Eval evidence
The pipeline, validated — every weekday
I sell “an eval harness with golden examples, latency and cost monitoring.” Here is one running in the open. The macro-data pipeline doesn’t just pull numbers — a validation pass checks every series against its source for freshness, value tolerance, and release-window timing, then commits the result. Same discipline I bring to a client AI feature: explicit thresholds, fail loud, keep the receipt.
Last run May 29, 2026 at 04:54 PM UTC · evaluated month 2026-05 · latest data 2026-05. Generated and committed by a GitHub Action in this repository.
Per-series checks
| Indicator | Source | Latest | Lag | Max Δ vs source | Tolerance | Release window | Status |
|---|---|---|---|---|---|---|---|
| Unemployment Rate | FRED:UNRATE | 2026-04 | 0 mo | 0.0000 | 0.15 | PASS | PASS |
| Avg Weeks Unemployeed | FRED:UEMPMEAN | 2026-04 | 0 mo | 0.0000 | 1.5 | PASS | PASS |
| Job Openings | FRED:JTSJOL | 2026-05 | 0 mo | 0.0000 | 350 | PASS | PASS |
| Unemployed 27 weeks | FRED:UEMP27OV | 2026-04 | 0 mo | 0.0000 | 300 | PASS | PASS |
| Unemployeed Count | FRED:UNEMPLOY | 2026-04 | 0 mo | 0.0000 | 400 | PASS | PASS |
| Fed Rate | FRED:FEDFUNDS | 2026-04 | 0 mo | 0.0000 | 0.2 | PASS | PASS |
| 15 year mortgage | FRED:MORTGAGE15US | 2026-05 | 0 mo | 0.0000 | 0.15 | PASS | PASS |
| 30 year mortgage | FRED:MORTGAGE30US | 2026-05 | 0 mo | 0.0000 | 0.15 | PASS | PASS |
| S&P 500 | FRED:SP500 | 2026-05 | 0 mo | 0.0000 | 125 | PASS | PASS |
| Labor Participation Rate | FRED:CIVPART | 2026-04 | 0 mo | 0.0000 | 0.2 | PASS | PASS |
| Housing Price Index | FRED:CSUSHPINSA | 2026-05 | 0 mo | 0.0000 | 1 | PASS | PASS |
| CPI | FRED:CPIAUCSL | 2026-04 | 0 mo | 0.0000 | 1 | PASS | PASS |
| GDP | FRED:GDP | 2026-05 | 0 mo | 0.0000 | 10 | PASS | PASS |
| Stock Market (b) | FRED:NCBCEL | 2026-02 | 0 mo | 0.0000 | 2500 | PASS | PASS |
| National Debt (b) | FRED:GFDEBTN | 2026-02 | 0 mo | 0.0000 | 400 | PASS | PASS |
| 10Y Treasury | FRED:DGS10 | 2026-05 | 0 mo | 0.0000 | 0.15 | PASS | PASS |
| 2Y Treasury | FRED:DGS2 | 2026-05 | 0 mo | 0.0000 | 0.15 | PASS | PASS |
| Consumer Sentiment | FRED:UMCSENT | 2026-04 | 0 mo | 0.0000 | 5 | PASS | PASS |
| Retail Sales | FRED:RSAFS | 2026-04 | 0 mo | 0.0000 | 15000 | PASS | PASS |
| Personal Savings Rate | FRED:PSAVERT | 2026-04 | 0 mo | 0.0000 | 2 | PASS | PASS |
| PCE Price Index | FRED:PCEPI | 2026-04 | 0 mo | 0.0000 | 1 | PASS | PASS |
| Avg Hourly Earnings | FRED:CES0500000003 | 2026-04 | 0 mo | 0.0000 | 0.3 | PASS | PASS |
| Initial Claims | FRED:ICSA | 2026-05 | 0 mo | 0.0000 | 30000 | PASS | PASS |
| Housing Starts | FRED:HOUST | 2026-04 | 0 mo | 0.0000 | 100 | PASS | PASS |
| Building Permits | FRED:PERMIT | 2026-04 | 0 mo | 0.0000 | 100 | PASS | PASS |
| Existing Home Sales | FRED:EXHOSLUSM495S | 2026-04 | 0 mo | 0.0000 | 200000 | PASS | PASS |
| Industrial Production | FRED:INDPRO | 2026-04 | 0 mo | 0.0000 | 1.5 | PASS | PASS |
How to read this: each weekday the harness re-pulls every series from FRED and compares it to what’s committed. Max Δ vs source is the largest discrepancy found; it must stay under tolerance. Lag is months behind the source. Release window confirms the data arrived when the official calendar said it would. A single failure flips the overall status and surfaces here.