feat(fred): triple the curated series list to the codes people actually search

[daniel3303]

Why

The economic-data pages are the best-ranking template on the deployment that consumes this module: Search Console puts /EconomicData/* at average position 5.4 over the 28 days to 2026-08-17. They are also the smallest surface we publish, 42 series out of the roughly 800,000 FRED carries.

This is the one place where the query is the identifier. Somebody types BAMLH0A0HYM2 and wants exactly that series, and that one code drew 1,064 impressions and 9 clicks in those 28 days. The constraint on the template is codes carried, not ranking.

What changes

The curated list goes from 42 to 114. Selection rule, now written into the file's header comment:

• chosen against FRED's own popularity score, floor of 50;
• confined to the categories already present, so no new taxonomy;
• every id resolved through the FRED series API before being written down. The first draft contained WILL5000INDFC, which does not exist. None of these are from memory.

What was missing was mostly structural:

Family	Gap
Interest rates	The curve started at the 2-year, so nothing could answer a question about the front end. Adds 1M through 20Y plus DFF, IORB, TB3MS.
Corporate bond spreads	The high-yield aggregate had no rating buckets under it, so a widening could not be told apart from a rotation within high yield. Adds BB, B, CCC, BBB, and the Moody's yields.
Housing	Starts and the 30-year rate, but no price. Adds MSPUS, HSN1F, PERMIT, MORTGAGE15US, existing-home sales, months' supply, homeownership.
Exchange rates	A trade-weighted index with no bilateral pairs beneath it. Adds the nine majors.
Money supply	No reverse repo facility, which is the most-watched item in that category.
Inflation, employment, output	The component and household-survey series that give the headline print its meaning.

Deliberate exclusions

Two series cleared the popularity floor and are kept out, with a test to keep them out:

• USSLIND stopped updating in April 2020.
• FPCPITOTLZGUSA is an annual World Bank restatement of US CPI that lags CPIAUCSL by a year.

A frozen series is worse than a missing one: it renders a current-looking page over a stale number, which is exactly the trap STLFSI2 sprang. That rule was a comment about one series; it is now Series_ExcludesTheSeriesFredHasStoppedUpdating, so the next curation pass cannot reintroduce a dead series on popularity alone.

Tests

Existing coverage already pins uniqueness, valid categories and every category being represented, so the additions are covered by it. New:

• Series_ExcludesTheSeriesFredHasStoppedUpdating
• InterestRates_CoverTheWholeTreasuryCurve (1M, 3M, 6M, 1, 2, 3, 5, 7, 10, 20, 30)
• CorporateBondSpreads_SeparateTheRatingBuckets (BB, B, CCC)

dotnet csharpier check . clean, dotnet build Equibles.sln -c Release 0 errors, dotnet test tests/Equibles.UnitTests 4,439 passed, 0 failed. Mutation-checked after committing: swapping one curve member for USSLIND failed both new guards, and the change was restored.

Cost

72 more series is 72 more rows in the daily refresh, well inside FRED's rate limit, and the observation history is fetched once.