find-data

name	find-data
description	Data discovery and assessment dispatching the explorer agent (finder) and data-quality-surveyor agent (critic). Searches public, administrative, survey, and novel data sources. Scores feasibility and measurement quality.
disable-model-invocation	true
argument-hint	[research question or data requirements description]
allowed-tools	["Read","Grep","Glob","Write","WebSearch","WebFetch","Task"]

Find Data

Discover and assess datasets by dispatching the Explorer (data finder) and data-quality-surveyor (data critic).

Input: $ARGUMENTS — research question, variable requirements, or description of needed data.

---

Workflow

Step 1: Context Gathering

1. Read research spec if it exists (quality_reports/research_spec_*.md)
2. Read strategy memo if it exists (quality_reports/strategy_memo_*.md)
3. Read .claude/rules/domain-profile.md for common data sources in the field
4. Understand what variables are needed: treatment, outcome, controls, time period, geography

Step 2: Launch Explorer Agent

Delegate to the explorer agent via Task tool:

Prompt: Find datasets for "[research question/requirements]".
Search across source categories:
  1. Public microdata (CPS, ACS, NHIS, MEPS, etc.)
  2. Administrative data (Medicare claims, tax records, court records)
  3. Survey data (RAND HRS, PSID, Add Health, NLSY)
  4. International (World Bank, OECD, Eurostat)
  5. Novel/alternative (satellite imagery, web scraping, proprietary)
For each dataset found:
  - Name, provider, access level (public/restricted)
  - Key variables available
  - Coverage (time period, geography, sample size)
  - Feasibility grade: A (ready to use), B (accessible with effort),
    C (restricted but obtainable), D (very difficult)
  - Strengths and limitations
Save to quality_reports/data_exploration_[topic].md

Step 3: Launch data-quality-surveyor Agent (Data Critic)

After Explorer returns, delegate to the data-quality-surveyor agent:

Prompt: Review the data assessment at quality_reports/data_exploration_[topic].md.
For each proposed dataset, check:
  1. Measurement validity — does the variable actually measure what we need?
  2. Sample selection — who's in the data? Who's missing?
  3. External validity — can we generalize from this sample?
  4. Identification compatibility — does this data support the proposed design?
  5. Known issues — documented problems with this dataset in the literature
Score each dataset. Flag deal-breakers.
Save critique to quality_reports/data_critique_[topic].md

Step 4: Synthesize Recommendations

After both agents return, present:

# Data Assessment: [Topic]
**Date:** [YYYY-MM-DD]

## Recommended Datasets (ranked)

### 1. [Dataset Name] — Feasibility: [A/B/C/D]
- **Why:** [Best match for research question]
- **Variables:** [Key variables available]
- **Concerns:** [data-quality-surveyor's critique]
- **Access:** [How to obtain]

### 2. [Dataset Name] — Feasibility: [A/B/C/D]
...

## Rejected Datasets
| Dataset | Reason for Rejection |
|---------|---------------------|
| [Name] | [data-quality-surveyor's deal-breaker] |

## Data Gaps
[Variables needed but not found in any dataset]

## Next Steps
[Concrete actions: apply for access, download, contact provider]

---

Principles

• Explorer finds, data-quality-surveyor critiques. Never skip the critique step.
• Feasibility matters. A perfect dataset you can't access is useless.
• Domain-profile aware. Check common data sources for the field first.
• Measurement validity is key. The data-quality-surveyor catches "this variable doesn't actually measure what you think."
• Be honest about limitations. Every dataset has them — document them upfront.