Objective
Independently replicate the distinct RecessionAlert row A U.S Weekly Leading Economic Index | RecessionAlert from the checked-in canonical snapshot, without a live-site request, credentials, private account data, or mutation of data/recessionalert.sqlite.
Approach
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:e81b7ab63069f9cbc94b4a57a526b55b69c32580e11801b1b2a09ba8f806f9fe; it is 86,958 bytes and was fetched at 2026-08-22T23:45:27Z. The article is dated 2016-02-23T16:01:55-0500. The 3833.1136-day article-to-snapshot interval is archival delay, not WLEI publication or release lag. The crawl log preserves two same-URL page-body vintages; the later digest is manifest-selected.
Disclosed rule and interpretation
The article describes a U.S. Weekly Leading Economic Index (WLEI) built from more than 20 weekly time series across six broad categories: Corporate Bond Market Composite, Treasury Bond Market Composite, Stock Market Composite, Labor Market Composite, Credit Market Composite, and mortgage-backed securities. It calls WLEI a weekly growth index, says it is weighted heavily toward financial-market data, and compares WLEI growth with ECRI WLIg. The article says the current format had run in-sample for seven years with subscribers and favors medium- to short-leading weekly inputs rather than monthly or long-leading inputs.
The source states that WLEI data have at most a one-week lag and are published Thursday afternoons, reportedly 18 hours before the ECRI Weekly Leading Index. It gives no component membership, formula, weights, normalization, lookback, threshold, state machine, missingness handling, or buy/sell/hold mapping. This is an economic indicator concept, not a disclosed playable portfolio signal. Direction for market use is undisclosed.
Native span, publication lag, and causal assumptions
The native evidence is one dated HTML article. Its seven-year subscriber in-sample statement is not a pinned observation series, and the target page has zero canonical asset rows. Six XLSX assets exist on the separate excel-data page but are not source-linked to this row and are not used. Native observation span and effective evaluation span are therefore undisclosed/not computable.
The source timing claims are at most one-week data lag and Thursday-afternoon publication. Exact recurring release timestamp, timezone, information cutoff, vintage, revision schedule, and point-in-time availability are unknown. The article-to-crawl interval is not release lag. Unknown dates and values remain unavailable rather than zero, neutral, false, stale-valid, or imputed. Any future overlay requires documented point-in-time input availability and at least one available-bar execution lag; same-close/lookahead is unsupported.
Validation and result
The focused offline suites tests/test_recessionalert_weekly_leading_economic_index.py and tests/test_recessionalert_1348_campaign.py pass 13 tests. They verify manifest/inventory/SQLite digest agreement, selected crawl history, disclosed component and timing text, zero target-page assets, read-only database access, unavailable-state preservation, source-labeled executable SignalSpecs, imperative-vs-DSL parity, nearby family/lookback/threshold variants, one-bar causal shifting, and same-close sensitivity labeling. The executed notebook is research/experiments/1348.ipynb; its offline HTML is generated at docs/experiments/1348-analysis.html by the canonical notebooks producer.
Because native WLEI observations and executable source semantics remain unavailable, the mandatory best-effort route used checked-in FRED INDPRO, UNRATE, GS10, BAA10Y and Yahoo SPY/IEF snapshots. Five labeled variants were run over the longest defensible weekly display window (1993-01-29/2026-06-05), with the common SPY/IEF overlay beginning 2002-08-09: weighted-growth v1, weighted-growth v2, diffusion v3, acceleration v4, and an explicitly non-causal same-close sensitivity. The v3 diffusion trial is retained as an all-risk-off failure; v4 is an interesting but speculative candidate; same-close is timing sensitivity only. These are descriptive proxy outputs, not native WLEI performance.
Native decision: not_reproducible; exploratory proxy disposition: publish_for_research. Publish the source-backed concept, provenance, timing metadata, causal policy, explicit gaps, and labeled proxy charts/results. Do not register a playable native Other-indicators signal, claim proxy/native equivalence, or present the same-close sensitivity as causal evidence.
Explicit gaps
No source-linked machine-readable WLEI component/composite observations or native span; no exact component formula, weights, transformations, normalization, growth/lookback window, initialization, score range, or threshold; no smoothing, equal-value, stale/missing/revised-input, or unavailable-state behavior; no recurring point-in-time release metadata, timezone, vintage, or revision policy; no buy/sell/hold or risk-on/risk-off mapping, target instrument, allocation, rebalance, holding period, costs, slippage, exit, or re-entry; no independent recession/market outcome window; and no captured chart/image bytes or source-linked redistributable data file.
Web plan and boundary
Web plan:
- canonical section: Other / Research via the experiments index;
- discoverable path: the generated campaign article and notebook analysis are built by
research.reportsand linked from existing research pages; - browser evidence: the generated source/provenance panel, selectable proxy variants, signal/equity charts, and disposition page are the appropriate observable surface;
- static/local boundary: this is an offline static exploratory proxy report, not a live WLEI evaluator; proxy overlays are research-only;
- exception: a native playable indicator workflow remains misleading and unsafe because the source supplies no source-linked series, thresholds, or action mapping. Reconsider only after a separately authorized refresh provides source-linked observations and executable semantics.
Public output excludes cached HTML, chart/image bytes, workbook bytes, credentials, cookies, private account data, and unverified workbook contents.
See research/findings/1348.md, research/findings/specs/recessionalert-weekly-leading-economic-index.json, research/experiments/1348.json, and research/experiments/1348.ipynb for the complete evidence record.