table-schema_skill

Overview

This skill designs evidence-first, two-layer table schemas for surveys so tables are answerable and verifiable before typesetting. It produces both internal index tables for coverage/debugging and reader-facing Appendix tables suitable for publication. The schema maps every column to concrete upstream evidence fields and enforces short, non-paragraph cells.

How this skill works

I inspect the outline, subsection briefs, and evidence drafts (plus an optional GOAL) to choose table questions, row units, and columns that can be filled without guessing. For each table I specify the question it answers, the row unit, column definitions and cell-style constraints, and an explicit evidence mapping to upstream fields. I output a two-layer schema: at least two index tables and two Appendix tables, each with publishable captions and fillability guarantees.

When to use it

• After evidence packs exist and you need verifiable tables before LaTeX (Stage C4).
• When you want reader-focused tables, not pipeline/internal logs.
• To ensure every table column can be filled from existing evidence without invention.
• When the paper outline and subsection briefs are stable and you can ground table axes.

Best practices

• Make each table answer one clear reader question; avoid mega-tables.
• Map every column to a specific upstream field (no TODOs or placeholders).
• Keep cell style to short phrases or controlled tokens; avoid paragraph cells.
• Separate two layers: exhaustive internal index vs. concise, publishable Appendix.
• Prefer small, fillable tables over broad, unfillable ones.

Example use cases

• Create an index table mapping H3 subsections to axes and representative works for internal evidence scans.
• Design an Appendix method map: one row per work with core idea, interface assumptions, and key refs.
• Produce an Appendix benchmark map listing task+metric, protocol constraints, and canonical refs.
• Validate that a planned Appendix column (e.g., 'key protocol constraint') is actually present in evidence_drafts before committing.