research-documentation

name	research-documentation
description	Unified documentation lookup and academic research skill for quant trading workflows. Covers API/library documentation retrieval, Context7 integration for versioned docs, and arXiv paper search for market microstructure, forecasting, and strategy research.
version	1.0.0
metadata	{"consolidates":["documentation-lookup-1","context7-documentation-lookup-2","arxiv-research-search"]}

Research & Documentation Lookup

Unified skill for finding, citing, and applying information from library documentation, versioned API references, and academic research papers. Tailored for a quantitative trading research system where correct API usage, up-to-date library behaviour, and awareness of recent academic work are all critical.

---

1. When to Use

Activate this skill when the task involves any of the following:

• Looking up the correct API signature, parameters, or return types for a library used in the trading stack (pandas, NumPy, scikit-learn, NautilusTrader, etc.).
• Producing a small, correct, self-contained code example that demonstrates library usage.
• Retrieving version-specific documentation via Context7 to ensure compatibility with pinned dependency versions.
• Searching arXiv for papers on market microstructure, alpha signals, portfolio optimisation, execution algorithms, risk modelling, or ML methods relevant to trading.
• Triaging and summarising academic papers to decide whether they are worth a full read.

---

2. Inputs to Gather

Input	Details
Library / API	Name, version, specific module or function of interest.
Query	What you need to know -- behaviour, edge cases, migration notes.
Context	How the result will be used (backtest, live engine, notebook).
arXiv topic	Keywords, categories (q-fin.TR, cs.LG, stat.ML), date range.
Constraints	Must match a pinned version, must work offline, etc.

---

3. API & Library Documentation Lookup

3.1 Workflow

1. Identify the library, version, and specific symbol (function, class, constant).
2. Retrieve the official documentation page or docstring.
3. Extract the signature, parameter descriptions, return type, and any caveats.
4. Produce a minimal, runnable code snippet that demonstrates correct usage.
5. Cite the documentation URL and version so the user can verify.

3.2 Best Practices

• Always confirm the version of the library installed in the project (pip show <pkg> or check pyproject.toml / requirements.txt).
• Prefer official docs over blog posts or Stack Overflow.
• When behaviour differs between versions, note the difference explicitly.
• For deprecated APIs, show the replacement and migration path.

3.3 Common Libraries in This Stack

Library	Typical Use
pandas	DataFrames, time-series indexing, resampling.
numpy	Vectorised math, array operations.
scikit-learn	Preprocessing, feature selection, model evaluation.
nautilus_trader	Backtest engine, order management, strategy base classes.
polars	High-performance DataFrames for large tick datasets.
xgboost / lightgbm / catboost	Gradient-boosted trees.
optuna	Hyperparameter optimisation.
shap	Feature importance / model explainability.
plotly / matplotlib	Visualisation.

---

4. Context7 Integration

Context7 provides versioned, structured documentation lookups that are more reliable than general web search for pinned-dependency projects.

4.1 Workflow

1. Resolve library ID: Use the Context7 resolve-library-id tool with the library name and a description of what you need. This returns a Context7-compatible library ID (format: /org/project).
2. Query documentation: Use the Context7 query-docs tool with the resolved library ID and a specific question. Be detailed -- e.g., "How to set up purged k-fold cross-validation in scikit-learn" rather than just "cross validation".
3. Apply: Integrate the retrieved code snippets and explanations into the current task, verifying that they match the pinned version.

4.2 Tips

• Select libraries based on name match, source reputation (High/Medium), snippet coverage, and benchmark score.
• If a specific version is pinned, include the version in the library ID (/org/project/version) for version-accurate results.
• Limit to 3 Context7 calls per question to avoid excessive API usage.

---

5. Academic Paper Search (arXiv)

5.1 Workflow

1. Define the research question: what phenomenon, method, or market behaviour are you investigating?
2. Construct an arXiv search query using relevant keywords and category filters:
   • q-fin.TR -- Trading and market microstructure.
   • q-fin.PM -- Portfolio management.
   • q-fin.RM -- Risk management.
   • q-fin.ST -- Statistical finance.
   • cs.LG -- Machine learning.
   • stat.ML -- Statistics and machine learning.
3. Retrieve candidate papers (title, authors, abstract, date).
4. Triage each paper:
   • Is the method applicable to the current trading context?
   • Is the data regime comparable (equities vs. crypto, HFT vs. daily)?
   • Are the results reproducible with available data?
5. Summarise the top 3-5 papers with a one-paragraph assessment each.
6. Cite with full arXiv IDs (e.g., arXiv:2301.12345) for traceability.

5.2 Search Tips

• Combine domain terms with method terms: "market microstructure" AND "transformer" or "pairs trading" AND "reinforcement learning".
• Filter by date range to focus on recent work (last 2-3 years).
• Check the "References" section of a relevant paper to find related work that the search may have missed.

5.3 Triage Template

For each candidate paper, produce:

Title:    <paper title>
arXiv ID: <arXiv:YYMM.NNNNN>
Authors:  <first author et al.>
Date:     <submission date>
Relevance: [High / Medium / Low]
Summary:  <2-3 sentences on method and key finding>
Applicability: <How it maps to our trading system; data requirements>
Action:   [Read in full / Skim methods section / Skip]

---

6. Outputs

For any documentation or research task, deliver:

• Concrete answer: the API signature, code snippet, or paper summary.
• Source citation: URL, version, or arXiv ID.
• Integration sketch: how to use the finding in the current codebase (files to change, imports to add, tests to run).
• Risk notes: version incompatibilities, deprecated APIs, unreproducible paper results, or known bugs.

---

7. Workflow Summary

1. Restate the information need in precise terms.
2. Choose channel: library docs, Context7 versioned lookup, or arXiv search.
3. Retrieve the information using the appropriate tool.
4. Verify against the project's pinned versions and constraints.
5. Deliver a cited, actionable answer with a code snippet or summary.
6. Log the lookup in the research journal if it informs a strategy decision.