Bellwether

The economy's earliest tell, made explorable. Read the ISM PMI the way the Fed does.

Bellwether is an interactive atlas of the ISM PMI suite — Manufacturing PMI (1948→present, 75+ years) and Services PMI (1997→present) — read alongside the US economic-policy decisions that shaped each era. The PMI is the economy's earliest monthly tell: a diffusion index that crosses 50 before GDP turns. Bellwether makes that signal something you can drive rather than a number you read once a month.

It's for the economically curious non-economist — anyone who sees "ISM Manufacturing PMI came in at 48.7" in the headlines and wants the machine behind it: what the number means, how it's built, and what it has caught (and missed) across seventy-five years of cycles.

Try it live → bellwether-nine.vercel.app

Status: live — 7 modes, Manufacturing PMI back to 1948 (75+ years), Services PMI to 1997, per-cell data provenance, FRED-backed, no signup.

---

Modes

Mode	Route	What it does
Timeline	/	INDPRO Y/Y % spine 1949→present, with hand-curated ISM PMI markers, a separate Services PMI figure (1997→), and a policy lane of major US economic decisions.
Decompose	/decompose · ?report=services	Reweight the headline subindices and backtest your blend against NBER recessions — Manufacturing (5 subindices) and Services (4).
Fed Chair	/fed-chair	Era-locked decision game with after-action review. Manufacturing PMI only; scenarios pre-date the Services PMI's 1997 launch.
Heatmap	/heatmap · ?report=services	18 industries × 5 policy regimes, for both Manufacturing and Services.
PMI Explained	/pmi-explained	How to read the ISM Reports on Business, with live Manufacturing and Services prints.
About the data	/about-the-data	Per-cell provenance ledger — every value traced to its source.
Background	/background	Illustrated timeline of the institutions and the seventeen Fed chairs.

Beyond the timeline — drive the signal

The headline PMI is a fixed-weight blend of five subindices. In Decompose, move the weights and watch your custom blend redraw against the real series — then see whether your mix would have caught the last twelve recessions. (static version.)

Fed Chair (/fed-chair)	Heatmap (/heatmap)
Five inflection points, five charts that hide everything past the day of decision. Pick your action, then see what actually happened — and how three schools of thought read it.	The headline number is an average of eighteen industries. The matrix maps each one against five policy regimes, so you can see who won and who lost under what.

---

Quickstart

npm install
cp .env.example .env.local         # paste your FRED key
npm run refresh-data               # fetch FRED snapshots into data/fred/
npm run dev                        # http://localhost:3000

The build doesn't require FRED at runtime — all market data is checked-in JSON in data/. Re-run npm run refresh-data when you want fresher FRED snapshots, and npm run import-prnewswire to backfill the most recent ISM headline prints from ISM's monthly press releases on PRNewswire. ISM Wayback re-imports (for the post-2014 history and the per-month subindex / industry tables) are documented in data/CONTRIBUTING.md.

Stack

• Frontend: Next.js 16 App Router · TypeScript (strict, noUncheckedIndexedAccess) · Tailwind CSS + shadcn-style primitives
• Charts: D3 (d3-shape, d3-scale, d3-array) for the timeline and backtests
• State: Zustand (Fed Chair scenario state)
• Data: checked-in JSON, validated with Zod at load time; sourced from FRED, ISM Reports on Business via the Internet Archive, ISM press releases on PRNewswire, and a forecasts.org historical mirror
• Testing: Vitest
• Analytics: Vercel Analytics

No database, no auth.

Repo layout

src/
  app/                            # Next.js App Router
    page.tsx                      # Timeline (default mode)
    decompose/page.tsx            # ?report=services for Services
    fed-chair/page.tsx            # scenario index
    fed-chair/[id]/page.tsx       # per-scenario runner
    heatmap/page.tsx              # ?report=services for Services
    pmi-explained/page.tsx
    about/page.tsx
    about-the-data/page.tsx
    background/page.tsx
    style-guide/page.tsx
  components/
    timeline/                     # PMITimeline, PolicyLane, etc.
    decompose/
    fed-chair/
    heatmap/
    background/
    pmi-explained/
    ui/                           # primitives — Button, Sheet, Card, Slider, etc.
  lib/
    data/                         # series loaders + Zod schemas
    content/                      # policies + glossary (data lives in `_policies-data.ts`, not MDX)
    decompose/                    # weighting math + backtest
    background/                   # illustrated timeline data
data/
  fred/*.json                     # FRED snapshots (INDPRO, IPMAN, FEDFUNDS, USREC, NAPM stubs)
  pmi-historical.json             # forecasts.org mirror, 1948 → 2014
  pmi-wayback.json                # ISM Mfg ROB via Internet Archive, 2014→present
  pmi-subindices-wayback.json     # the five Mfg subindices
  pmi-curated.{csv,json}          # Mfg headline rows from ISM press releases (PRNewswire) + hand entries
  nmi-wayback.json                # ISM Services ROB via Internet Archive
  nmi-subindices-wayback.json     # the four Services subindices
  nmi-curated.json                # Services headline rows from ISM press releases (PRNewswire)
  industry-monthly-wayback.json   # Mfg per-month industry growth/contraction lists
  services-industry-monthly-wayback.json
  sectors.json                    # Mfg heatmap (18 industries × 5 regimes)
  sectors-services.json           # Services heatmap
scripts/
  fetch-fred.ts                   # FRED refresh
  import-wayback-ism.ts           # Mfg headline scraper (Wayback Machine)
  import-wayback-subindices.ts    # Mfg subindices (Wayback Machine)
  import-wayback-industries.ts    # Mfg industry lists (Wayback Machine)
  import-wayback-nmi*.ts          # Services equivalents (Wayback Machine)
  import-prnewswire-ism.ts        # current-vintage Mfg + Services headlines from PRNewswire
  import-forecastsorg.ts          # historical mirror
  import-pmi-csv.ts               # CSV → JSON for hand-curated rows
  reconcile-ism.ts                # 3 checks/sector: cross-source, composite, spot-check
  verify-nber.ts                  # link health on NBER citations

The data

Bellwether's edge is its provenance. Nothing is scraped live at request time — every series is a checked-in JSON snapshot, backfilled from several primary sources and cross-checked against each other:

• FRED for the macro spine (INDPRO, IPMAN, FEDFUNDS, USREC).
• ISM Reports on Business via the Internet Archive for the post-2014 headline, subindex, and per-industry history.
• ISM press releases on PRNewswire for current-vintage Manufacturing and Services prints.
• A forecasts.org mirror for the deep 1948→2014 history.

Those sources are held against each other rather than trusted individually. npm run check-data requires that months carried by two sources agree, that each headline equals the mean of its own subindices, and that hand-read anchor values match what ships — so a scraper that starts quietly returning the wrong number fails the build instead of publishing.

Every value is reproducible, and the per-cell provenance ledger traces each one to its source. See data/CONTRIBUTING.md for how the imports work and data/LICENSING.md for the per-source licensing posture before you redistribute the data.

Conventions

• Files: kebab-case
• Components: PascalCase
• All exports get JSDoc
• Strict TypeScript — noUncheckedIndexedAccess is on, deal with undefined
• Max ~400 lines per file; split when bigger
• Data is validated with Zod at load time, not at use time

Documentation

• docs/pmi-explainer.md — the long-form explainer on reading the ISM Reports on Business.
• SETUP.md — full local setup, environment variables, and the data-refresh workflow.
• DISCLAIMER.md — what Bellwether is and is not (it is not investment advice).
• SECURITY.md — how to report a vulnerability.
• CITATION.cff — how to cite the project.

Contributing

Contributions are welcome — especially data corrections, new policy events, and deeper provenance. Good first contributions: file a data correction when you spot a value that disagrees with the primary source, or improve the explainer for one of the subindices. See CONTRIBUTING.md to get started and CODE_OF_CONDUCT.md for community expectations; bug reports and feature requests have their own issue templates.

Before opening a PR: npm run lint && npm run typecheck && npm test.

License

The project's own code and content are released into the public domain under CC0 1.0 — use it for anything, no attribution required.

The data layer is different and is not dedicated to the public domain: economic values are facts reproduced with attribution to their primary sources, and quoted ISM commentary is included as fair-use excerpts. See data/LICENSING.md for the per-source posture before redistributing the data.