Semantic Search (Codex)

Search claims and sources using natural language queries. Use when looking for related claims or finding existing analysis.

Invocation

$search <query>

Note: Codex reserves /... for built-in commands. Use $search instead.

Search claims and sources using natural language queries.

Usage

rc-db search "query" --limit 10
# or: uv run python scripts/db.py search "query" --limit 10

Note: Semantic search requires embeddings. If search fails with embedding errors:

• •Ensure sentence-transformers is installed (included in realitycheck dependencies)
• •Check embedding status: rc-db embed status or uv run python scripts/embed.py status
• •Generate missing embeddings: rc-db embed generate or uv run python scripts/embed.py generate
• •Do not set REALITYCHECK_EMBED_SKIP=1 unless you explicitly want to disable embeddings (used for CI/testing)

Options

• •--domain: Filter by domain (TECH/LABOR/ECON/GOV/SOC/RESOURCE/TRANS/GEO/INST/RISK/META)
• •--limit: Maximum results to return (default: 10)
• •--format: Output format - json (default) or text
• •--type: Filter by record type (claim or source)

How It Works

1. •Convert query to embedding using sentence-transformers
2. •Find nearest neighbors in the claims/sources vector index
3. •Return ranked results with similarity scores

Output

Results include:

• •Claim/Source ID
• •Text (truncated)
• •Type, Domain, Evidence Level
• •Credence score
• •Similarity score

Examples

rc-db search "AI automation labor displacement"
rc-db search "training costs" --domain TECH --limit 5
rc-db search "economic transition" --type claim

Related Skills

• •$stats
• •$validate