Objective
Independently replicate the distinct RecessionAlert row MODELS: A Recession Forecasting Diffusion (RFD) | RecessionAlert from the checked-in canonical snapshot without a live-site request, credentials, private account data, or canonical crawl-database mutation. Because the native series and execution semantics are incomplete, this result records the source-fidelity boundary and the required labeled best-effort proxy campaign separately.
Approach and provenance
The canonical input is the manifest-selected page body in data/recessionalert.sqlite, cross-checked against research/findings/recessionalert_inventory.md and research/artifacts/recessionalert/manifest.json. The page SHA256 is sha256:e705e5996621202e0e72744fdf229c4a15a8d85d1de3eff8427c7ac76bd99066; it is 91,931 bytes and was fetched at 2026-08-22T23:48:08Z. The article timestamp is 2022-03-18T11:27:57-0400. The 1,618-day article-to-snapshot interval is archival delay, not an RFD publication or economic-data release lag. The crawl log preserves two same-URL page-body vintages; the later body is manifest-selected. The target page has no captured chart/GIF asset rows or subscriber workbook bytes.
The exploratory runner reads tracked public market data through read-only SQLite access. The selected macro proxy chain is FRED INDPRO, UNRATE, DGS10, DGS3MO, and HOUST; the common macro overlap is 1981-09 through 2026-06. Yahoo SPY and IEF provide the primary overlay window, with BIL as a secondary defensive sensitivity. Proxy identifiers, spans, transformations, and redistribution limits are recorded in the campaign provenance and lab sidecar.
Disclosed rule and causal boundary
The article says fifteen diversified quantitative recession models are combined into a diffusion counting how many are in their respective recession territories. RFD ranges from 0 to 15. The named components span NBER Big-Four, GDPI, SuperIndex coincident/leading/diffusion and recession-syndrome models, Anxious Index, labor and housing composites, CMHI, Stylized Headwinds, WLEI, USMLEI, USHMI, RAVI, and USLONG.
The source configures an “optimum actionable trigger” above five, meaning at least six component models in recession. It claims approximately six months of lead time, approximately two months variability, a 4.5–7.5-month Golden Lead Zone, and probability mappings at 6/12/18/24 months. The chart is described as updating and alerting subscribers whenever RFD rises, with historical data in a Monthly PRO Excel Data File. These are source claims and monthly context, not an independently measured release clock or validated performance result.
RFD is a recession-warning concept, not a disclosed buy-only, sell-only, or dual-sided trading signal. No instrument, allocation, rebalance, holding period, cost, slippage, entry, exit, re-entry, or persistence rule is disclosed. For the inferred proxy only, the causal baseline applies one available-month publication lag followed by the engine's one-bar execution lag; same-available-period lag 0 is retained only as a research-only sensitivity.
Best-effort proxy results
The inferred public diffusion analogue derives a transparent four-state recession score from the selected macro series, scales the available count to a 0–15-like score, and tests warning thresholds around the documented > 5 trigger. Six nearby labeled variants were executed across lookbacks, thresholds, and available lags, with SPY-versus-IEF primary overlays and SPY-versus-BIL secondary sensitivities. The strongest causal primary result was the six-month smoothing / threshold-6 / one-month availability-lag variant, with SPY/IEF total return 5.1588 versus aligned SPY 11.3974. A lag-0 sensitivity returned 5.7128, while a lag-2 conservative variant returned 3.2668; lag 0 is not a causal publication assumption. All tested public proxy overlays trailed aligned SPY, and long-lookback event counts were low. These are exploratory proxy observations, not native RFD validation.
Decision
Exploratory publication under Other / Research, with no source-fidelity promotion and no official RecessionAlert indicator or strategy registration. The campaign satisfies best-effort policy with selected public proxies, an explicitly inferred rule, six executed neighboring variants, longest defensible local history, causal timing treatment, diagnostics, and retained negative outcomes. Native RFD observations, proprietary component definitions, publication timing, and portfolio semantics remain unresolved.
Files and validation
- Finding and source-fidelity evidence:
research/findings/1346.md - Task-specific inferred SignalSpec:
research/signal_spec/fixtures/recessionalert-rfd-proxy.json - Executable campaign runner:
research/research_campaign/recessionalert_1346.py - Campaign artifacts:
research/campaigns/recessionalert-rfd/ - Executed notebook and generated analysis:
research/experiments/1346.ipynb,docs/experiments/1346-analysis.html - Public campaign page and payload:
docs/research/recessionalert-rfd.html,docs/site-data/research/recessionalert-rfd.json - Public result note and generated result surfaces:
research/results/1346.md,docs/site-data/results.json,docs/results/1346.html - Focused validation:
python3 -m pytest -q tests/test_recessionalert_rfd_campaign.py tests/test_recessionalert_rfd.py - Campaign validation and reproducibility:
python3 -m research.research_campaign validate --campaign recessionalert-rfd,python3 -m research.research_campaign.recessionalert_1346 --run --output-dir research/campaigns/recessionalert-rfd - Static/public checks:
python3 research/scripts/check_static_reports.py docs,python3 -m research.artifacts check --mode full --offline
Explicit gaps
No source-linked RFD observations, component histories, captured Monthly PRO workbook, exact component formulas or thresholds, missing-data handling, native observation span, synchronized release clock, point-in-time availability, vintage/revision policy, complete probability table, independent outcome series, buy/sell/hold mapping, instrument, allocation, holding, costs, exit, re-entry, or transition rule is available. The page’s six-month lead, probability, variability, and zero-error claims therefore remain unvalidated. The public proxy is deliberately not presented as preserving native RFD identity.
Public boundary
Public outputs contain only redacted source location, source hashes, disclosed rule summaries, proxy identifiers and derived metrics, timing metadata, causal assumptions, validation status, and gap notes. They exclude cached HTML, GIF/chart bytes, workbook bytes, authenticated or nonce-bearing URLs, credentials, cookies, private account data, and any official indicator or investment recommendation.