"Silver price" returned US CPI

Published 2026-10-11. The before-and-after results are read from series-lookup-2026-10-11.json, produced by running both versions of the lookup on the same requests.

The bug

Agents rarely pass the exact identifier a data source wants. They pass words: "10 year treasury yield", "core inflation", "silver price". So most tools that wrap a data source carry a small resolver from words to an identifier.

In canli-markets-mcp 0.1.0, economic_series took a FRED series ID or plain words. The plain-word path scored each of its 47 common series by the share of the request's words found in the series description, and took the best one if at least half matched:

const words = s.toLowerCase().match(/[a-z0-9&]+/g) ?? [];
const scored = COMMON_SERIES
  .map(([id, text]) => [id, words.filter((w) => text.split(" ").includes(w)).length / words.length])
  .sort((a, b) => b[1] - a[1]);
if (!scored.length || scored[0][1] < 0.5) throw new NotFound(/* ... */);
return { id: scored[0][0] };

A request of two words needs only one shared word to reach half. The consumer price index is described as "cpi consumer price index inflation all items", so "price" alone was enough:

Every wrong answer came back as a real series with real numbers. The result did carry the series title, so the mistake was there for anyone who read it. Nothing made an agent read it.

The words that decide each question ("silver", "japan", "uk") are the ones that matched nothing. The words that did match ("price", "inflation", "rate") say what kind of number is wanted, not which one.

The rule that replaced it

One function now serves three resolvers: FRED series, World Bank indicators and OECD outlook measures. Each word of a request falls into one of three groups:

• Filler is ignored: "what is", "show me", "since", "this year", and four-digit years.
• Kind words say what sort of number: rate, growth, index, level, daily, monthly. They count toward the best match but are never required.
• Specific words are everything else, and every one of them must appear in the entry.

export function matchWords(entries, textOf, request, vocab) {
  const asked = split(request).filter((w) => !vocab.filler.has(w) && !/^(1[89]|20)\d\d$/.test(w) && !/^[a-z]$/.test(w));
  const specific = asked.filter((w) => !vocab.kind.has(w));
  if (!specific.length) return null;
  let best = null, most = 0;
  for (const entry of entries) {
    const has = new Set(split(textOf(entry)));
    if (!specific.every((w) => has.has(w))) continue;
    const shared = asked.filter((w) => has.has(w)).length;
    if (shared > most) { best = entry; most = shared; }
  }
  return best;
}

When nothing qualifies, the tool says so and says where to look: another country's economy is in world_economy or economic_outlook, and the price of a commodity, currency, crypto or stock is in price_history. An agent that gets that refusal can call the right tool next. An agent that got consumer prices had nothing telling it to.

Test the refusals first

The old tests checked that good requests found their series, and they all passed. None asked whether a request for something the tool does not have gets refused. The test file now opens with 9 requests that must be refused, then 42 plain requests that must each resolve to one exact series. The list of common series grew to 73 at the same time.

A resolver from words to identifiers is a search engine that returns one result. It needs a way to return none, and the tests that prove it can are the ones to write first. The code is in mcp-markets/src/words.mjs in the canlicapital repository, and the tool is part of canli-mcp.