Initializing SQL engine...
SQL Verification Demo NightClaw ↗

Select a question to begin

Both sides are simulated to illustrate a pattern. The left side represents typical AI search behavior: retrieving data from inconsistent web sources, applying unstated methodology, producing non-reproducible results. The right side runs deterministic SQL against canonical FRED data, then uses AI only for narrative. SQL is real; LLM outputs are scripted.

AI Search

Simulated
Tokens
0
$0.000
Waiting for race to start

Deterministic + AI

Simulated
Tokens
0
$0.000
Waiting for race to start

Headline Cross-Reference

400 tokens (Tier 2 model) — $0.001

Why This Matters

AI Search Pattern
✗ Sources vary between runs
✗ Methodology unstated
✗ Results non-reproducible
✗ No verification step before output
Deterministic + AI Pattern
✓ Canonical data source (FRED)
✓ SQL methodology visible and auditable
✓ Same query, same result every time
✓ Numbers verified before narrative generated
The point is not that AI search is bad — it’s that for questions requiring precise numerical answers, the source and methodology must be explicit and reproducible. AI adds value in interpretation, context, and narrative. Deterministic code handles the math. The pattern works when each does what it’s suited for.