Objective
Independently replicate the distinct RecessionAlert inventory row for Generation-2 Market Probability Models from the checked-in data/recessionalert.sqlite snapshot without live-site requests, credentials, private account data, or canonical crawl-database mutation.
Source and provenance
Page https://recessionalert.com/gen2-prob-models/ bound to SHA256 sha256:9ec872ee30f34cf39a5b6e390a6363866e36667a64580ea35afcbdefe4ed2f46, fetched 2026-08-22T23:48:10Z, published 2022-09-12. Public proxies loaded read-only from data/market.sqlite.
Disclosed rule, native span, and publication lag
Documented fragments include Zig-Zag x=5/x=17 OHLC move math since 2000, rank-based probabilities, 60/80/90% thresholds, twelve-point diffusion score, and explicit non-actionable disclaimer. No native probability observations are pinned. Publication lag is undisclosed. Causal default is lag>=1 on daily closes; lag-zero is control_only.
Approach and variants
Primitives in research/research_campaign/recessionalert_1412.py. Eight default variants plus sixteen grid neighbors (24 trials) across lag, threshold, diffusion, and x-parameter grids on ^GSPC 2000+ training window with optional SPY/IEF overlays.
Results (exploratory)
Peak-probability SPY/IEF overlay at 90% threshold shows the best tested Sharpe among causal variants on the proxy reconstruction; documented diffusion>=6 baseline is sparse but aligns with the application note's six-event emphasis. Disposition: exploratory_tested_proxy_no_playable_indicator — source explicitly rejects actionable trading use.