explain-paper

name	explain-paper
description	Explain the paper, methods, factors, tables, figures, and code paths for this repo. Use when Claude needs to answer a paper question, trace an exhibit back to code, or connect saved results to equations and manuscript claims.
argument-hint

Explain Paper

Read only the docs needed for the question:

• AGENTS.md
• docs/agent-context/exhibits/README.md
• docs/agent-context/exhibits/
• docs/manifests/exhibits.csv
• docs/manifests/manuscript_exhibits.csv
• docs/manifests/paper_claims.csv
• docs/acceptance/prompt_harness.csv
• .claude/paper-context.md
• docs/paper/co-pricing-factor-zoo.ai-optimized.md
• docs/agent-context/paper-reading-guide.md
• docs/agent-context/paper-method.md
• docs/agent-context/paper-results-main.md
• docs/agent-context/paper-results-ia.md
• docs/agent-context/tables-guide.md
• docs/agent-context/figures-guide.md
• docs/agent-context/factor-interpretation.md
• docs/agent-context/factors-reference.md
• docs/agent-context/factors/
• docs/agent-context/noisy-proxy-guide.md
• docs/agent-context/treasury-component-guide.md
• docs/agent-context/time-varying-guide.md
• docs/agent-context/ia-robustness-guide.md

Use When

• the user asks what the paper finds or how the method works
• the user wants a table, figure, factor, or equation traced back to code
• the user wants to understand which saved objects feed a result
• the user wants to compare manuscript claims with executable repo coverage

Do Not Use When

• the task is to set up dependencies or diagnose the runtime environment
• the task is to execute the replication pipeline rather than explain it
• the question is a general coding task unrelated to the paper or outputs

Inputs

• the paper question, exhibit id, factor name, or code path of interest
• optional request for high-level intuition versus exact equation-level tracing

Outputs

• concise explanation of the requested method or result
• exact exhibit-to-code path when relevant
• saved-object names or generated output paths when relevant
• explicit note when manuscript prose and executable coverage diverge

Workflow

1. Start from the empirical question or exhibit, not from sampler internals.
2. Use the relevant file in docs/agent-context/exhibits/ first when a dossier exists for that exhibit.
3. Use docs/manifests/exhibits.csv to locate the entry script and helper before reading source deeply.
4. Use docs/manifests/manuscript_exhibits.csv for repo coverage questions and docs/manifests/paper_claims.csv for headline manuscript claims.
5. Load the full paper only when exact equation, appendix, table, or figure references matter.
6. Prefer stored saved-object metadata such as nontraded_names, bond_names, and stock_names over reconstruction.
7. When the question matches a canonical explanation prompt, use docs/acceptance/prompt_harness.csv as the minimum answer contract before you elaborate.

Example Prompts

• "Fully explain Table 1."
• "Explain Table 5 and show the code path."
• "Why does the paper say the SDF is dense?"
• "Which equation is the treasury decomposition using?"
• "What does it mean for a factor to be included in the SDF?"
• "Is PEAD just a small-cap effect?"

Failure Boundaries

• do not invent factor classifications or unsupported model configurations
• do not treat stale prose as authoritative when code and docs disagree
• do not describe paper-only IA exhibits as executable repo outputs
• do not bulk-load the full paper when a smaller shared doc answers the question