Replicate RecessionAlert indicator: A new monthly leading US index

Objective

Independently replicate the distinct RecessionAlert inventory row A new monthly leading US index | RecessionAlert from the checked-in canonical snapshot without a live-site request, credentials, private account data, or canonical database mutation. Because the source omits observations and an executable market rule, the required best-effort path evaluates clearly labeled public proxies separately from native source fidelity.

Result

The SHA-verified March 3, 2016 article says the index uses monthly time-series context and names 23 labor, housing, yield-curve, money-supply, market, credit, claims, state, sales, manufacturing, and delinquency components. It describes four recession-probability interpretations, a preferred average of the highest two probabilities, and separate Long-leading and Short-leading views from 21 components.

Those disclosures are narrative context, not a complete executable signal. The source provides no source-linked observations, transformations, weights, probability calibration, numeric threshold, state transition, release clock, missing-input policy, buy/sell/hold mapping, instrument, sizing, rebalance, holding, cost, exit, or re-entry rule. The 23-component list versus later 21-component reference is retained as an explicit gap. Native disposition remains insufficient_evidence; no official RecessionAlert indicator is registered or promoted.

As a separate exploratory exercise, nine inferred variants were run with FRED INDPRO, UNRATE, and T10Y3MM public macro proxies and descriptive SPY risk-on versus IEF/BIL defensive sleeves. Lookbacks 6/12/21, threshold neighbors, and execution lags 0/1/2 are recorded in the trial ledger over the 2007-05 through 2026-06 BIL-limited common window. Negative and underperforming outcomes remain preserved, and these aggregate proxies do not recover the source's 23-component identity.

Provenance and native span

The canonical input is the page body selected by research/artifacts/recessionalert/manifest.json, research/findings/recessionalert_inventory.md, and read-only data/recessionalert.sqlite. Its full SHA256 is sha256:deedcf22d1d280248542375b8093cf84c98462b27f8eb4f92d09c0d5e1c8b246; it is 85,937 bytes and was fetched at 2026-08-22T23:45:29Z. The crawl log preserves an earlier same-URL body with digest sha256:db385905b85b3f1b684da94e119fd90bb6e3a914aa7b831fabc482ec5c93777d.

The native span is one dated HTML article snapshot whose narrative reaches January 2016. The approximately 3,824-day interval from the article date to the archival fetch is snapshot delay, not economic-data publication lag. The underlying publication lag, recurring release clock, timestamp, timezone, synchronized cutoff, vintage, revision policy, and point-in-time availability are unknown.

Causal execution assumptions

The causal proxy baseline uses one available monthly bar of execution lag; lag two is a conservative sensitivity. Lag zero is retained only as a same-close research-only timing sensitivity and is not causal evidence. Warm-up and missing proxy inputs remain unavailable rather than being treated as zero, neutral, false, forward-filled, or imputed. Any future native overlay requires verified point-in-time availability and at least one available-bar execution lag.

Evidence and publication

• research/campaigns/recessionalert-new-monthly-leading-us-index/campaign.json records source claims, proxy provenance, native span, unknown lag, redistribution boundaries, nine variants, diagnostics, synthesis, and fidelity-only decision.
• research/campaigns/recessionalert-new-monthly-leading-us-index/state-space.json records the nine executed variants and their preserved ledger outcomes; diagnostics and lab-series artifacts retain timing, regime, event, and combination results.
• research/research_campaign/recessionalert_1344.py is the reproducible read-only campaign runner, and tests/test_recessionalert_1344_campaign.py provides focused offline coverage.
• docs/research/recessionalert-new-monthly-leading-us-index.html and its payload provide the public Other / Research visual proxy page with explicit inference labels and limitations. The executed notebook is research/experiments/1344.ipynb, converted to docs/experiments/1344-analysis.html.
• The Results/Changelog note and detail page are generated by python3 -m research.reports.results; python3 research/scripts/check_static_reports.py docs validates the static bundle.

Validation includes the complete focused campaign test module, the offline campaign runner, result generation, and static-reports check. No live-site request, credential access, tracked-crawl write, or market-cache mutation was used.

Access and limitations

The supported user path is the discoverable Other → Research page, where the static article explains the source-fidelity boundary and the Research Lab exposes the tested proxy variants. The local executable path is python3 -m research.research_campaign.recessionalert_1344; its outputs are exploratory research, not trading advice or an official catalog signal.

The campaign cannot validate the source's claimed index or promote a signal until a separately authorized, redistributable, source-linked monthly panel or chart supplies observations, transformations, component membership, probability equations, thresholds, release timing, revisions, and execution semantics. Keep all inferred variants labeled as proxies and do not present their metrics as native RecessionAlert fidelity.