Same Job Title, Different Job: Inside Federal vs Private Tech Hiring

Two weeks ago we priced the federal tech bargain: lower pay and almost no mobility, in exchange for near-total job security, with a back-loaded pension holding the trade together. That explained why federal tech workers so rarely leave.

It left a thread hanging. Even the youngest, un-vested federal workers quit at only ~5% — far below their private peers — which we attributed to "non-transferable experience binding from day one, before any pension handcuff exists." Data scientist Abigail Haddad put it more sharply: "lack of movement is about skill set / tech stack as much as it's about pension… you'd see more movement if you looked at folks who were more hands-on-keyboard coders."

She's right — and this shows what that non-transferable experience actually is. There's a second wall, independent of the pension: under identical titles, federal and private tech postings ask for different tools, and several role categories that define private tech barely exist as federal jobs.

The roles that barely exist

How many of each data role does federal hiring actually post? In our federal tech index (~1,345 postings) versus the private index:

Title	Federal	Private
Data Scientist	129	2,181
Data Analyst	2	1,038
Data Engineer	17	1,758
Machine Learning Engineer	3	2,555
AI Engineer	5	848
MLOps (any title)	0	—

Data Engineer, ML Engineer, AI Engineer, MLOps — the backbone of a modern private data organization — are functionally not federal jobs. And our federal index is already filtered for tech relevance, which would tend to keep such roles if they existed, so single-digit counts are a signal, not an artifact. The federal data workforce is the analyst-and-researcher lineage: Data Scientists, Operations Research Analysts, and Statisticians.

Same title, different job: the federal Data Scientist

Take the one title that exists on both sides and read what the postings ask for. (Federal N=143, private N=2,000; every gap is statistically significant.)

The split is clean. Federal Data Scientist postings over-index on statistics-and-reporting — statistics (96% vs 54%), data visualization (44% vs 23%), dashboards (41% vs 27%), SAS/SPSS (13% vs 3%). Private postings over-index on code-experiment-deploy — Python (63% vs 43%), experimentation / A-B / causal inference (34% vs 6%), MLOps / deployment (40% vs 29%), cloud (16% vs 6%), LLM/GenAI (16% vs 5%).

The tell is the middle: both sectors say "machine learning" at nearly the same rate (~40%). Federal data scientists name ML as often as anyone — they just lack the hands-on toolchain to operationalize it. It stays a concept, not a deployed system. Even federal "Data Scientist (AI)" roles turn out to be about AI standards and evaluation policy, not building models — verbatim: "promotes the adoption of standards, guides, best practices, and policy for measuring and evaluating AI technology."

A federal Data Scientist and a private Data Scientist share a title and a vocabulary, but not a job: one describes machine learning, the other ships it.

The pattern repeats: Software and Security

Software Engineer (federal N=98) — the milder split. Federal software engineers use Python about as often as private ones (29% vs 31%), but they're markedly less cloud-native (AWS 13% vs 24%, Kubernetes/Docker 10% vs 21%, TypeScript 5% vs 19%) and carry a compliance overlay private software lacks (RMF/NIST/FISMA 7% vs 1%). Federal software modernizes defensively more than architecturally.

Cybersecurity (federal N=814) — the sharpest divergence in the data. Federal security is clearance-and-governance: security clearance in 92% of postings, RMF 11%. But the hands-on toolset is nearly absent — Python 2% vs 41%, threat modeling 0% vs 24%, Kubernetes/Terraform 1% vs 28%, SIEM 2% vs 20%. A private "Security Engineer" is a cloud DevSecOps builder; a federal "Cybersecurity Specialist" is a cleared compliance professional.

Why matched titles cover so little: the composition inverts

Collapse both sides into role families and the shares invert:

Role family	Federal tech	Private tech
Software/Dev (hands-on)	4.4%	43.2%
Data/ML/AI	13.5%	18.4%
Security/Cyber	37.8%	3.1%
IT/Ops/Infra	31.0%	12.3%
Program/Product/Mgmt	13.3%	23.1%

Hands-on software development is ~10× rarer in federal tech; security is ~12× more common. And one occupational series — 2210 (IT Management) — is 81% of federal tech postings (Data Science, series 1560, is ~9%). The hands-on-keyboard coders most able to arbitrage a pay gap by moving are exactly the population federal tech barely employs. Low mobility isn't only a handcuff on the people who are there — it's partly a question of who's there to begin with.

Two more walls: clearance and the closed remote door

Clearance. 58% of federal tech postings require a real security clearance (Secret / Top Secret / SCI) — 75% in security roles — versus low single digits private. A private engineer without one can't take most federal tech jobs regardless of skill.

Remote — the door just slammed shut. (This one panel uses the longer Loyola/USAJOBS history, since our own index doesn't reach before 2026.) USAJOBS carries an official "Remote job" designation (work-from-anywhere). Federal tech remote hiring held at ~6–10% from 2022 through January 2025 — then collapsed to ~0.3%, dropping precisely in Q1–Q2 2025: 7.5% in January → 1.3% in March → ~0% after. That's the federal return-to-office order (January 2025). Private tech still designates ~31% of postings fully-remote. The field is demonstrably real, not a scraping glitch: it read a healthy 6–10% for three years, then collapsed exactly when policy changed.

What it means

If you're a federal tech worker thinking of leaving: the pension is the handcuff you can feel; the skills gap is the one you can't. Operations Research, statistics-and-reporting data science, clearance-gated compliance security — the private market runs on Python, cloud, experimentation, and DevSecOps. The move is a retraining project, not just a resignation, which is why so few make it, vested or not.

If you're a private engineer eyeing government: several roles you'd recognize barely exist as federal jobs, and most that do want a clearance you can't get on your own.

For the government: the pipeline problem is deeper than pay. Even a fully-funded, competitively-paid hiring push would be hiring into role categories — analyst data science, compliance security — that don't match where technical talent is trained. Closing the pay gap wouldn't make a modern ML-engineering workforce appear. The job architecture has to change first.

---

Federal and private postings from the Skillenai job index (2026), both under the same relevance filter; the remote-work time series uses the open loyoladatamining/usajobs USAJOBS corpus (2022–2026), which reaches further back than our index. Skill measurement is raw-text phrase matching applied identically to both sides. Full methodology and code.