The questions you'd ask if you distrusted us.

Every answer carries a number and a limitation. If a question can't be answered that way, we say so.

You already prompt your IDE with these. Here are the answers, source-first, with the failure modes stated.

Plain terms

What is tail risk in plain terms?

Tail risk is the probability of an outcome far from the mean — the part of the distribution that standard models truncate. It matters because a single fat-tailed move can erase a decade of compounding; the Spitznagel framing of cost-effective hedging treats this as the geometric cost of drawdowns, where one −50% event mathematically outweighs years of +10% gains. See our tail-risk primer for the full distributional argument.

VaR vs Expected Shortfall

VaR vs Expected Shortfall — what’s the actual difference?

Value-at-Risk asks “what is my loss at the 95th percentile?” — it stops at the threshold and ignores everything beyond it. Expected Shortfall answers the harder question: “given that I breach that threshold, what is my average loss past it?” ES is coherent and tail-sensitive (Almeida et al. 2017), which is why it has become the Basel-mandated risk measure where VaR understates fat tails.

Concentrated positions

How do I hedge a concentrated single-stock position?

You decouple price exposure from ownership without triggering a lock-up breach or a tax event. The standard architecture is a programmatic put-option overlay sized to your true concentration — when 60–90% of net worth sits in one ticker, even a 5–10% protective allocation changes the geometric outcome materially. This is the named core case, not an edge case.

The diagnostic

What does a portfolio diagnostic actually produce?

A 48-hour document, not a call. It returns your concentration ratio, an Expected Shortfall estimate across modelled regimes, and the cost of a calibrated hedge — quoted in plain numbers before any commitment. If your tail is smaller than you feared, the document says so.

Method

Why diffusion models over Monte Carlo?

Classical Monte Carlo inherits its tails from the distribution you assume — usually Gaussian, which systematically under-weights extreme co-movements. Diffusion models learn the joint structure from data, so correlation breakdown and fat-tailed clustering survive into the simulated paths; the quantitative method behind our stress engine documents the calibration. The engineering analogy: you are fuzzing the real failure modes, not replaying a sanitised spec — which is why it tracks the current risk environment rather than a static spec.

Cost

What does it cost, and what happens when I click?

The first diagnostic run is free; any ongoing carry is quoted in the document before you decide. Clicking submits three fields (one required: email) — no call, no sales sequence, no calendar. Context is collected after the first output, never before.

Failure modes

When does the model fail?

Every model has a failure rate, and we publish ours: roughly a 4.2% false-comfort rate — cases where the system understates realised tail loss under specific regime shifts. We state it up front because a risk tool that hides its own error band is the thing it claims to protect you from.

Run the diagnostic on your own book.

Run my diagnostic →

3 fields · 48-hour document · no call, no sequence.