How The Supply Map works

The Supply Map is a connected encyclopedia of how the world supplies America. It covers 36 essential goods, the inputs they depend on, the countries those inputs come from, and (as our extraction pipeline runs) the companies and facilities that produce them. Every signal we surface is traceable to a U.S. federal government API or database.

We do not use news articles, social media, or third-party aggregators as primary signal inputs. Each source has a defined ingestion cadence and a staleness threshold; data older than the threshold is flagged stale and excluded from the live surface.

All signals originate from U.S. federal government APIs and databases. We do not use news articles, social media, or third-party aggregators as primary signal inputs. Each source has a defined ingestion cadence and a staleness threshold — data older than the threshold is flagged as stale and excluded from live scoring.

Each source runs through a four-stage pipeline:

1. 01
   Fetch. The ingestion task calls the source API and pages through all available records. Raw API responses are stored as-is in risk_raw_observations with a content hash. Unchanged records are detected by hash comparison and skipped — only changed or new records proceed.
2. 02
   Normalize. Each source has a dedicated normalizer that converts the raw record into a canonical signal schema: severity (0–100), confidence (0–100), subject name, signal type, risk vector, observed date, and evidence URL. Normalization logic is source-specific but outputs a uniform structure.
3. 03
   Upsert. Normalized signals are upserted to risk_normalized_signals using the source's external ID as the deduplication key. Existing records are updated if content has changed; new records are inserted. Signal status (active / historical) is preserved across runs.
4. 04
   Match + score. After upsert, the matching engine runs deterministic matching to link signals to essential goods. After matching, new risk score snapshots are written for all goods using the formulas below.

Each normalized signal is matched against the 39 essential goods catalog using deterministic rules — not embeddings or fuzzy NLP. This makes matches auditable and correctable.

Each good has a curated list of matching terms and optional category rules. For example, the Beef & Ground Beef good matches signals whose subject name contains “beef”, “ground beef”, “hamburger”, “bison”, “veal”, or “steak”. USDA FSIS signals are additionally filtered by the product category field in the recall record.

Each match produces a match_score (0–100) used in scoring. Match score reflects the strength of the association: a direct product-name match scores higher than a category-only match. Matches are stored in essential_good_signal_matches and are re-run on every ingest to pick up new goods added to the catalog.

• Coverage gaps: only goods with at least one active data source are shown. Many essential goods (water infrastructure, housing materials, grid reliability) are not yet covered.
• Lag: scores reflect data as of the last successful ingestion. Government APIs may themselves lag behind real-world conditions by hours to days.
• False negatives: the absence of a government signal does not mean a good is safe. Shortages and disruptions sometimes precede official reports.
• Geographic granularity: national scores do not reflect regional variation. Gasoline in Houston may behave differently than gasoline in rural Maine. State and metro views are a future priority.
• Substitutes: scoring does not model cross-good substitution. If insulin is at severe risk and metformin is at low risk, the dashboard does not reflect that households may switch.
• Structural factor calibration: vulnerability factor ratings are set at catalog creation and updated periodically — not in real time. They reflect informed editorial judgment, not econometric estimation.

We want this data to be cited, challenged, and improved. If you find an error in a score, a recall record, or the methodology itself, please reach out. We will investigate and correct publicly.

Jon Ragsdale is a FINRA-registered investment adviser representative (CRD# 8129537, verifiable on FINRA BrokerCheck). The Supply Map scores are supply risk assessments, not investment or financial advice.