Risk Watch

Reading stress before the filings arrive

Private-credit stress rarely arrives as one clean, timely number. Traded markets move quickly but provide proxies. Regulatory filings can reveal pressure in the vehicles themselves, but report on an earlier period. Risk Watch keeps the two layers distinct so a market warning is not mistaken for filed confirmation, or a quiet market for proof that private portfolios are healthy.

The dashboard image above is a dated capture. Live readings depend on the configured feeds and the observations they return.

Two layers of evidence

Early warning combines high-yield and CCC spreads, bank-credit indicators, listed BDCs, liquid credit, software equities, and leveraged-loan ETFs. These are market and bank proxies, not direct measurements of private portfolios.

Confirmation draws on SEC vehicle disclosures and public BDC data, N-PORT interval-fund reports, completed tender outcomes, and ICI high-yield fund flows. Reporting periods and source links stay attached to the evidence. Together, the layers distinguish an unconfirmed market warning, filed stress amid calm markets, broad stress, and insufficient evidence.

Scores with visible coverage

The dashboard shows three experimental scores across tactical, cyclical, and structural horizons: private-credit stress, credit opportunity, and AI-equity opportunity. Each score exposes its inputs, fixed weights, contributions, freshness, and evidence coverage. Below 50% coverage, the card says Insufficient evidence instead of displaying a score. Missing inputs do not inherit the weight of available ones.

An evidence timeline places traded prices, official macro releases, and SEC filings on one dated axis. A contribution waterfall reconciles the selected score, while signal rows show source cadence, stale or unavailable readings, raw thresholds, and active exceptions. Snapshots are stored when observations change and streamed to connected users.

Context beyond private credit

Separate views cover inflation and growth, Treasury and global sovereign stress, bond and equity divergence, semiconductor breadth, and rate sensitivity. An EIA energy panel shows crude and refined-product prices, same-date crack-spread proxies, inventories, refining activity, and demand proxies. Energy readings are explanatory evidence and do not enter risk scores or alerts. The optional economic forecast is a research model and also stays outside the scores.

Built in Rust

An Axum server collects provider data, assembles and scores snapshots, serves the dashboard, and sends WebSocket updates and alerts. The application and its optional offline model trainer are Rust programs. SQLite supports a local run; the Docker setup uses PostgreSQL with TimescaleDB and Redis. Provider caches, retry and circuit state, replay, structured logs, Prometheus metrics, and separate health and readiness endpoints support operation when a feed fails.

Evidence has limits

Filed NAV marks, non-accruals, fund flows, and tender outcomes remain lagged to their reporting periods. Public prices and bank series are proxies. The horizon scores remain experimental until point-in-time outcomes provide enough history to validate them.

Current-day NAV, live redemption queues, liquidity terms, borrowing availability, covenant headroom, and vehicles without adequate public filings still require internal administrator or portfolio feeds.