Methodology

How Nexus links federal contract data to public company information

Overview

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   SEC EDGAR     │     │    SAM.gov      │     │  USAspending    │
│  (Public Cos)   │     │ (Fed Vendors)   │     │ (Fed Contracts) │
│                 │     │                 │     │                 │
│  CIK + Ticker   │     │      UEI        │     │      UEI        │
│  89K companies  │     │  876K entities  │     │  179M awards    │
└────────┬────────┘     └────────┬────────┘     └────────┬────────┘
         │                       │                       │
         └───────────┬───────────┴───────────┬───────────┘
                     │                       │
                     ▼                       ▼
              ┌─────────────────────────────────────┐
              │         ENTITY CROSSWALK            │
              │  SEC CIK ↔ SAM UEI ↔ USAspending   │
              └─────────────────────────────────────┘
                              │
                              ▼
                         NEXUS UI

Data Sources

SEC EDGAR

Public company filings

89K companies

Key Fields:

CIK, company name, tickers, SIC code, state, EIN

Update Frequency:

Daily

SAM.gov

Federal vendor registry

876K entities

Key Fields:

UEI, CAGE code, legal name, DBA, address

Update Frequency:

Monthly

USAspending

Federal contract awards

179M awards

Key Fields:

Award ID, recipient UEI, agency, amount, dates

Update Frequency:

Daily

Entity Resolution

The Problem

No common identifier exists across all three sources. SEC uses CIK, SAM/USAspending use UEI, and company names are inconsistent.

"Apple Inc.""APPLE INC""APPLE COMPUTER INC"

Step 1: Name Normalization

Standardize company names by removing legal suffixes, punctuation, and converting to uppercase.

"Apple Inc."→"APPLE"

"Microsoft Corporation"→"MICROSOFT"

Step 2: Blocking

Group by first 3 characters to reduce comparison pairs from 78 billion to ~50 million.

Block "APP": Apple, Applied Materials, Applebee's...

Step 3: Trigram Similarity

Use PostgreSQL's pg_trgm extension for fuzzy matching. Match threshold: similarity > 0.7

similarity('APPLE', 'APPLE') = 1.00

similarity('MICROSOFT', 'MICROSFT') = 0.82

Step 4: Score Boosting

Additional signals improve accuracy: same state (+0.1), same EIN (+0.2), exact match (1.0)

Expected Results

10-20K

SEC ↔ SAM matches

85-90%

Overall accuracy

500K+

SAM ↔ USAspending (UEI exact)

<5%

False positive rate

Data Refresh Strategy

Source	Frequency	Method
SEC EDGAR	Weekly	Delta update from submissions.zip
SAM.gov	Monthly	Full refresh from monthly extract
USAspending	Monthly	Incremental award updates
Crosswalk	Monthly	Re-run after source updates

Limitations

•Fuzzy matching is imperfect — Some false positives/negatives will exist
•Subsidiaries — SEC data is at parent level; contracts may be to subsidiaries
•Name changes — Companies change names; former_names field helps but isn't complete
•International — Focus is US federal contracts; international subsidiaries may be missed
•Private companies — SEC only covers public companies; many large contractors are private