Credibility is the product.

Integrity in practice

These aren't aspirations pinned to a wall — they're implemented behaviors. Compet 2026 suppresses any score that falls below its confidence threshold rather than publish bad data when a vendor feed fails. The Global Exchange Planner reports its own negative finding — a naïve baseline beat every forecasting model, so it recommends a risk-managed split instead of fabricating a timing edge. EconOS marks stale series instead of hiding them.

The discipline comes from the day job. Nine years of federal compliance — 2 CFR Part 200, subrecipient monitoring, procurement review — is work where a wrong number has real consequences. Software built by someone who signs off on drawdowns treats data integrity as a professional obligation, not a feature.

Working principles

• Only credible public data: federal statistical agencies, city open-data programs, and documented research datasets.
• Licenses honored: non-commercial and attribution terms are flagged in each repository and respected in each product.
• Computed vs. proposed: calculated outputs stay separate from planned enhancements. No fabricated metrics or impact.
• Honest limits: data-sparse geographies, modeled estimates, and baseline comparisons are surfaced, not smoothed over.
• Reproducible where repo-backed: portfolio pipelines rebuild with one command and validate output against schema gates.
• No personal data: products use aggregate public statistics; nothing identifies an individual.