Data & sources

Sector data and issuer/platform mappings come from rwa.xyz; protocol TVL from DeFiLlama; per-product yields from issuer feeds (Ondo, Apollo, etc.) when available, with a fall-back to issuer-published / configured rates when a live feed isn't connected; the risk-free curve from FRED. The on-chain NAV/price oracle board on the Market page reads Chainlink SmartData feed contracts directly on Ethereum mainnet via a public JSON-RPC endpoint — no API, no key; per-feed freshness is the oracle's own last on-chain update timestamp. The RWA tokens list on the Market page comes from CoinGecko's free, keyless coins/markets endpoint (category: real-world-assets-rwa) and mixes governance tokens with asset-backed tokens. All upstream sources are free/public. Where a metric truly cannot be computed we render an em-dash (—); where a yield is a configured fall-back (not a live feed) we surface it with a tilde (~) and the label indicative in the Holdings table.

Portfolio health score

A 0–100 composite computed by compute_portfolio_health_score(ai_agent.py:1796): Sharpe ratio contribution (0–30 pts; Sharpe 0 → 0 pts, Sharpe 1.5 → 30 pts, capped), diversification via the Herfindahl index across the book's holdings (0–25 pts; 25 pts = perfectly diversified, 0 = a single asset), liquidity(0–25 pts; weighted-average liquidity tier), and regulatory standing (0–20 pts; weighted-average regulatory score). It is a scanning aid, not investment advice, and is intentionally conservative — a high score never implies safety.

Risk metrics (Sharpe · Sortino · VaR · CVaR)

Sharpe = (annualised return − risk-free) ÷ annualised volatility. The risk-free leg is the live FRED 3-month T-bill. Volatility is rating-derived: each asset's duration/credit/liquidity ratings map to an annualised vol % (a 1–10 score → vol mapping), and every live metrics call reports vol_source: "rating-derived". A λ = 0.94 EWMA realised-vol estimator over a per-asset return panel is implemented in the codebase but has no callers today — volatility is not currently computed from realized prices. Sortino substitutes downside deviation against MAR = risk-free, using the same rating-derived volatility.

VaR / CVaR: VaR is Cornish-Fisher modified — the Gaussian α-quantile is adjusted for the portfolio's skew and excess kurtosis, then scaled by the same rating-derived volatility (a relative tail-dispersion measure; the drift term is intentionally excluded so a high-carry book doesn't collapse its VaR to zero). CVaR (Expected Shortfall) applies the closed-form Gaussian-ES multipliers to VaR — ×1.254 at 95% and ×1.146 at 99% (the ratio φ(z_α)/α ÷ z_α).

Monte-Carlo path engine: forward paths are simulated with geometric Brownian motion plus a Merton jump-diffusion overlay (Poisson-timed jumps) and Student-t(4)innovations, so the fan captures fat tails and discrete shocks rather than assuming log-normality. Per-tier drift and volatility are calibrated from the same rating-derived inputs above.

Read these in context: tokenized-treasury RWAs are front-of-curve, low-volatility instruments, so their Sharpe is mechanically high — a large Sharpe here reflects low variance, not a large edge. Don't compare it to an equity Sharpe.

Portfolio concentration (Herfindahl index)

Concentration is captured via the Herfindahl–Hirschman Index (HHI = Σ(weightᵢ²)), computed as the diversification component of the Portfolio Health Score (compute_portfolio_health_score, ai_agent.py:1825) across the book's holdings/asset-class weights — not per issuer. There is currently no separate issuer-level HHI calculation and no automatic 10%/20% single-issuer concentration flag in the codebase. Lower HHI = more diversified.

Forward simulation (Monte-Carlo)

Forward outcomes are N simulated paths (default 2,000) using historical-style volatility with jump risk, holding the current tier allocation fixed. P5/P50/P95 bands are the 5th/50th/95th percentile terminal values. What drives the fan width:the fixed-income core (tokenized treasuries) is low-variance, so the visible spread is dominated by the portfolio's volatile sleeves (private credit, real estate, commodities) — not the T-bill core. Treat the fan as illustrative planning, not a forecast. This is hypothetical performance under SEC Marketing Rule 206(4)-1; it does not reflect actual trading and is not a guarantee.

Disclaimers

VYLOS Ground is a research/demo application using a fictional illustrative portfolio. Nothing here is investment, legal, or tax advice. Figures are for evaluation only.