A world event is rated Low, Middle or High. Here is how.

The base rate is an editorial estimate of how likely the event is over the next year, given long-run history. It answers: “if you knew nothing else about today, how likely is this?” It does not move with the market.

Then author-written rules look at signals. A rule says: if this signal reads this way, the event is more (or less) likely than usual. Rules that measure the same thing are not counted twice: the strongest one wins, and two overlapping rules are discounted. The result is clamped so one noisy reading can never push a scenario to certain or to impossible. Every band opens to a “Why?” list of which rules fired and what each signal read.

The signals today:

• VIX close: the stock market’s fear gauge (FRED VIXCLS).
• Oil volatility: how wildly the United States Oil Fund has moved over 30 days. It captures the oil tail without paying for an options feed.
• US Treasury rates: 10-year, 2-year, the gap between them and the Fed funds rate (FRED).
• Conflict news by region: for each of eleven regions, the share of news events that are conflict rather than diplomacy, weighted by severity, from GDELT event codes. A region only counts when it runs hotter than its own history, so a permanently busy region does not read as a permanent emergency.

Step one, channels. Each scenario is authored as the set of things it moves: the crude oil price, European gas, a shipping lane, policy rates, the dollar, equity volatility, credit spreads, sanctions, advanced chips, industrial metals, grain. Each line has a direction, a size and one sentence of why. The map is in lib/scenarios/transmission/channels.ts, in plain English.

Step two, shocks. Small, medium and large become numbers in one place only, lib/scenarios/transmission/shocks.ts. Oil moves 5, 10 or 20 percent. Gas 10, 20 or 40. A shipping lane loses 30, 60 or 100 percent of its usual transits. The whole stock market moves 2, 4 or 8 percent, calibrated to published research that the median geopolitical drawdown is about 4 percent over roughly two weeks.

Step three, reach the holding. For each stock the engine uses the first of four layers that has a real basis, and says on screen which one:

• Declared exposure. The company itself disclosed something the scenario touches: a trade lane, a commodity it makes or buys, revenue in an affected region. Half of a commodity move reaches a producer, 30 percent reaches a buyer, and half of revenue in a hit region is treated as impaired. Several rules about one company take the strongest, not the sum.
• Sector history. No company basis, but we have measured this kind of event: the scenario’s market shock plus how much better or worse the sector did than the market, only where at least five past events exist.
• Beta. The weakest basis: the stock’s sensitivity to the market times the scenario’s market move.
• None. No basis. The holding gets no number and is counted as uncovered, never as zero, because zero would read as “not affected” when the truth is “we do not know”.

Step four, the what-if dial. The severity slider multiplies the step-two numbers, between a quarter and double, so every layer moves together.

One deliberate choice: the history layer uses the authored market shock plus the sector’s excess move, not the raw historical sector median. Those raw medians are positive at one and three months for every event class because markets recovered, and would have published a Taiwan invasion as a gain.

Worst case is the scenario at its authored full severity. It is not a value-at-risk number or a tail estimate. It answers: if this happened as written, what would it be worth to me.

The weighted case compares a large impact that is unlikely with a small one that is probable. It is an average of two outcomes, so you will never see it happen: the outcome is the event or no event.

It does not scale the total up to cover holdings it could not price. If we priced 60 percent of your portfolio, the figure covers that 60 percent and the other 40 is reported as uncovered. It does not model correlation, second-round effects, your trading, options, leverage or hedges, and it does not adjust for time: the number is the move, not a path. Percentages are over your whole portfolio, so a thinly covered portfolio reads as a small number beside a large uncovered line.

Row confidence is computed, not asserted: low when under half the portfolio could be priced or when beta carries most of it, high when most priced value rests on company-specific disclosure, otherwise moderate.