Texas ITSAC Transparency Project

Open-source analysis of Texas DIR IT Staffing Augmentation Contracts (ITSAC).
FY2020-2026 | 182,591 invoices | $2.27B total spend | 10,352 workers

ITSAC Dashboard

Comprehensive analysis of IT staffing contracts across Texas state agencies. Time-series trends, vendor analysis, ethnicity distribution, HUB compliance, and worker-level data.

$2.27B
Total Spend
10,352
Workers
7 FYs
FY2020-2026

Billing Anomaly Analytics

Deep-dive into multi-vendor billing anomaly patterns. Ethnicity overindex analysis, vendor networks, agency exposure, buyer analysis, and investigation resources.

167
Fraud Cases
$130M+
Flagged Spend
1.46x
SA Overindex

Contract Viewer

Individual contract details parsed from LBB CMS purchase orders. Worker names, job titles, hourly rates, vendor/agency contacts. Static dataset: FY2025 through April 10, 2026.

1,044
Contracts
PDF
Source
LBB CMS
Origin

Legal Disclaimer

This site presents publicly available data for transparency and research purposes only. Nothing on this site constitutes an accusation of wrongdoing.

The term "billing anomaly" refers to data patterns where a worker appears to bill through multiple vendors in the same reporting period. There are many legitimate explanations for these patterns, including vendor transitions, subcontracting arrangements, corrections, and data entry errors.

All individuals and organizations referenced in this data are presumed innocent of any wrongdoing. The presence of a name on this site does not imply guilt, fraud, or any illegal activity. Only authorized government investigators and courts of law can make such determinations.

Ethnicity classifications are probabilistic inferences from names and will contain errors. They are included solely to identify potential systemic patterns in public contracting, not to characterize any individual.

The authors of this site are not affiliated with any Texas state agency. This is an independent public transparency project using open data.

Data Notes & Methodology

Rate/hr field: The "Unit Price" field from state data does not always represent an hourly rate. In some cases it is a monthly rate, lump sum payment, or per-deliverable price. Entries showing rates above $300/hr should be interpreted with caution. For example, Protiviti Government Services bills at $700/hr (plausible for Big 4 consulting), while some entries show $10,000+ "rates" that are actually lump-sum payments reported as 1 unit.

Avg Rate calculation: Average rates are computed from the unit_price field across all invoices where rate > 0. Outlier lump-sum entries inflate the average. The median rate across all invoices is approximately $90/hr.

Ethnicity classification: Inferred from worker names using GPT-4o-mini. This is probabilistic and will contain errors, particularly for ambiguous names. Categories are broad cultural/regional groupings, not precise ethnic identities. ~2% of names remain unclassified.

Billing anomaly detection: A "billing anomaly" means a worker billing through 2+ different vendors in the same reporting month. This can indicate legitimate subcontracting (LOW risk = same agency), vendor transitions, or potential irregularities (HIGH risk = different agencies). It is not proof of wrongdoing — it is a pattern that may warrant further review by authorized investigators.

FY2026 data: Only 5 months available (Sep 2025 - Jan 2026). Use "Same Period" mode on the Overview tab for fair YoY comparison.

Worker names: Some source data contains pipe-delimited names, comma-reversed names (Last,First), or partial names. These are normalized at display time. A small number of entries have job titles or descriptions in the name field.

Contract Viewer: Data parsed from ~1,044 LBB CMS purchase order PDFs using regex, OCR, and LLM extraction. This is supplemental to the official DIR data and may contain parsing errors. Fields like vendor/agency contacts, purchaser names, and PO numbers are not available in the Socrata API. Some agencies (TWC) redact worker names. DFPS contracts use large multi-page documents where extraction is less reliable.