Objective
Explore the distinct RecessionAlert NBER co-incident inventory row from checked-in offline sources without live requests, credentials, private data, raw source republication, or SQLite mutation.
Source and provenance
Page https://recessionalert.com/the-nber-recession-model-project/ is bound to SHA256 sha256:a0a711709c28e199e85e0a6d79b9cafc44aafeef39d4df6ec183e3bf5dc8a682. Twelve chart GIF embeds are uncrawled (linked_assets=0); revision/GDP-GDI companions are pointers, not coefficients. Only derived monthly signals and metrics are published.
Approach and results
Eighteen labeled variants across five families used a monthly SPY/IEF overlay on INDPRO/CPIAUCSL/UNRATE/RRSFS (1992-01..2026-06, 402 months). Primary claims require lag_bars >= 1; lag 0 is control-only. Documented-intent syndrome lag-1 (Sharpe 0.743) trailed SPY buy-hold (0.769); composite lag-1 (0.866) and best neighbor (0.900) cut drawdown. Native NBER parity remains NO-GO.
Disposition
exploratory_tested_proxy — Other/Research publication only. NO-GO for native NBER co-incident fidelity or promotion to a playable Indicator because the source omits the Moore smoothing formula, deflator series pins, Probit coefficients, ALFRED vintage pins, and asset SHA256.