MacroGauge

A daily, independent view of US inflation—built from live market data, official releases, and auditable vintage history.

Open MacroGauge · Explore Data Center Inflation · Read the methodology

MacroGauge re-prices the CPI basket using higher-frequency market and alternative data, compares the result with official inflation, forecasts upcoming prints, and publishes the evidence behind every number. The Python pipeline performs all collection and calculation; the Next.js site renders pre-built, schema-validated JSON and does no analytical work in the browser.

Data Center Inflation

The dedicated Data Centers page tracks a category that has no official all-in price index: the cost to build, equip, and operate US data centers.

Available now

• DC Build Index — construction labor, steel, concrete, copper and aluminum, switchgear, transformers, generators, HVAC equipment, and pumps.
• DC Ops Index — industrial electricity, facilities and operations labor, and machinery maintenance.
• DC Hardware Index — transaction-sensitive official price series for compute, storage and memory, and networking equipment.
• Live market tails — copper and aluminum futures extend the relevant monthly PPIs beyond their latest official print, and DRAMeXchange NAND spot prices extend the storage-device PPI during large memory-price swings, without overwriting official history.
• State cost parity — build and operating-cost multipliers combine QCEW construction wages and EIA industrial electricity prices with nationally priced inputs.
• Full receipts — component weights, contributions, last observations, quality holds, and the contrast between transaction-sensitive and hedonically adjusted hardware series.

All three indexes are rebased to 2018-01 = 100. Build and Ops remain separate because capex and opex have different cost drivers; the project does not manufacture a blended total-cost-of-ownership number from an arbitrary capex/opex split.

The construction boom, in real terms

The page also carries monthly Census C30 data-center construction spending — and a series no one else publishes, because it requires a data-center-specific cost deflator:

• a keyless Census XLSX connector for the seasonally adjusted annual-rate and not-seasonally-adjusted construction series;
• nominal construction spending in dollars and NSA same-month YoY growth;
• real data-center construction spending, calculated by deflating Census nominal spending with MacroGauge's own DC Build Index into constant January 2018 dollars;
• revision-aware ingestion into the append-only vintage store, preserving preliminary, revised, and final Census values;
• a nullable construction block inside datacenter.json, so the section degrades cleanly if the source breaks;
• source-drift checks and an isolated CENSUS failure domain, so a workbook-layout change cannot break the core inflation gauge or the rest of the Data Center page.

The design is documented in DC Construction Boom Design.

AI capacity tracker

The AI capacity page prices neoclouds, ex-miners and hyperscalers per megawatt against how much of their capacity is energized, opens each company's full record, schedules dated construction by quarter (searchable by ticker, company or customer), and maps every campus.

What you can explore

Area	Pages and capabilities
Inflation gauge	Supercore, cost of living, Gauge-vs-BLS gap decomposition, official comparison, component heatmaps
Personal inflation	Custom basket reweighting, grocery prices, since-date calculator, real-wage analysis
Forecasts	CPI preview, next-print nowcast, 12-month outlook, forecast scoreboard, model matrix, release log
Macro conditions	Economic heat check, consumer stress, recession risk
Data centers	Build, operating, and hardware inflation; state parity; component-level receipts
Transparency	Live source status, QA results, methodology, vintage replay, and forecast accountability

How it works

Official + market + alternative sources
                  │
                  ▼
       isolated source connectors
                  │
                  ▼
   append-only vintage observation store
                  │
                  ▼
 pure engines: rebase → blend/splice → gate → aggregate
                  │
                  ▼
 schema-validated JSON artifacts + QA receipts
                  │
                  ▼
       Next.js static site → Vercel

The weekday workflow is triggered by an external scheduler at 8:45 AM, 10:45 AM and 1:45 PM Eastern, with GitHub crons (8:40 AM Eastern plus backups) as the fallback for missed triggers; a publish gate keeps it to one publish per day. It collects new observations, recomputes the products, validates every artifact, and commits the resulting store and site data. A new data commit is the pipeline heartbeat; a green workflow that skipped its publication gate is not counted as a publish. On CPI/PPI release days the gate lets a later firing republish when the day's print has not yet reached the store, so a cron that lands before FRED propagates the release cannot leave the site a month stale.

Reliability is built around a few hard rules:

• Source isolation: a failed connector is reported in sources_status.json; it does not stop unrelated sources.
• Phase isolation: the gauge, nowcast, outlook, composites, and Data Center Index publish independently and report failures through qa.json.
• Contract safety: JSON Schema validation failures stop deployment. Invalid artifacts never ship.
• Staleness over silence: stale observations carry forward within explicit limits and remain visible in source and component receipts.
• No browser-side analytics: published JSON is the result; the site only formats and visualizes it.
• No live network calls in tests: connectors accept injected HTTP functions and use recorded or generated fixtures.

Repository map

pipeline/
  connectors/        one external source per connector
  engine/             pure calculation stages and product engines
  publish/            JSON builders, writers, and validation
  store/              append-only vintage storage
config/               source registry, baskets, composites, and calendars
schemas/              one JSON Schema contract per published artifact
store/obs/            monthly JSONL vintage partitions
site/
  src/app/            static routes
  src/components/     reusable React and chart components
  src/lib/            presentation utilities and client-only helpers
  public/data/        pipeline-generated JSON consumed by the site
tests/                pytest suite and network fixtures
docs/                 architecture, design specs, and implementation plans

The main architecture reference is macrogauge-design.md. The Data Center index methodology and pipeline architecture are described in Data Center Cost Index Design.

Run locally

Prerequisites

• Python 3.12+
• Node.js 22+
• npm
• A FRED API key for pipeline runs
• Optional EIA, BLS, FMP, and USDA keys for their respective live sources

Build and run the site

The repository includes the latest published JSON, so the site can run without collecting fresh data first.

cd site
npm ci
npm run dev

Open http://localhost:3000. To produce the static export:

npm run build

Run the data pipeline

From the repository root:

python3 -m venv .venv
source .venv/bin/activate
pip install --require-hashes -r requirements.lock

export FRED_API_KEY="..."
export EIA_API_KEY="..."   # optional, enables EIA sources
export BLS_API_KEY="..."   # optional, enables registered BLS sources
export FMP_API_KEY="..."   # optional, enables market-data tails
export USDA_API_KEY="..."  # optional, enables USDA sources

python -m pipeline.run_daily --store store --out site/public/data

FRED_API_KEY is required to start a daily run. Missing optional credentials surface as isolated source failures; inspect site/public/data/sources_status.json and site/public/data/qa.json after the run.

Test and verify

Pipeline tests run from the repository root:

pytest -q
pytest tests/test_dcindex.py -q
pytest tests/test_run_daily.py -q

Frontend checks run from site/:

npm run build
npm test
npm run e2e

CI runs the full Python suite, static build, Vitest suite, and Playwright smoke tests on pushes to main and on pull requests.

Data contracts and vintage policy

Every file in site/public/data/ has a corresponding contract in schemas/. Writers validate their output before it becomes deployable, and array alignment or cross-field invariants that JSON Schema cannot express are pinned in tests.

Rows in store/obs/*.jsonl are immutable and schema-versionless:

• new Observation fields may be added;
• existing fields are never renamed, removed, or retyped;
• readers provide defaults for fields absent from older partitions;
• committed partitions are never rewritten;
• merge conflicts preserve both sets of observation rows.

This policy keeps old vintages replayable and makes source revisions auditable over time.

Deployment and project status

• Production: macrogauge-cloudten.vercel.app
• Daily publisher: .github/workflows/daily.yml
• CI: .github/workflows/ci.yml
• Current work: a market-data memory nowcast tail for the DC Hardware Index (no official DRAM price index exists) and a cost-of-compute section (GPU rental and AI inference prices)

MacroGauge is an analytical project, not investment advice. Source data can be revised, delayed, or unavailable; the site exposes freshness and QA state so those limitations remain visible.