How the system works, what data is fresh, and what is configurable

Regime Signal is a revision-aware weekly intelligence framework that helps investors detect where automation intensity is rising across labor clusters, then translate that neutral signal through capture profile into likely tailwinds, headwinds, or mixed outcomes. It is a directional decision-support system, not a causal engine or a standalone trading model.

Traditional labor data is broad, lagged, and often noisy. Regime Signal exists to give institutional users a faster, more structured way to interpret whether automation intensity is broadening, narrowing, or shifting across roles and sectors before consensus margins, staffing assumptions, and sell-side narratives fully adjust.

Current portfolio mapping universe includes 520 public tickers, 756 exposure occupations, and 342 BLS occupation profiles. Portfolio v5 layers curated company attributes, workforce relevance weights, capture profiles, software-enabler displacement risk, and efficiency confirmation for selected names. SEC disclosure scoring remains on its separate 517-name backend universe until that layer is expanded. Immutable weekly vintages are now written for point-in-time review and future backtests. Unemployment confirmation, SEC filings, and occupation structure support interpretation, but do not all enter the top-line composite directly.

1. Smooth high-frequency labor demand signals.
2. Standardize them relative to trailing baseline context.
3. Aggregate them into a composite regime score.
4. Translate the score into Stable / Watch / Warning / Alarm states.

s_t = smooth(x_t)
z_t = (s_t - mean_t) / stdev_t
C_t = Σ w_i z_i / Σ w_i

Confidence is reported on a 0–100 scale and reflects data freshness, coverage quality, and cross-signal agreement. Uncertainty bands show likely ranges around the current read rather than pretending the system is more precise than it is.

Each cell score blends overall conditions, weekly change, AI exposure, spread/breadth, role sensitivity, and industry sensitivity. Intensity tiers break at 45 / 60 / 75 and each cell carries a coverage label so users can distinguish stronger support from thinner support.

Lead-lag probabilities are evaluated across 30 / 60 / 90 / 180 day horizons using multiple signals, including top-line labor-demand intensification, divergence, exposure persistence, and corroborating evidence quality.

Divergence compares hiring demand against employment levels. When postings weaken while payrolls stay firmer, the model can flag a latent margin signal before broader labor confirmation catches up. Labor confirmation then checks whether unemployment and underemployment momentum are beginning to agree with the weaker hiring picture.

UNRATE_3m = UNRATE_t - UNRATE_t-3
UNRATE_6m = UNRATE_t - UNRATE_t-6
U6_3m = U6_t - U6_t-3
U6_6m = U6_t - U6_t-6

SEC disclosure intensity ingests 8-K, 10-Q, and 10-K filings and scores AI/labor language using frequency, recency, and context weighting. It is a corroboration layer that adds company-level evidence to the broader macro and cluster reads.

Exposure matching uses exact SOC matches first, then broader SOC prefix matches, then title-based fallback when needed.

Tickers map to clusters through a direct mapping hierarchy first, then industry fallback, then an explicit unmapped state. Portfolio v3 adds a company-attributes layer so business model and labor intensity can modify the blended signal instead of treating every same-cluster company as identical. Portfolio v4 adds curated workforce relevance weights, which rebalance the contributor mix and blended score when company-specific labor relevance appears stronger or weaker than the broad proxy mapping alone. Portfolio v5 adds capture profiles, separate mapping and capture confidence, software-enabler displacement risk, and efficiency confirmation for selected names. Capture profiles currently resolve names into labor buyers, labor sellers, software enablers, hardware enablers, or mixed / unclear cases.

The weekly read surfaces regime state, weekly deltas, capture-aware translations, early signals, action guidance, portfolio mapping, data quality context, and research-note export support. It is designed to support analyst workflow, not replace analyst judgment.