

• define the unified core Q&A;
• evaluate Significancy, Theorization, and Generalization;
• create claim-evidence map;
• decide what evidence is needed.

---

If the user asks whether the idea is worth continuing

Use:

• workflows/go-no-go-evaluation.md
• templates/go-no-go-report.md

Generate:

• reports/03_go-no-go-report.md

Goal:

• evaluate feasibility, novelty, evidence leverage, target-tier potential, and stopping conditions;
• decide whether to proceed, revise, pause, or stop.

---

If the project needs to reproduce an anchor paper or baseline

Use:

• workflows/reproduction.md
• templates/reproduction-plan.md
• templates/reproduction-report.md

Generate:

• reports/04_reproduction-plan.md
• reports/05_reproduction-report.md

Goal:

• decide whether reproduction is needed;
• define reproduction target;
• execute or plan reproduction;
• compare reproduced results to anchor paper;
• update evidence and baseline credibility.

---

If the project needs to define the proposed method

Use:

• workflows/method-specification.md
• templates/method-spec.md

Generate:

• reports/06_method-spec.md

Goal:

• specify method before implementation;
• define data interface, target object, method components, novelty-bearing components, mechanism of improvement, beyond-anchor design requirements, fairness constraints, theory needs, and experiment implications.

---

If the project needs experiments designed

Use:

• workflows/experiment-design.md
• templates/experiment-plan.md

Generate:

• reports/07_experiment-plan.md

Goal:

• map claims to experiments;
• design numerical, synthetic, dataset, benchmark, ablation, robustness, and sensitivity experiments;
• define baselines, metrics, success criteria, failure criteria, logging, and execution plan.

---

If experiment results are available

Use:

• workflows/result-interpretation.md
• templates/result-interpretation-report.md

Generate:

• reports/08_result-interpretation-report.md

Goal:

• check execution validity;
• describe main results;
• compare results against expected mechanism;
• interpret baseline failures, ablations, robustness, unexpected results, and boundary conditions;
• update claims.

---

If the project needs an evidence audit

Use:

• workflows/claim-evidence-audit.md
• templates/claim-evidence-audit.md

Generate:

• reports/09_claim-evidence-audit.md

Goal:

• extract claims;
• normalize claims into auditable form;
• map evidence to claims;
• audit validity, strength, and scope;
• identify overclaims and contradictions;
• decide final claim status and project readiness.

---

If the project is ready to become a paper package

Use:

• workflows/paper-ready-packaging.md
• templates/paper-ready-package.md

Generate:

• reports/10_paper-ready-package.md

Goal:

• create final paper-level Q&A;
• define final contribution statement;
• position target journal;
• map evidence to paper claims;
• select main tables and figures;
• define theory status;
• create paper outline, abstract skeleton, introduction blueprint, limitations, and remaining TODOs.

---

If the user wants to learn from anchor-paper writing style

Use:

• workflows/anchor-writing-style-analysis.md
• templates/anchor-writing-style-note.md

Generate:

• reports/11_anchor-writing-style-note.md
• or reports/anchor-writing-style-notes/[paper-short-name].md

Goal:

• decompose the anchor paper's writing logic section by section, subsection by subsection, and paragraph by paragraph;
• extract rhetorical functions, transition logic, reusable patterns, and non-reusable paper-specific moves.

---

If the user wants to write into LaTeX

Use:

• workflows/latex-manuscript-writing.md
• templates/latex-manuscript-plan.md
• templates/section-draft-plan.md
• references/manuscript-section-rules.md
• references/manuscript-section-blueprints.md
• references/latex-writing-rules.md

Generate:

• reports/12_latex-manuscript-plan.md
• reports/section-plans/[section-name]-plan.md
• updated .tex files only if the user approves editing.

Goal:

• inspect the user-provided LaTeX template;
• map paper-ready content into template locations;
• plan sections before drafting;
• draft or edit one section at a time;
• preserve labels, citations, macros, theorem environments, figure paths, table formatting, and build system;
• prevent unsupported claims from entering the manuscript.

---

Required Artifacts and Their Roles

Reports

Artifact	Role
reports/00_project-card.md	Initial project definition
reports/01_anchor-paper-map.md	Anchor paper structure and reuse map
reports/02_contribution-design.md	Unified Q&A and contribution design
reports/03_go-no-go-report.md	Project feasibility and continuation decision
reports/04_reproduction-plan.md	Reproduction planning
reports/05_reproduction-report.md	Reproduction outcome and baseline credibility
reports/06_method-spec.md	Proposed method specification
reports/07_experiment-plan.md	Experiment design
reports/08_result-interpretation-report.md	Interpretation of experiment results
reports/09_claim-evidence-audit.md	Claim-evidence audit
reports/10_paper-ready-package.md	Paper-ready research package
reports/11_anchor-writing-style-note.md	Anchor paper writing-logic analysis
reports/12_latex-manuscript-plan.md	Full manuscript-to-template plan
reports/section-plans/[section-name]-plan.md	Section-level paragraph plan

References

Artifact	Role
references/contribution-rubric.md	Contribution evaluation through Significancy, Theorization, and Generalization
references/positioning.md	Scope, target research class, and MineProof boundary control
references/manuscript-section-rules.md	Section-level manuscript writing rules
references/manuscript-section-blueprints.md	Default paragraph blueprints for manuscript sections
references/latex-writing-rules.md	LaTeX editing, integration, and safety rules

Supporting Docs

Artifact	Role
docs/file-map.md	Human-readable map of skill architecture

Workflows

Workflow	Role
workflows/project-intake.md	Start and classify a project
workflows/anchor-paper-analysis.md	Analyze anchor papers as research references
workflows/contribution-design.md	Design unified contribution and evidence path
workflows/go-no-go-evaluation.md	Decide whether to proceed
workflows/reproduction.md	Plan and evaluate reproduction
workflows/method-specification.md	Define the method before implementation
workflows/experiment-design.md	Design experiments
workflows/result-interpretation.md	Interpret experiment results
workflows/claim-evidence-audit.md	Audit claims against evidence
workflows/paper-ready-packaging.md	Package the project for manuscript writing
workflows/anchor-writing-style-analysis.md	Analyze anchor paper writing logic
workflows/latex-manuscript-writing.md	Plan and draft LaTeX manuscript sections

---

Stage Gates

The assistant should stop at explicit checkpoints after major stages.

After Project Intake

Stop if:

• research object is unclear;
• data availability is unclear;
• anchor papers are missing;
• target contribution is too vague.

After Contribution Design

Stop if:

• unified core Q&A is unclear;
• contribution is only a collection of technical components;
• required evidence is infeasible;
• MineProof needs are central but unavailable.

After Go / No-Go Evaluation

Stop if:

• project is low-significance and low-feasibility;
• target tier is unrealistic without major evidence;
• idea is a minor tweak without mechanism or data leverage.

After Reproduction Planning

Stop if:

• reproduction is not necessary;
• reproduction is infeasible;
• user must choose whether to reproduce before proceeding.

After Method Specification

Stop if:

• method is not linked to the core Q&A;
• novelty-bearing component is unclear;
• beyond-anchor design requirements are unresolved;
• fairness constraints are unclear.

After Experiment Design

Stop if:

• essential experiments are not linked to claims;
• success and failure criteria are not defined;
• baselines or metrics are unfair;
• user has not approved execution.

After Result Interpretation

Stop if:

• execution validity is weak;
• results are uninterpretable;
• results contradict the expected mechanism;
• additional experiments are needed before audit.

After Claim-Evidence Audit

Stop if:

• core claims are unsupported;
• overclaims are unresolved;
• MineProof-pending claims are being treated as established;
• project is not ready for paper-ready packaging.

After Paper-Ready Packaging

Stop if:

• final Q&A is unstable;
• target positioning is unrealistic;
• main evidence does not match final contribution;
• manuscript writing would reintroduce unsupported claims.

Before LaTeX Drafting

Stop if:

• user-provided LaTeX template has not been inspected;
• section draft plan has not been generated;
• claims are not mapped to evidence;
• template location is unclear.

---

File and Template Discipline

When generating artifacts:

• use the specified template when available;
• preserve the report numbering convention;
• mark uncertain items as Unclear, Needs user decision, Evidence pending, MineProof pending, or TODO;
• do not invent data, results, citations, proof status, or file paths;
• do not treat placeholders as real evidence;
• do not proceed automatically past a checkpoint unless the user explicitly asks.

When writing Markdown artifacts:

• avoid nested code fences inside large Markdown code blocks;
• use indented blocks for example file structures or pseudocode;
• keep tables readable;
• use explicit status labels.

When writing LaTeX:

• inspect the user-provided template first;
• preserve existing structure;
• do not impose a default file tree;
• do not delete labels, citations, macros, bibliography entries, theorem environments, or figure paths;
• do not invent BibTeX keys;
• add explicit TODO comments for missing citations or missing results;
• do not write MineProof-pending results as theorems;
• do not insert claims rejected by Claim-Evidence Audit.

---

Common Failure Modes to Avoid

Avoid:

1. turning an idea into a contribution without evidence;
2. treating reproduction as novelty;
3. treating a pipeline as a methodological contribution;
4. treating private data as automatic journal-level contribution;
5. using prediction metrics to claim decision value without downstream evidence;
6. using simulation to claim real-world validity;
7. writing broad generalization from one dataset;
8. stating MineProof-pending theory as established;
9. ignoring negative results or boundary conditions;
10. designing experiments unrelated to claims;
11. adding many low-value robustness checks;
12. copying anchor-paper prose;
13. overfitting the manuscript to one anchor paper;
14. forcing a decision-making frame when no action variable exists;
15. writing a manuscript before Claim-Evidence Audit;
16. writing LaTeX before inspecting the template;
17. creating a default LaTeX structure when the user provides a template.

---

Default Response Behavior

When the user asks what to do next:

• identify the current stage;
• recommend the next workflow or template;
• explain why it is next;
• avoid expanding the system unnecessarily.

When the user asks for a file:

• provide the requested file directly;
• keep it consistent with the existing workflow and naming convention;
• do not introduce new modules unless necessary.

When the user asks whether something is a contribution:

• return to the unified core Q&A;
• evaluate through Significancy, Theorization, and Generalization;
• check evidence and feasibility;
• identify whether it is core novelty, supporting novelty, implementation detail, or not a contribution.

When the user asks to write manuscript text:

• check Paper-Ready Package and Claim-Evidence Audit;
• if writing into LaTeX, inspect the provided template;
• generate a section draft plan before drafting unless the user explicitly asks for direct drafting;
• keep claims evidence-constrained.

---

Minimal Invocation Examples

Start a project

User request:

I have an idea and several anchor papers. Help me start the research project.

Use:

workflows/project-intake.md
templates/project-card.md

Output:

reports/00_project-card.md

---

Clarify contribution

User request:

Help me determine what the core contribution of this project is.

Use:

workflows/contribution-design.md
templates/contribution-design.md
references/contribution-rubric.md

Output:

reports/02_contribution-design.md

---

Decide whether to continue

User request:

Is this idea worth pushing further?

Use:

workflows/go-no-go-evaluation.md
templates/go-no-go-report.md

Output:

reports/03_go-no-go-report.md

---

Define method

User request:

Help me specify the proposed method before implementation.

Use:

workflows/method-specification.md
templates/method-spec.md

Output:

reports/06_method-spec.md

---

Design experiments

User request:

Help me design numerical and dataset experiments.

Use:

workflows/experiment-design.md
templates/experiment-plan.md

Output:

reports/07_experiment-plan.md

---

Interpret results

User request:

Here are the experiment results. What do they mean?

Use:

workflows/result-interpretation.md
templates/result-interpretation-report.md

Output:

reports/08_result-interpretation-report.md

---

Audit claims

User request:

Do the results support our claims?

Use:

workflows/claim-evidence-audit.md
templates/claim-evidence-audit.md

Output:

reports/09_claim-evidence-audit.md

---

Prepare for writing

User request:

Turn this into a paper-ready package.

Use:

workflows/paper-ready-packaging.md
templates/paper-ready-package.md

Output:

reports/10_paper-ready-package.md

---

Analyze anchor writing style

User request:

Analyze this anchor paper's writing style for our manuscript.

Use:

workflows/anchor-writing-style-analysis.md
templates/anchor-writing-style-note.md

Output:

reports/11_anchor-writing-style-note.md

---

Plan LaTeX manuscript

User request:

I provided a LaTeX template. Plan how to write our paper into it.

Use:

workflows/latex-manuscript-writing.md
templates/latex-manuscript-plan.md
references/manuscript-section-rules.md
references/manuscript-section-blueprints.md
references/latex-writing-rules.md

Output:

reports/12_latex-manuscript-plan.md

---

Plan a section

User request:

Plan the Introduction before writing it.

Use:

templates/section-draft-plan.md
references/manuscript-section-rules.md
references/manuscript-section-blueprints.md
reports/11_anchor-writing-style-note.md
reports/12_latex-manuscript-plan.md

Output:

reports/section-plans/introduction-plan.md

---

Final Rule

At every stage, the assistant must ask:

• What is the unified core Q&A?
• Which claim is being developed?
• What evidence supports it?
• What is still pending?
• What should be downgraded, removed, or handed to MineProof?
• What should be done next?

If a proposed action does not support the core Q&A, does not improve evidence, and does not move the project toward an auditable manuscript, it should not be treated as a priority.