China Origin Component Risk Score: How to Screen a Bill of Materials

A China origin component risk score condenses several sourcing questions about one part number into a single value: where the manufacturer sits, whether the part's category counts as critical, and how many suppliers can actually fulfil orders. In the Electronic Components Supply Chain & Risk Dataset the score never travels alone - each row pairs it with an is_chinese boolean origin flag, a confidence_score observed between 0.7 and 0.95, and a risk_factors list of reason codes. Three codes appear across the published rows: chinese_origin, critical_category and limited_suppliers.

Observed risk_score values span 0.1 to 0.6, with nulls on rows where nothing triggered. The spread matters operationally. A 0.6 tagged chinese_origin reads nothing like a 0.1 whose only trigger was limited_suppliers: one is a geography verdict, the other is a scarcity count. No published documentation explains the weighting behind the decimal, so treat the number as triage input that orders human attention - not as an auditable rating.

Exactly once in the electronic-components slice: the Electronic Components Supply Chain & Risk Dataset, cataloged by Datadory from source Hugging Face. It holds 791 rows, one per manufacturer part number, each stacking four layers on the same key. Identity: mpn, manufacturer, manufacturer_country. Catalog: category label, free-text technical description, datasheet references. Commerce: a suppliers array deepened by a nested seller-detail block down to individual offers with SKUs, price points and inventory levels. Risk: is_chinese, confidence_score, risk_score and risk_factors.

Coverage notes set expectations honestly. Manufacturer origins span the US, EU, Japan and China; distributor offers come from DigiKey, Mouser, TME, Conrad, Win Source and independent brokers. Only 14 of the 791 rows carry is_chinese=true, so the panel screens a curated analog-and-embedded shortlist rather than the whole market - the Nexar API (Octopart) alone reaches millions of components, FindChips Aggregated Price & Inventory Search spans roughly 41,473 public MPN landing points, and the LCSC Electronics Catalog lists about eight million product pages per storefront language. Within its own scope the record is unusually well documented: Datadory scores it 7 out of 10 against a cross-catalog mean of 7.81 across all 1,744 datasets, with field definitions verified against published sample rows.

Manufacturer concentration is heavy and easy to quote: Analog Devices holds 195 of the 791 rows, Texas Instruments 184, STMicroelectronics 100, Espressif Systems 63, GigaDevice 48, Microchip 41, Seeed Studio 28 and Adafruit Industries 22. Three sampled rows show how differently parts reach the same score scale:

mpn=ESP8266EX   mfr=Espressif Systems   country=China   is_chinese=true
category=RF Receivers, Transceivers   confidence=0.95
risk_score=0.60   risk_factors=[chinese_origin]

mpn=DFR1063     mfr=DFRobot             risk_score=0.25  risk_factors=[critical_category]
mpn=GPY         mfr=Pycom               risk_score=0.10  risk_factors=[limited_suppliers]

ESP8266EX is the only sampled row pairing a high score with a China origin flag: 0.6 at 0.95 confidence on chinese_origin. DFR1063 reaches 0.25 through critical_category, and GPY sits at 0.1 through limited_suppliers - neither trigger involves geography at all. That is exactly why the reason codes deserve reading beside the number: the same decimal can mean origin exposure, category criticality or plain supplier scarcity.

Ranking a whole bill of materials on the raw score alone therefore misorders work. Sort by risk_score descending within each reason-code group - chinese_origin rows first, then critical_category, then limited_suppliers - and attach the manufacturer_country value so a reviewer sees geography and scarcity in one screen. Across 791 rows that sort is a one-pass job, and Datadory returns it already applied to whatever subset you name.

You send the part-number list; the screen comes back as typed rows. The mechanics, in the order they matter:

1. Match on the MPN. Both sides normalize to uppercase with package-code suffixes stripped before joining; matched rows return the five working columns - manufacturer_country, is_chinese, confidence_score, risk_score and risk_factors. The manufacturer part number is the join key across the entire slice.
2. Read nulls as unscored, not safe. With only 14 flagged rows out of 791, most joins come back clean by absence. The delivered sort puts nulls last, score descending, confidence alongside so a reviewer can weigh how loudly each verdict speaks.
3. Take the seller detail pre-flattened. The nested seller block carries full offer objects with price-break arrays and inventory levels; delivered relationally it arrives as pipeline-ready columns instead of a nesting problem that explodes row counts.
4. Get the verdict and its evidence together. Every scored row ships with the category label, description and supplier list that sit behind it, so a challenged score can be argued with on the merits rather than taken on faith.

The score orders attention; four live layers decide whether acting on it is actually possible. All join on the same MPN key:

The Nexar API (Octopart) contributes multi-distributor depth - per-seller offers, quantity-break pricing, inventory levels, lifecycle status and factory lead days on the same part record. FindChips Aggregated Price & Inventory Search spreads the same question across authorized, independent and industrial/MRO channels, with alternates and watch alerts riding alongside. The LCSC Electronics Catalog brings Shenzhen-side volume - per-warehouse stock rows and USD/CNY price ladders skewed toward Asian domestic brands Western catalogs underrepresent. The TME Electronic Components Catalog & API attaches forward-looking supply dates and waiting periods from one major European distributor's own warehouse positions.

Together they answer the question a flag cannot: if this part needs a second source, does one exist on a shelf near your build, at what price and lead time? The second table below lays the stack out layer by layer.

Four limits bound what the number can tell you.

Breadth. 791 rows against catalogs of millions means absence of a flag is not evidence of safety - the panel is a shortlist, and unlisted parts simply carry no verdict. Methodology comes second: no published documentation explains how risk_score, confidence_score or the three reason codes are computed, and sampled scores run only 0.1 to 0.6 with some rows null. Third, freshness: the scored panel is a single-pass snapshot whose prices and stock are as-of-collection values, while the live layers beside it reflect current state - which is precisely why the verdict and the confirmation should arrive in the same delivery. Fourth, semantics: because triggers mix origin, category and supplier count, a mid-score part may need a second source rather than a redesign.

Use the score to order human attention, then confirm supplier reality against live rows before committing to a redesign or a dual-source qualification. The electronic component price history data walkthrough shows how repeated capture timestamps those live quotes into a series worth reviewing monthly.

Five teams run on these rows, ranked by how directly a scored panel settles their day job:

1. Supply-chain risk managers screen BOMs for geographic exposure part by part, and because every flag ships with reason codes and a confidence score, the result survives audit - the explanation travels with the verdict.
2. Component and sourcing engineers run second-source checks before design freeze: how many distributors carry the candidate part, at what price and stock. A part with one offer object is a different procurement decision than one with six, which is what limited_suppliers fires on.
3. Competitive intelligence and product teams map manufacturer concentration per category - who owns the supply base and where second-source gaps sit across RF, MCU, sensor and power lines. Their workflow continues at competitive intel product teams.
4. Data scientists and quant teams calibrate origin-classification models on the flag-plus-confidence structure and treat the nested seller arrays as ground truth for pricing work. Their workflow: data scientists.
5. Investors and analysts quantify semiconductor sourcing exposure at the component level, beneath company-level disclosure. Their workflow: investors & quants.

Because the hard part was never locating a score - it is reconciling a static judgment layer with live commerce layers that define stock, currency and granularity differently, then keeping the join alive as catalogs churn.

Every delivery ships with the field dictionary unchanged and sample rows for validation. Name the MPNs, manufacturers or categories you care about and request a sample scoped to your own bill of materials: the schema in the sample is the schema you ship against.

Start with the electronic components data guide, the pillar that scores and groups all eight records in the slice, or browse the ranked best electronic-components datasets. For adjacent workflows, see the electronic component MPN search database comparison on the resolution side of the same join and the electronic component price history data walkthrough on turning quotes into series. The full pooled view lives on the electronic components data hub, and the Electronic Components Supply Chain & Risk Dataset record stands behind every figure above.