# How to Build the Skills a Defense Data Platform Job Asks For, and What It Pays

A Secret-level data engineer posting advertised $182,000 to $235,000 this week. Every skill on the job spec, Iceberg to lineage to GitOps, is free to practice at home.

Author: J.A. Watte
Published: July 18, 2026
Source: https://jwatte.com/blog/becoming-defense-data-platform-engineer/

---

A job description crossed my desk in mid-July 2026 for a defense data-platform role, and it is the clearest picture I have seen of where cleared data work is heading. The listing wants Apache Iceberg (or Delta or Hudi) on S3-compatible object storage, Postgres with pgvector, columnar engines like Trino, DuckDB and ClickHouse, change data capture with Debezium, REST catalogs from the Polaris and Gravitino and Nessie family, Kubernetes awareness, Terraform or Pulumi with Flux or ArgoCD, familiarity with the CNCF ecosystem including which projects are foundation-governed and which belong to a single vendor, and, above all, data provenance, lineage and governance. That last one carries a line most job specs never include: depth here will be weighted heavily. It also asks for something you cannot download: an active Secret clearance and familiarity with Air Force systems of record like MILPDS and ARMS.

Strip out the clearance and the personnel systems and here is the surprising part: every single technology on that list is open source, free to run, and small enough to practice on a laptop you already own. The same stack runs at Netflix, Shopify, Bloomberg, Cloudflare and the Pentagon's own Advana platform, so nothing you build at home is a toy version of the real thing. It is the real thing, minus the accreditation paperwork. This post maps the listing into five layers plus a defense context layer, and for each one gives you the learning sources I would actually use, a runnable test environment with the commands as they appear in the official docs today, and the organizations that publicly run it. Salaries, with sources, are at the end.

I read job specs this way because it is how I learned to build my own systems: my HomeStats project pulls Census, FRED, BEA, HUD and FBI data through one pipeline into 21,000+ pages, and Corvus folds 745+ news sources into a single corpus. Neither is a defense system, but both taught me the lesson this job spec is really testing: moving data is easy, and knowing where every row came from is the hard part employers pay for.

## The five layers, and one you cannot practice at home

The twelve requirements collapse into five buildable layers: storage (the lakehouse), movement (change data capture), query (the engines), trust (catalogs, lineage, access control), and the platform underneath it all (Linux, Kubernetes, infrastructure as code, GitOps). The sixth layer, defense context, is knowledge rather than tooling: what Advana is, why the Air Force's personnel data flows the way it does, and what a Secret clearance gates. You can read your way into the sixth layer, and I will point at the primary documents.

## 1. The lakehouse: Iceberg on object storage, plus Postgres

Apache Iceberg is a table format: it turns a pile of files in object storage into database-grade tables with transactions, schema evolution and time travel. It was created at Netflix and contributed to the Apache Software Foundation in 2018, and it has since become the closest thing data infrastructure has to a consensus: Snowflake made Iceberg tables generally available in June 2024, AWS launched S3 Tables in December 2024 as "the first cloud object store with built-in Apache Iceberg support," Google BigQuery offers managed Iceberg tables, and Microsoft Fabric serves Delta tables with virtual Iceberg metadata. Databricks, the company behind rival format Delta Lake, agreed in June 2024 to acquire Tabular, the startup founded by Iceberg's original creators. Adobe wrote up its migration of the Experience Platform data lake to Iceberg while processing about a million batches a day. When one format is supported by every major warehouse vendor at once, learning it is not a bet, it is table stakes.

**Learn it:** the [official Iceberg docs](https://iceberg.apache.org/docs/latest/) (version 1.11.0 as I write) plus the free digital copy of [*Apache Iceberg: The Definitive Guide*](https://www.dremio.com/guides/apache-iceberg-the-definitive-guide/) (Shiran, Hughes and Merced, O'Reilly, 2024) that Dremio gives away with no credit card. For the conceptual foundation underneath all of this, the second edition of [*Designing Data-Intensive Applications*](https://martin.kleppmann.com/2026/03/24/designing-data-intensive-applications-2e.html) by Martin Kleppmann and Chris Riccomini landed in March 2026, and [*Fundamentals of Data Engineering*](https://openlibrary.org/isbn/9781098108304.json) (Reis and Housley, 2022) is the field's course textbook.

**Test environment:** the [official Spark quickstart](https://iceberg.apache.org/spark-quickstart/) is a Docker Compose file with four services: a `tabulario/spark-iceberg` container with Spark and notebooks, an `apache/iceberg-rest-fixture` REST catalog on port 8181, MinIO as the S3-compatible object store, and an `mc` sidecar that creates the warehouse bucket. Then:

```bash
docker-compose up
docker exec -it spark-iceberg spark-sql
```

```sql
CREATE DATABASE IF NOT EXISTS demo.nyc;
CREATE TABLE demo.nyc.taxis (
  vendor_id bigint, trip_id bigint, trip_distance float,
  fare_amount double, store_and_fwd_flag string
) PARTITIONED BY (vendor_id);
INSERT INTO demo.nyc.taxis VALUES
  (1, 1000371, 1.8, 15.32, 'N'), (2, 1000372, 2.5, 22.15, 'N');
```

That is a genuine lakehouse on your laptop: object store, REST catalog, engine, ACID table. Poke at the `metadata/` folder MinIO now holds and you will understand Iceberg better than most people who only read about it.

The job spec also names Postgres with pgvector, which matters for the AI/ML awareness requirement: pgvector (version 0.8.5) adds vector similarity search to Postgres, it ships in Supabase, and Amazon RDS has supported it since May 2023 in every region including AWS GovCloud, which is exactly where defense workloads live. The lab is three lines once you pull the official image:

```bash
docker pull pgvector/pgvector:pg18-trixie
```

```sql
CREATE EXTENSION vector;
CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
```

For the object-storage layer itself, know Ceph by name: its Object Gateway speaks a large subset of the Amazon S3 API, and the Ceph Foundation sits under the Linux Foundation. On-prem defense environments that cannot use commercial S3 reach for exactly this.

## 2. Change data capture: Debezium

Change data capture is how a lakehouse stays current without hammering production databases: instead of polling, you tail the database's own transaction log and stream every insert, update and delete as an event. Debezium is the open-source standard here. Its public users page lists Reddit, Shopify, Strava, Ubisoft, Vimeo and Zalando among its production references, and its stewardship story is worth knowing on its own: Red Hat sponsored it exclusively from 2015, then in November 2024 the project moved to the vendor-neutral Commonhaus Foundation. The current stable series is 3.6, released July 1, 2026.

**Learn it:** the [official tutorial](https://debezium.io/documentation/reference/stable/kc-tutorial.html) end to end. One warning about stale guides elsewhere: the current tutorial runs Kafka as a single node in KRaft combined mode. There is no Zookeeper anywhere in it, so any walkthrough that starts a Zookeeper container is describing the past.

**Test environment:** three containers from the tutorial, verbatim from the current docs:

```bash
docker run -it --rm -p 9092:9092 --name kafka --hostname kafka \
  -e CLUSTER_ID=<YOUR_UNIQUE_CLUSTER_IDENTIFIER> -e NODE_ID=1 -e NODE_ROLE=combined \
  -e KAFKA_CONTROLLER_QUORUM_VOTERS=1@kafka:9093 \
  -e KAFKA_LISTENERS=PLAINTEXT://kafka:9092,CONTROLLER://kafka:9093 \
  -e KAFKA_ADVERTISED_LISTENERS=PLAINTEXT://kafka:9092 \
  quay.io/debezium/kafka:3.6

docker run -it --rm --name mysql -p 3306:3306 \
  -e MYSQL_ROOT_PASSWORD=debezium -e MYSQL_USER=mysqluser -e MYSQL_PASSWORD=mysqlpw \
  quay.io/debezium/example-mysql:3.6

docker run -it --rm --name connect -p 8083:8083 -e GROUP_ID=1 \
  -e CONFIG_STORAGE_TOPIC=my_connect_configs -e OFFSET_STORAGE_TOPIC=my_connect_offsets \
  -e STATUS_STORAGE_TOPIC=my_connect_statuses --link kafka:kafka --link mysql:mysql \
  quay.io/debezium/connect:3.6
```

Register the MySQL connector with one POST to port 8083, watch the change topic, then open a second terminal and UPDATE a row in MySQL. Watching your edit arrive as a structured before-and-after event is the moment CDC stops being a buzzword. The prove-it project: land those events into the Iceberg table from layer 1, and you have built the front half of a real ingestion pipeline.

## 3. Query engines: Trino, DuckDB, ClickHouse

The job spec names three engines because they cover three different jobs. Trino is the federated SQL engine you put in front of a lakehouse: it began at Facebook in 2012 as Presto (created by Martin Traverso, Dain Sundstrom, David Phillips and Eric Hwang), the creators left and forked it, and the fork became Trino in December 2020 after a trademark fight. Its users page lists Netflix, LinkedIn, Goldman Sachs, Stripe, Shopify, Lyft and Pinterest among roughly 64 organizations. DuckDB is the opposite: an in-process analytical database you embed anywhere, now at version 1.5.4, with a commercial cloud (MotherDuck) built on top of it. ClickHouse is the throughput monster: built at Yandex for its web-analytics platform, and famously run by Cloudflare, which wrote up handling 6 million HTTP requests per second of analytics on a 36-node cluster; its adopters page also lists Uber for logging, eBay and Microsoft Clarity.

**Learn them:** [Trino's docs](https://trino.io/docs/current/) (version 483), plus the free [*Trino: The Definitive Guide*, 2nd edition](https://www.starburst.io/info/oreilly-trino-guide/) from Starburst, with the honest caveat that the book is pinned to Trino 392 from 2022, so trust it for concepts and the current docs for specifics. DuckDB and ClickHouse are both learnable straight from their getting-started pages in an afternoon each.

**Test environment:** each engine is one command.

```bash
# Trino
docker run --name trino -d -p 8080:8080 trinodb/trino
docker exec -it trino trino

# DuckDB
curl https://install.duckdb.org | sh

# ClickHouse
docker run -d --name some-clickhouse-server \
  --ulimit nofile=262144:262144 clickhouse/clickhouse-server
docker exec -it some-clickhouse-server clickhouse-client
```

The DuckDB detail worth knowing in 2026: its `iceberg` extension is no longer read-only. `INSTALL iceberg; LOAD iceberg;` gives you `iceberg_scan()` for standalone tables, and attaching an Iceberg REST catalog with `ATTACH` now supports create, insert, update, delete, merge, schema evolution and time travel. That means the smallest possible lakehouse lab is DuckDB pointed at the REST catalog from layer 1, no Spark required. That sentence would have been false a year ago, which is exactly why this post quotes versions.

## 4. Catalogs, lineage and access: the part they weight heavily

Here is the layer the job spec flags as the differentiator, and it is the least glamorous and least practiced one. Provenance answers "where did this number come from"; lineage answers "what happens downstream if I change this table"; governance answers "who is allowed to see it, and can we prove it." In a defense setting these are not nice-to-haves, they are the difference between data someone will act on and data nobody may legally touch.

The catalog landscape reshuffled recently, so current status matters. The Iceberg REST catalog API is an OpenAPI spec anyone can implement, and the implementations are competing hard: Apache Polaris (open-sourced by Snowflake in mid-2024) graduated to an Apache top-level project in February 2026; Apache Gravitino, a federated "metadata lake," graduated in June 2025; Project Nessie, Dremio-originated, adds git-like branches and tags to catalog state; Databricks open-sourced Unity Catalog in June 2024 under Apache 2.0, and as of this month it sits at Sandbox stage in LF AI & Data; and Lakekeeper is a fast Rust implementation of the REST spec. You do not need to master all five. You need to be able to say what a REST catalog is, run one (you already did, in the Iceberg quickstart), and explain why five of them exist.

For lineage, the open standard is OpenLineage, an LF AI & Data graduated project that models every pipeline as datasets, jobs and runs, with integrations for Airflow, Spark, Flink, dbt and Trino. Marquez, also LF graduated, is its reference server. The metadata-catalog products a defense program might buy or build on, DataHub (born at LinkedIn, open-sourced in 2020) and OpenMetadata, both have free local quickstarts.

**Test environment:** lineage with a visual payoff in two commands:

```bash
git clone https://github.com/MarquezProject/marquez && cd marquez
./docker/up.sh --seed
```

The UI at `localhost:3000` renders a lineage graph of a seeded food-delivery pipeline: click any dataset and see every job that feeds it. DataHub is nearly as quick if you have 8 GB of RAM to give it: `datahub docker quickstart`, UI at `localhost:9002`. The prove-it project that would genuinely stand out in an interview for this role: wire OpenLineage events from the CDC pipeline you built in layer 2 into Marquez, then answer "where did this row come from" with a screenshot instead of a promise.

Access control rounds out the trust layer. The job spec says ABAC/RBAC and OIDC/OAuth, and the canonical free sources are NIST SP 800-162 (the government's own definition of attribute-based access control, where authorization is computed from subject, object, operation and environment attributes) and a hands-on hour with Keycloak, the identity server that is currently a CNCF incubating project:

```bash
docker run -p 127.0.0.1:8080:8080 \
  -e KC_BOOTSTRAP_ADMIN_USERNAME=admin -e KC_BOOTSTRAP_ADMIN_PASSWORD=admin \
  quay.io/keycloak/keycloak:26.7.0 start-dev
```

Create a realm, a user and a client, and you have issued yourself real OIDC tokens. Open Policy Agent, CNCF graduated since 2021, is the policy engine you would use to enforce ABAC-style rules in front of a catalog. If you want a formal credential for this layer, DAMA's CDMP exists ($311 per exam, DMBOK2-based), but for this role I would spend the money on the Kubernetes certs below and prove the governance skill with the Marquez lab instead.

## 5. The platform underneath: Linux, Kubernetes, IaC, GitOps

Everything above runs on the layer the spec calls "Kubernetes awareness" plus Linux fundamentals, and this is the layer with the clearest certification ladder. The Linux Foundation's [LFS101 Introduction to Linux](https://training.linuxfoundation.org/training/introduction-to-linux/) is free and covers roughly 60 hours of command line, scripting and networking. For Kubernetes, KCNA ($250, 90-minute multiple choice) is the awareness-level cert that matches this job spec's language, and CKA ($445, two-hour hands-on exam, currently on Kubernetes v1.35) is the operator-level one; note the CKA was revised in February 2025 to add Gateway API, Helm and Kustomize, so study materials older than that are missing content. Both include two attempts and are valid two years, and the CKA bundles two sessions of the killer.sh exam simulator. When you want to understand what the managed clusters are hiding, Kelsey Hightower's [Kubernetes the Hard Way](https://github.com/kelseyhightower/kubernetes-the-hard-way) walks you through assembling a cluster from parts.

**Test environment:** a real cluster on your laptop in about a minute, then GitOps on top of it:

```bash
# kind: Kubernetes in Docker (v0.32.0)
curl -Lo ./kind https://kind.sigs.k8s.io/dl/v0.32.0/kind-linux-amd64
chmod +x ./kind && sudo mv ./kind /usr/local/bin/kind
kind create cluster

# Argo CD, straight from the current getting-started docs
kubectl create namespace argocd
kubectl apply -n argocd --server-side --force-conflicts \
  -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml
kubectl port-forward svc/argocd-server -n argocd 8080:443
argocd admin initial-password -n argocd
```

(If you have a spare Linux box or VM, `curl -sfL https://get.k3s.io | sh -` gives you k3s instead, in about thirty seconds.) Point Argo CD at a Git repo of manifests, change a file, and watch the cluster converge on it: that loop is the entire GitOps idea, and Flux (`flux bootstrap github ...`, docs at version 2.9) is the same idea with a different shape. The Linux Foundation certifies this layer too: Certified GitOps Associate and Certified Argo Project Associate, $250 each.

For infrastructure as code, HashiCorp's own Docker get-started tutorial is the gentlest on-ramp: a dozen lines of HCL with the `kreuzwerker/docker` provider, `terraform init`, `terraform apply`, and nginx appears at `localhost:8000` because a text file said so. The Terraform Associate exam is now version 004 and costs $70.50. Pulumi, the spec's alternative, is free and open source with get-started guides for every major cloud, and pairs optionally with Pulumi Cloud, which has a free individual tier. And know OpenTofu (currently 1.12): why it exists is the next section, and this job spec is quietly testing whether you know.

## Foundation projects versus single-vendor projects

The strangest line in the job description is also my favorite: familiarity with the CNCF ecosystem, "including the distinction between foundation projects and single-vendor projects." That is not trivia. It is risk management, and 2023 to 2026 handed us three case studies.

Case one: on August 10, 2023, HashiCorp switched future Terraform releases from the open MPL 2.0 license to the Business Source License. Six weeks later the Linux Foundation announced OpenTofu, a community fork of the last open release, backed by 140+ organizations. Every team that had bet on Terraform-the-open-project woke up owning a dependency on Terraform-the-product.

Case two: MinIO, the S3-compatible store that half the tutorials on the internet (including the official Iceberg quickstart) use as their object layer. Its community GitHub repo now opens with the banner "THIS REPOSITORY IS NO LONGER MAINTAINED," the AGPLv3 community edition is distributed as source only, and the official docs describe an enterprise container that requires a license file. The quickstart still runs today because the old Docker Hub image still exists. But a platform engineer who chose MinIO in 2023 is doing migration math in 2026.

Case three runs the other direction: Debezium spent nine years as a Red Hat-sponsored project, then moved to the vendor-neutral Commonhaus Foundation in November 2024, and OpenLineage and Marquez live under LF AI & Data. Meanwhile DataHub's old project domain now redirects to a commercial site, and OpenMetadata's trademark belongs to a company, even though the code is Apache 2.0. None of that makes any of these projects bad. It makes them different risk profiles, and the interview question hiding in the job spec is: can you tell which is which before you build a defense program on one? The tell is governance, not license: CNCF describes its mission as sustaining "vendor-neutral projects," and the Apache Way states that no organization can gain special privileges or control a project's direction. When a single company owns the trademark, the roadmap and the only paid support channel, plan for the day its incentives change.

## Where this stack already runs in defense

The defense context layer is readable in public documents, and reading them is the cheapest interview preparation available.

Start with the Deputy Secretary of Defense's May 5, 2021 "Creating Data Advantage" memo, which designates Advana as "the single enterprise authoritative data management and analytics platform" for DoD and orders components to publish data assets in a federated catalog and make data visible, accessible, understandable, linked, trustworthy, interoperable and secure (yes, that spells VAULTIS, and yes, people say it out loud). By March 2023, sworn congressional testimony from the DoD's Chief Digital and AI Officer put Advana at more than 390 connected data sources, almost two petabytes, 31,000+ users across 190+ organizations, and more than $11 billion in savings attributed to improved auditability. The CDAO itself was created in February 2022 by merging the DoD Chief Data Officer, the Joint AI Center, the Defense Digital Service and the Advana office. When the job spec asks for "provenance, weighted heavily," this is why: the department's flagship analytics platform exists in large part to survive audits.

On the Air Force and Space Force side, three names to know. The Unified Data Library is Space Systems Command's cloud repository for space-domain-awareness data, hosting more than 300 data types under a program office managing a roughly $4 billion portfolio. Platform One is the Air Force's DevSecOps software factory, and Iron Bank, its hardened container registry, describes itself as "the DoD's supply chain for mission critical software"; if you have practiced pulling vetted images and deploying through GitOps, you have practiced the civilian version of exactly this. Kessel Run, formally Air Force Life Cycle Management Center Detachment 12, runs air-operations software under the DoD's software acquisition pathway. And the AI churn is real here too: NIPRGPT, the Air Force Research Laboratory's experimental chatbot that reached 700,000+ DoD users, was sunset on December 31, 2025 in favor of the Pentagon-wide GenAI.mil.

The two systems of record named in the listing are personnel systems, and their public paper trail teaches a provenance lesson all by itself. MILPDS, the Military Personnel Data System, is the system of record for all Air Force military personnel actions, a commercial Oracle EBS-based product hosted in a DISA data center, and its published privacy impact assessment lists about 31 system-to-system interfaces exchanging personnel data. ARMS, the Automated Records Management System, is where the Air Force consolidated its personnel records, with scanning centralized at the Air Force Personnel Center at Randolph. Here is the kicker: MILPDS's own interface list contains both the "Automated Records Management System Legacy Conversion (ARMS LC)" and a completely different "Aviation Resource Management System (ARMS)." The Air Force runs two systems with the same acronym, and both touch the same personnel database. If you ever wanted a one-sentence answer to "why does lineage tooling matter," it is that a job description can cite ARMS and an engineer's first task is figuring out which one.

The clearance requirement is the one part of this you cannot lab your way into: an active Secret clearance requires sponsorship by an employer with a contract that needs it. What you can do is understand what it gates and what it is worth, which brings us to money.

## What the jobs pay

Every figure here is from a source I fetched and checked this week; links are in the fact-check section, and the survey years differ, so read them as anchors rather than one number.

Government statistics first. The Bureau of Labor Statistics does not have a "data engineer" occupation code, but its 2025 occupational wage data brackets the role: database architects, the closest match, show a median of $139,500 (mean $144,440) across 67,140 jobs nationally; data scientists median $120,230; software developers median $135,980 across about 1.69 million jobs. Market surveys sit around and above that: Dice's 2025 Tech Salary Report puts the average US tech salary at $112,521 (2024 data) with a 17.7 percent premium for professionals doing AI work; Robert Half's 2026 guide projects data engineer starting salaries from $127,000 (limited experience) through $156,250 (mid) to $180,750 (extensive); and levels.fyi, which skews toward large tech companies, shows a US data engineer median total compensation of $160,000 across 1,243 reports, with the 75th percentile at $222,500.

Now the clearance premium, which is the reason this particular job spec pays attention. ClearanceJobs' 2026 compensation report (2025 data) puts average total compensation for cleared professionals at a record $126,125, up nearly 6 percent in a year. The ladder inside that average is the interesting part: DoD Secret averages $107,439, Top Secret $126,839, TS/SCI $139,261, and polygraph-backed roles reach $149,875, roughly $30,000 above cleared peers without a polygraph. Geography stacks on top: cleared professionals average about $139,000 in Maryland and $138,700 in Virginia. And specialized data roles clear these averages easily: on the day I wrote this, live cleared data-engineer postings advertised bands like $119,900 to $179,800 (TS/SCI, KBR), $122,000 to $253,000 (TS/SCI with polygraph, Nightwing), and one Secret-level data engineering role in Columbia, Maryland at $182,000 to $235,000. Those are point-in-time postings, not statistics, but they show the ceiling: a Secret clearance plus exactly the lakehouse-and-lineage stack in this post is currently pricing at roughly double the average Secret-holder's package.

One honest caveat about that arithmetic: the cleared average and the all-tech average come from different surveys in different years, so treat the comparison as directional. The direction, though, is unambiguous, and it is the same one I wrote about in [The W-2 Trap](https://thew2trap.com/): a bigger salary is the beginning of a financial plan, not the end of one. The skills ladder in this post is how you raise the number; what you do with the raise is a different book.

## The rank translation: what those salaries equal in uniform

Readers coming out of the military asked the natural next question: when you count the benefits, what rank do these civilian packages actually equal? There is an official way to answer it. Regular Military Compensation, defined in 37 U.S.C. 101(25), is basic pay plus the housing allowance (BAH) plus the subsistence allowance (BAS) plus the federal tax advantage that comes from BAH and BAS being excluded from gross income (IRS Publication 3 confirms both are untaxed). DoD publishes average RMC by grade, and the January 1, 2026 figures reproduced by the Congressional Research Service run: E-1 $60,810, E-5 $89,148, E-8 $129,715, O-1 $85,567, O-4 $172,477, O-6 $237,978.

To fill in the whole ladder, I computed the 2026 cash package for a service member with dependents stationed in the Washington DC metro area, the same corridor where the cleared salaries above cluster, using the DFAS 2026 pay table, the official 2026 BAH file (DC053 rates), and 2026 BAS. These figures are my arithmetic from those three published tables, and they understate the uniformed side because they exclude the federal tax advantage:

| Grade (years of service) | 2026 DC-metro cash | DoD average RMC (2026) |
|---|---|---|
| E-5 (over 6) | $92,600 | $89,148 |
| O-1 (over 2) | $94,300 | $85,567 |
| E-6 (over 10) | $107,900 | |
| E-7 (over 14) | $122,000 | |
| O-2 (over 4) | $126,800 | |
| E-8 (over 20) | $137,200 | $129,715 |
| W-3 (over 14) | $142,800 | |
| O-3 (over 8) | $149,700 | |
| E-9 (over 24) | $160,300 | |
| O-4 (over 14) | $179,400 | $172,477 |
| O-5 (over 20) | $204,600 | |
| O-6 (over 24) | $234,500 | $237,978 |

Now lay the civilian anchors from the previous section against that ladder. The average Secret-clearance holder's $107,439 is an E-6 with ten years in. Dice's all-tech average of $112,521 sits between E-6 and E-7. The cleared-workforce average of $126,125 is an O-2 who just pinned on, or a senior E-7. The TS/SCI average of $139,261 and the BLS database-architect median of $139,500 are both an E-8 over 20, edging toward W-3 money. Robert Half's mid-level data engineer at $156,250 lands between an O-3 and an E-9; the levels.fyi median of $160,000 is almost exactly an E-9 with 24 years. Robert Half's experienced band at $180,750 and the top of the KBR posting at $179,800 are an O-4 over 14. And that Secret-level Columbia posting at $182,000 to $235,000 spans O-4 to O-6: colonel pay, at the command-a-brigade point of a 24-year career, for a data engineering job that requires the same clearance a brand-new E-3 can hold. The junior-officer rungs matter just as much at the other end: the $80,000 to $90,000 entry cleared data roles are O-1 and E-5 money, which means a first cleared data job pays about what a lieutenant makes, and the civilian ladder climbs from there far faster than promotion boards do.

Three things the table hides, all of which favor the uniformed column. First, retirement: under the Blended Retirement System the government contributes up to 5 percent of basic pay to the TSP (1 percent automatic plus 4 percent matching) and pays a pension of 2 percent times years served times the highest-36-months basic pay, with an inflation COLA, for anyone who reaches the 20-year cliff, plus a mid-career continuation-pay bonus of 2.5 to 13 times monthly basic pay between years 8 and 12. A civilian data engineer's 401(k) match is portable from year one; the pension needs two decades. Second, healthcare: active-duty families pay no Tricare Prime enrollment fees, while the average employer family premium hit $26,993 in 2025 with the worker paying $6,850 of it, the invisible wedge I covered in [the COLA versus W-2 wages piece](/blog/cola-vs-w2-wages/). Third, the raise itself is indexed: military basic pay rose 3.8 percent in 2026 by statute, not by a manager's mood, which is the whole thesis of [the written-raise article](/blog/public-sector-union-jobs-beat-inflation/). Bonuses can also stack on the uniformed side (the Air Force's retention bonus for cyber fields runs up to $180,000 per contract with a $360,000 career cap), while the allowances most people ask about are narrower than folklore says: CONUS COLA is taxable and tiny (about 127,000 members share roughly $99 million in 2026, and only 18 housing areas qualify), and the tax-free overseas versions, OCONUS COLA and OHA, apply only while stationed abroad.

## The disability stack: the pay most job seekers forget to count

There is a second income stream that a large share of prior-service readers will carry into any of these jobs, and almost no salary comparison counts it. VA disability compensation is common among recent-era veterans: VA's own FY2025 benefits report shows 4,004,299 of 9,111,600 Gulf War era veterans, 43.9 percent of everyone who served from August 1990 on, receiving compensation, and CBO puts it at 40 percent of post-9/11 veterans (across all living veterans it is about 36 percent). The ratings skew high, too: the single most common rating today is 100 percent (29.1 percent of recipients), a majority are rated 70 percent or higher, and the 30-to-60 band covers another 22.1 percent.

The numbers, at the rates effective December 1, 2025 for a veteran with no dependents: 30 percent pays $552.47 a month, 60 percent pays $1,435.02, and 100 percent pays $3,938.58. All of it is tax-free under 38 U.S.C. 5301 and none of it is reduced by your salary. CBO states it flatly: recipients "receive the same disability compensation regardless of whether they work or what their income is," and most working-age recipients are employed. This deserves saying plainly, because people get it wrong constantly: a VA disability rating, including a schedular 100 percent, is not a statement that you cannot work, and it carries no earnings limit whatsoever. The one exception is Individual Unemployability, which pays at the 100 percent rate precisely because the veteran cannot maintain substantially gainful employment, defined in 38 CFR 4.16 by the one-person poverty threshold. Do not confuse any of this with Social Security disability, which does restrict work ($1,690 a month is the 2026 substantial-earnings line). And if you are a 20-year retiree rather than a separatee, the retired-pay offset and its CRDP and CRSC exceptions apply; a veteran who separated short of retirement simply receives the compensation on top of a paycheck, with the narrow caveat that separation or severance pay gets recouped first.

So redo the math for the 43.9 percent. A 60 percent rating adds $17,220 a year of tax-free income, worth about $22,000 pre-tax at a 22 percent marginal rate (that gross-up is my arithmetic). Stack it on the TS/SCI average and $139,261 becomes roughly $161,000 in effective compensation, an E-9-to-O-4 jump in the table above. At 100 percent the stack adds $47,263 tax-free, about $60,600 pre-tax equivalent: the person who takes that $235,000 Columbia job with a 100 percent rating is functionally at $295,000, which is beyond the O-6 row and into general-officer territory, before counting a single state benefit.

## The portable benefits: the VA loan and the GI Bill

Two more parts of the package travel with the veteran into any of these jobs and never show up on a salary line. The VA home loan needs zero down payment and no private mortgage insurance, and since January 2020 it carries no loan limit for a borrower with full entitlement; nearly 90 percent of VA loans are made with nothing down ([VA](https://www.va.gov/housing-assistance/home-loans/)). The piece that stacks with a disability rating: at any 10 percent or higher service-connected rating the VA funding fee is waived entirely, so the 2.15 percent first-use fee other VA borrowers pay, about $8,600 on a $400,000 loan, simply disappears ([VA](https://www.va.gov/housing-assistance/home-loans/funding-fee-and-closing-costs/)).

The Post-9/11 GI Bill is the education half, and for a data engineer it is the cheapest path to the master's or the cloud certifications the senior version of this role rewards, with no student debt attached. It pays full tuition and fees at a public in-state school (up to $29,920.95 a year at private or foreign schools for 2025-2026), a monthly housing allowance set at the E-5-with-dependents rate for the school's ZIP, and up to $1,000 a year for books, across 36 months ([VA rates](https://www.va.gov/education/benefit-rates/post-9-11-gi-bill-rates/)). Third parties commonly value the full benefit north of $100,000, and it is transferable to a spouse or children after six years of service, so it can fund the family's degrees instead of the veteran's.

## The 100 percent example: Florida, and the states with no property tax

State benefits are where a 100 percent Permanent and Total rating compounds hardest, and Florida is the worked example because its stack is statute-deep. Some in Florida call the result, only half as a joke, "super citizen" status; whatever you call it, every line below is a citation, not folklore. In Florida a 100 percent P&T veteran pays no property tax on their homestead at all (F.S. 196.081, and the exemption survives to an un-remarried surviving spouse), gets one free DV license plate with registration (F.S. 320.084, minus a few dollars of service charges the statute preserves), pays no driver-license fees (F.S. 322.21, against the standard $48), holds free hunting, freshwater and saltwater fishing licenses (F.S. 379.353, extended in 2024 to any 50-percent-plus rating), and carries a free lifetime family pass to every state park (F.S. 258.0145, any service-connected rating). Florida also levies no personal income tax, and its constitution effectively bars one, so neither the VA compensation nor the $235,000 salary gives the state a cut. One myth to skip: the Florida toll waiver is not a general disabled-veteran perk; it requires a specially equipped vehicle and certified upper-limb impairment.

Florida is not alone. Verified against statutes and official state pages, a 100 percent P&T veteran's primary residence is fully exempt from property tax with no income cap in at least Florida, Texas (Tax Code 11.131, which requires both 100 percent compensation and a 100 percent or IU rating), Alabama, New Jersey, Virginia, Michigan (including IU ratings), Oklahoma (plus a sales-tax exemption capped at $25,000 of purchases a year), Maryland, and Nebraska, with Honolulu exempting everything except a $300 minimum tax. Several more states appear on consolidated lists but with caps or means tests (Illinois caps home value, Pennsylvania tests income), and New Hampshire's total exemption applies only to specially adapted homes, so read your own state's fine print. Texas and Florida are the headline pair because they combine the full exemption with zero state income tax. On a $500,000 Florida or Texas home at a typical 1.5 to 2 percent effective rate, the exemption alone is worth $7,500 to $10,000 a year, after-tax, forever, on top of everything above. My [state-by-state veteran housing table at HomeStats](https://homestats.app/veteran-housing-by-state/) ranks all fifty states on exactly this stack: VA-loan math on the median home, exemption tier, registration and permit waivers, and income-tax status.

Two portable levers widen the geography play past the 100 percent case. Retired military pay is now untaxed in roughly 38 states, the 29 income-tax states that fully exempt it plus the 9 with no income tax at all ([military.com](https://www.military.com/benefits/military-pay/state-retirement-income-tax.html)), so where a veteran retires materially changes the after-tax pension. And the VA loan is assumable: under [38 U.S.C. 3714](https://www.law.cornell.edu/uscode/text/38/3714) a creditworthy buyer can take over the veteran's existing loan and its interest rate with VA approval, which in a high-rate market turns a low-rate VA mortgage into a resale asset a conventional loan cannot match. Partial property-tax exemptions also scale below 100 percent, so the state benefit starts well before P&T: Texas, for one, gives fixed exemptions of $5,000 to $12,000 at ratings from 10 to 90 percent ([Tax Code 11.22](https://statutes.capitol.texas.gov/Docs/TX/htm/TX.11.htm)).

At a 100 percent Permanent and Total rating the package widens past the veteran to the family. CHAMPVA covers a spouse and children who are not eligible for Tricare, with no premium, a $50-per-person deductible, and a $3,000 household cap on out-of-pocket costs ([VA](https://www.va.gov/health-care/family-caregiver-benefits/champva/)); the Chapter 35 benefit pays those same dependents about $1,574 a month for full-time college, 36 months of it ([VA](https://www.va.gov/education/survivor-dependent-benefits/dependents-education-assistance/)), and many states pile their own free dependent tuition on top. One gate is wider than folklore says: base commissary and exchange shopping, sales-tax-free and about 25 percent below grocery prices, extends to any service-connected rating since January 2020, not only 100 percent ([VA](https://news.va.gov/67974/commissary-military-service-exchange-mwr-access-extended-veterans-beginning-january/)); Space-Available military flights, by contrast, do require the permanent 100 percent rating. And 100 percent is not even the ceiling: Special Monthly Compensation pays above it for aid-and-attendance needs ([VA](https://www.va.gov/disability/compensation-rates/special-monthly-compensation-rates/)).

The honest closing caveat: the clearance and the rating are different things, earned different ways, and none of this section is career advice to chase a rating. It is accounting. Two engineers can sit in the same SCIF doing the same job at the same $139,261, and the one who left service with a 60 percent rating and a Florida homestead is $25,000 to $30,000 a year ahead in effective compensation. A salary comparison that ignores that is not describing the cleared workforce as it actually exists, in which nearly half of recent-era veterans carry the second income stream. It is also only fair to name the price behind these numbers: under the legacy system only about one in five who serve reach the twenty-year pension, and the disability compensation that anchors the stack is payment for real and lasting harm, measured in a veteran-suicide toll the VA still reports at more than 6,000 a year ([Military OneSource](https://www.militaryonesource.mil/resources/millife-guides/blended-retirement-system/); [VA 2024 report](https://news.va.gov/137221/va-2024-suicide-prevention-annual-report/)). This is accounting, not a recruiting brochure.

## How I would sequence it

Run one lab at a time and a 16 GB laptop handles all of this; none of the official quickstarts states a hard memory requirement, and only the Spark-based Iceberg compose and DataHub (which asks for 8 GB) are heavyweights, so that sizing is my assessment rather than a vendor's promise. Weekend one: the Iceberg quickstart, until the metadata files make sense. Weekend two: Debezium, ending with change events landing in your Iceberg table. Weekend three: Trino and DuckDB over that same table, because querying one copy of the data with two engines is the lakehouse pitch made tangible. Weekend four: Marquez, wiring lineage through the pipeline you now own. Weekends five and six: kind, Argo CD and the Terraform tutorial, then re-deploy your pipeline the GitOps way. That is six weekends to a portfolio project that answers every technical line on a real defense job description, for the cost of electricity, with certs (KCNA or CKA, Terraform Associate) as optional receipts on top. Then read the Advana memo and the MILPDS privacy assessment on weekend seven, and you can talk about the sixth layer too.

## Related reading

- [How to build the skills behind a lead AI/ML platform engineer](/blog/becoming-ai-ml-platform-engineer/). The same job-spec-to-capability-map read for the agentic AI and model-serving side of the house.
- [The expert's trap: when knowing fifteen things is the depreciating asset](/blog/the-experts-trap-depreciating-knowledge/). Why this whole skill stack depreciates on three fronts at once, and what indexed and owned assets to bolt underneath it.
- [What $200k AI jobs actually ask for, and how to practice every skill free](/blog/blog-ai-ml-github-projects-200k-jobs/). The pay-band view across cleared and commercial postings, project by project.
- [Modern integration environments: CI/CD with Jenkins, Kubernetes, and the glue between](/blog/modern-cicd-kubernetes-jenkins/). A deeper pass on the platform layer this post compresses into weekend five.
- [Streaming pipelines and platform engineering](/blog/streaming-and-platform-engineering/). Kafka and Flink fundamentals, which is the substrate Debezium rides on.
- [One playbook, many starting hands: set-asides in government contracting](/blog/cleared-veteran-disabled-minority-govcon-playbook/). What the clearance economy looks like from the business-owner side instead of the W-2 side.

## Fact-check notes and sources

- **Job description:** a real listing shared with me in July 2026; the employer is deliberately not named, and requirement wording is paraphrased.
- **Salaries (government):** BLS Occupational Employment and Wage Statistics, 2025 data, retrieved 2026-07-18 via the [BLS Public Data API](https://www.bls.gov/developers/) for [Database Architects 15-1243](https://www.bls.gov/oes/current/oes151243.htm) (median $139,500, mean $144,440, employment 67,140), [Data Scientists 15-2051](https://www.bls.gov/oes/current/oes152051.htm) (median $120,230) and [Software Developers 15-1252](https://www.bls.gov/oes/current/oes151252.htm) (median $135,980, employment 1,687,890).
- **Salaries (surveys):** [Dice 2025 Tech Salary Report](https://www.dice.com/technologists/ebooks/tech-salary-report/salary-trends.html) ($112,521 average, 17.7% AI premium; 2,835 respondents surveyed Aug-Nov 2024); [Robert Half data engineer guide](https://www.roberthalf.com/us/en/job-details/data-engineer) ($127,000 / $156,250 / $180,750); [levels.fyi US data engineer](https://www.levels.fyi/t/software-engineer/title/data-engineer) (median total comp $160,000, n=1,243, page updated 2026-07-18).
- **Clearance premium:** ClearanceJobs 2026 Compensation Report coverage: [record $126,125 average](https://news.clearancejobs.com/2026/03/31/new-report-reveals-security-clearance-compensation-climbs-to-record-high-amid-federal-workforce-uncertainty/) and [by clearance level and polygraph](https://news.clearancejobs.com/2026/05/13/from-secret-to-ts-sci-to-full-scope-poly-how-security-clearances-impact-compensation/) (Secret $107,439, TS $126,839, TS/SCI $139,261, full-scope poly $149,875 vs $118,680 no-poly). Posting bands from the live [ClearanceJobs data engineer search](https://www.clearancejobs.com/q-data-engineer), snapshot 2026-07-18.
- **Iceberg and lakehouse:** [ASF project spotlight](https://news.apache.org/foundation/entry/asf-project-spotlight-apache-iceberg) (Netflix origin, 2018 donation); [Databricks-Tabular announcement](https://www.databricks.com/company/newsroom/press-releases/databricks-agrees-acquire-tabular-company-founded-original-creators) (June 4, 2024, no price disclosed); [Snowflake Iceberg GA](https://docs.snowflake.com/en/release-notes/2024/other/2024-06-10-iceberg-tables); [Amazon S3 Tables](https://aws.amazon.com/about-aws/whats-new/2024/12/amazon-s3-tables-apache-iceberg-tables-analytics-workloads/); [BigQuery Iceberg tables](https://docs.cloud.google.com/bigquery/docs/iceberg-tables); [Fabric OneLake Iceberg](https://learn.microsoft.com/en-us/fabric/onelake/onelake-iceberg-tables); [Iceberg at Adobe](https://medium.com/adobetech/iceberg-at-adobe-88cf1950e866); [Iceberg vendors page](https://iceberg.apache.org/vendors/); [Spark quickstart](https://iceberg.apache.org/spark-quickstart/); [pgvector](https://github.com/pgvector/pgvector); [RDS pgvector announcement](https://aws.amazon.com/about-aws/whats-new/2023/05/amazon-rds-postgresql-pgvector-ml-model-integration/); [Ceph Object Gateway](https://docs.ceph.com/en/latest/radosgw/) and [Ceph Foundation](https://ceph.io/en/foundation/).
- **CDC:** [Debezium users page](https://debezium.io/community/users/); [Commonhaus move](https://debezium.io/blog/2024/11/04/debezium-moving-to-commonhaus/); [releases page](https://debezium.io/releases/) (3.6 stable, 2026-07-01); [current tutorial](https://debezium.io/documentation/reference/stable/kc-tutorial.html) (all commands quoted from this page as fetched 2026-07-18).
- **Engines:** [Trino rename announcement](https://trino.io/blog/2020/12/27/announcing-trino.html) (Presto history, four creators); [Trino users](https://trino.io/users.html); [Trino container docs](https://trino.io/docs/current/installation/containers.html); [Trino Definitive Guide 2e via Starburst](https://www.starburst.io/info/oreilly-trino-guide/) (pinned to Trino 392); [DuckDB install](https://duckdb.org/install/) (1.5.4); [DuckDB iceberg extension](https://duckdb.org/docs/current/core_extensions/iceberg/overview.html) (REST-catalog read/write); [MotherDuck](https://motherduck.com/product/); [ClickHouse history](https://clickhouse.com/docs/about-us/history); [ClickHouse Docker](https://clickhouse.com/docs/install/docker); [Cloudflare's ClickHouse post](https://blog.cloudflare.com/http-analytics-for-6m-requests-per-second-using-clickhouse/); [ClickHouse adopters](https://clickhouse.com/docs/about-us/adopters) (self-described as compiled from public sources).
- **Catalogs, lineage, access:** [Iceberg REST OpenAPI spec](https://raw.githubusercontent.com/apache/iceberg/main/open-api/rest-catalog-open-api.yaml); [Polaris TLP graduation](https://polaris.apache.org/blog/2026/02/19/apache-polaris-graduates-to-top-level-project/) and [Snowflake's post](https://www.snowflake.com/en/blog/apache-polaris-top-level-project/); [Gravitino TLP announcement](https://news.apache.org/foundation/entry/the-apache-software-foundation-announces-two-new-top-level-projects) (June 3, 2025); [Project Nessie](https://projectnessie.org/); [Unity Catalog open-sourcing](https://www.databricks.com/blog/open-sourcing-unity-catalog) and [LF AI & Data project listing](https://lfaidata.foundation/projects/) (Sandbox as of 2026-07-18; OpenLineage and Marquez Graduated); [Lakekeeper](https://docs.lakekeeper.io/); [OpenLineage](https://openlineage.io/docs); [Marquez quickstart](https://marquezproject.ai/docs/quickstart); [DataHub quickstart](https://docs.datahub.com/docs/quickstart); [DataHub's LinkedIn origin](https://www.linkedin.com/blog/engineering/open-source/open-sourcing-datahub-linkedins-metadata-search-and-discovery-p); [OpenMetadata](https://open-metadata.org/); [NIST SP 800-162](https://csrc.nist.gov/pubs/sp/800/162/upd2/final); [OPA graduation](https://www.cncf.io/announcements/2021/02/04/cloud-native-computing-foundation-announces-open-policy-agent-graduation/); [Keycloak at CNCF](https://www.cncf.io/projects/keycloak/) (incubating) and [Keycloak Docker guide](https://www.keycloak.org/getting-started/getting-started-docker) (26.7.0); [CDMP exams](https://cdmp.info/exams/) ($311).
- **Platform layer:** [LFS101](https://training.linuxfoundation.org/training/introduction-to-linux/) (free); [KCNA](https://training.linuxfoundation.org/certification/kubernetes-cloud-native-associate/) ($250); [CKA](https://training.linuxfoundation.org/certification/certified-kubernetes-administrator-cka/) ($445, Kubernetes v1.35) and [Feb 18, 2025 exam changes](https://training.linuxfoundation.org/certified-kubernetes-administrator-cka-program-changes/); [Kubernetes the Hard Way](https://github.com/kelseyhightower/kubernetes-the-hard-way); [kind quick start](https://kind.sigs.k8s.io/docs/user/quick-start/) (v0.32.0); [k3s](https://k3s.io/); [Argo CD getting started](https://argo-cd.readthedocs.io/en/stable/getting_started/); [Flux install](https://fluxcd.io/flux/installation/) and [bootstrap](https://fluxcd.io/flux/installation/bootstrap/github/); [CGOA](https://training.linuxfoundation.org/certification/certified-gitops-associate-cgoa/) and [CAPA](https://training.linuxfoundation.org/certification/certified-argo-project-associate-capa/) ($250 each); [Terraform Docker tutorial](https://developer.hashicorp.com/terraform/tutorials/docker-get-started/docker-build); [HashiCorp certifications](https://developer.hashicorp.com/certifications/infrastructure-automation) (Associate 004, $70.50); [Pulumi get started](https://www.pulumi.com/docs/iac/get-started/); [OpenTofu docs](https://opentofu.org/docs/) (1.12).
- **Foundation vs vendor:** [HashiCorp BUSL announcement](https://www.hashicorp.com/blog/hashicorp-adopts-business-source-license) (Aug 10, 2023); [Linux Foundation OpenTofu announcement](https://www.linuxfoundation.org/press/announcing-opentofu) (Sep 20, 2023, 140+ organizations); [MinIO community repo](https://github.com/minio/minio) ("no longer maintained" banner, AGPLv3 source-only) and [AIStor container docs](https://docs.min.io/aistor/installation/container/install/) (license file required); [CNCF who we are](https://www.cncf.io/about/who-we-are/) (vendor-neutral mission); [The Apache Way](https://www.apache.org/theapacheway/); [CNCF projects list](https://www.cncf.io/projects/) (maturity levels as of 2026-07-18).
- **Defense platforms and systems of record:** [Deputy SecDef "Creating Data Advantage" memo, May 5, 2021](https://media.defense.gov/2021/May/10/2002638551/-1/-1/0/DEPUTY-SECRETARY-OF-DEFENSE-MEMORANDUM.PDF) (Advana designation, data decrees, VAULTIS); [CDAO testimony to HASC, March 9, 2023](https://democrats-armedservices.house.gov/_cache/files/5/8/5859348e-f628-42d3-b1ca-30a6cc588df9/E26568F90E61B18EDE9F8CCAB9E9D18E.martell-testimony.pdf) (390+ sources, ~2 PB, 31,000+ users, $11B, CDAO's February 2022 formation); [SSC on the Unified Data Library](https://www.ssc.spaceforce.mil/Newsroom/Article/4039108/space-systems-commands-udl-provides-data-solutions-at-the-speed-of-battle) (300+ data types); [Platform One Big Bang docs](https://docs-bigbang.dso.mil/latest/bigbang-training/docs/1_bigbang-101/p1-introduction/) and [Iron Bank docs](https://docs-ironbank.dso.mil/); [Kessel Run strategy approval](https://www.aflcmc.af.mil/NEWS/Article/3035345/kessel-runs-software-pathway-strategy-approved/) (AFLCMC Det 12); [NIPRGPT sunset](https://defensescoop.com/2025/12/18/air-force-sunsetting-niprgpt-generative-ai-platform/) (Dec 31, 2025, GenAI.mil); [MILPDS privacy impact assessment](https://www.privacy.af.mil/Portals/26/Miliraty%20Personnel%20Data%20System%20(MILPDS).pdf) (system of record, Oracle EBS-based, interface list including both ARMS entries); [ARMS/PRDA](https://www.af.mil/News/Article-Display/Article/467187/airmen-directed-to-visit-prda-to-view-personnel-records/) and [ARMS scanning consolidation](https://www.arpc.afrc.af.mil/News/Article-Display/Article/567337/arms-document-processing/). Several .mil pages block automated fetching; content was verified via the Internet Archive's snapshots of the same URLs.
- **Military pay and RMC (rank translation):** RMC definition at [37 U.S.C. 101(25)](https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title37-section101&num=0&edition=prelim); average RMC by grade as of Jan 1, 2026 from [CRS Defense Primer IF10532](https://www.everycrsreport.com/reports/IF10532.html) (updated July 7, 2026, reproducing DoD's Selected Military Compensation Tables); [DoD Greenbook](https://militarypay.defense.gov/Portals/3/GreenBook%202023.pdf) for the RMC methodology; 2026 basic pay cells from the DFAS tables ([enlisted](https://www.dfas.mil/MilitaryMembers/payentitlements/Pay-Tables/Basic-Pay/EM/), [warrant](https://www.dfas.mil/MilitaryMembers/payentitlements/Pay-Tables/Basic-Pay/WO/), [officer](https://www.dfas.mil/MilitaryMembers/payentitlements/Pay-Tables/Basic-Pay/CO/)); [2026 BAS](https://www.dfas.mil/MilitaryMembers/payentitlements/Pay-Tables/bas/) ($476.95 enlisted / $328.48 officer); DC-metro BAH from the official [2026 BAH rate file](https://www.travel.dod.mil/Allowances/Basic-Allowance-for-Housing/BAH-Rate-Lookup/) (MHA DC053, with dependents) and the [DoD 2026 BAH release](https://www.war.gov/News/Releases/Release/Article/4357076/department-of-war-releases-2026-basic-allowance-for-housing-rates/) (4.2% average increase); BAH/BAS untaxed per [IRS Publication 3](https://www.irs.gov/publications/p3); the 3.8% 2026 raise per [CRS IF10260](https://www.congress.gov/crs-product/IF10260); BRS mechanics from the official [defined benefit](https://militarypay.defense.gov/Portals/3/Documents/BlendedRetirementDocuments/Fact%20Sheet-Defined%20Benefit.pdf), [defined contribution](https://militarypay.defense.gov/Portals/3/Documents/BlendedRetirementDocuments/Fact%20Sheet-Defined%20Contribution.pdf) and [continuation pay](https://militarypay.defense.gov/Portals/3/Documents/BlendedRetirementDocuments/Fact%20Sheet-Continuation%20Pay.pdf) fact sheets; [Tricare Prime enrollment fees](https://tricare.mil/Plans/Enroll/Prime/EnrollmentFees) (none for active-duty families) vs the [KFF 2025 Employer Health Benefits Survey](https://www.kff.org/health-costs/annual-family-premiums-for-employer-coverage-rise-6-in-2025-nearing-27000-with-workers-paying-6850-toward-premiums-out-of-their-paychecks/) ($26,993 family premium); bonus ceilings at [37 U.S.C. 331](https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim-title37-section331&num=0&edition=prelim), [Navy SRB policy](https://www.mynavyhr.navy.mil/Career-Management/Community-Management/Enlisted-Career-Admin/SRB-SDAP-Enl-Bonus/), [Air Force FY25 SRB list](https://www.afpc.af.mil/News/Article/4022949/air-force-releases-fy25-selective-retention-bonus-list/) and [FY26 reporting](https://www.airandspaceforces.com/air-force-fewer-reenlistment-bonuses-2026-list/); [CONUS COLA](https://www.travel.dod.mil/Allowances/CONUS-Cost-of-Living-Allowance/) (taxable) with the [2026 rates release](https://www.war.gov/News/Releases/Release/Article/4366432/the-department-of-war-releases-the-2026-continental-us-cost-of-living-allowance/) (~127,000 recipients, ~$99M, 18 areas), [OCONUS COLA](https://www.travel.dod.mil/Allowances/Overseas-Cost-of-Living-Allowance/) and [OHA](https://www.travel.dod.mil/Allowances/Overseas-Housing-Allowance/) (both untaxed, overseas only). The "2026 DC-metro cash" column and every rank-to-salary pairing are my arithmetic from those tables (basic pay + DC053 BAH + BAS, excluding the federal tax advantage), rounded to the nearest $100.
- **VA disability rates, prevalence and work rules:** monthly rates effective Dec 1, 2025 from [VA's compensation rates page](https://www.va.gov/disability/compensation-rates/veteran-rates/) (2.8% COLA, matching the [SSA 2026 COLA](https://www.ssa.gov/cola/)); tax-free per [IRS Publication 525](https://www.irs.gov/publications/p525) and [38 U.S.C. 5301](https://www.law.cornell.edu/uscode/text/38/5301); "regardless of whether they work or what their income is" and the SSDI contrast from [CBO's Income of Working-Age Veterans Receiving Disability Compensation](https://www.cbo.gov/publication/59836) and [CBO report 59380](https://www.cbo.gov/system/files/2023-12/59380-Veterans.pdf) (40% of post-9/11 veterans); [TDIU rules](https://www.va.gov/disability/eligibility/special-claims/unemployability/) with marginal employment defined at [38 CFR 4.16](https://www.law.cornell.edu/cfr/text/38/4.16); SSDI's 2026 $1,690 substantial-earnings line from [SSA](https://www.ssa.gov/benefits/disability/work.html); prevalence and ratings distribution from the [VA FY2025 Annual Benefits Report, compensation section](https://www.benefits.va.gov/REPORTS/abr/docs/2025-compensation.pdf) (6,338,253 recipients; 4,004,299 of 9,111,600 Gulf War era veterans = 43.9%, my division of the report's own counts; 100% is the modal rating at 29.1%); [DFAS CRDP](https://www.dfas.mil/retiredmilitary/disability/crdp/) and [CRSC](https://www.dfas.mil/retiredmilitary/disability/crsc/) for the retiree offset exceptions; severance recoupment under [10 U.S.C. 1174](https://www.law.cornell.edu/uscode/text/10/1174). The 22% gross-up equivalents are my arithmetic, stated as such.
- **State benefits for 100% P&T veterans:** Florida statutes [196.081](http://www.leg.state.fl.us/statutes/index.cfm?App_mode=Display_Statute&URL=0100-0199/0196/Sections/0196.081.html) (total homestead exemption), [320.084](https://www.flsenate.gov/Laws/Statutes/2024/320.084) (free DV plate, minus preserved service charges), [322.21](https://www.flsenate.gov/Laws/Statutes/2024/322.21) (license fees waived), [379.353](https://www.flsenate.gov/Laws/Statutes/2024/379.353) (free hunting/fishing licenses), [258.0145](https://www.flsenate.gov/Laws/Statutes/2024/258.0145) (free lifetime state-park pass), and [338.155](https://www.flsenate.gov/Laws/Statutes/2024/338.155) (the narrow toll waiver); [Texas Tax Code 11.131](https://statutes.capitol.texas.gov/Docs/TX/htm/TX.11.htm); [Alabama H-3](https://www.revenue.alabama.gov/property-tax/homestead-exemptions/); [New Jersey](https://www.nj.gov/treasury/taxation/lpt/lpt-disabledvet.shtml); [Virginia 58.1-3219.5](https://law.lis.virginia.gov/vacode/title58.1/chapter32/section58.1-3219.5/); [Michigan MCL 211.7b](https://www.legislature.mi.gov/Laws/MCL?objectName=MCL-211-7B); [Oklahoma](https://oklahoma.gov/veterans/benefits.html) (Okla. Const. Art. 10 Sec. 8E; sales-tax cap $25,000/yr); [Maryland](https://veterans.maryland.gov/benefits-services/tax-exemptions/state-property-tax-exemptions); [Nebraska Category 4V](https://www.sarpy.gov/165/Homestead-Exemption); [Honolulu](https://realproperty.honolulu.gov/tax-relief-and-forms/exemptions/totally-disabled-veterans/) ($300 minimum tax remains); New Hampshire's adapted-homes-only scope at [RSA 72:36-a](https://gc.nh.gov/rsa/html/V/72/72-36-a.htm); consolidated candidate list (Illinois/Pennsylvania caveats) via [Veterans United](https://www.veteransunited.com/futurehomeowners/veteran-property-tax-exemptions-by-state/), a secondary source used only for enumeration. The $7,500-to-$10,000 exemption-value figure is my illustration at a 1.5-to-2% effective rate, not a statute number. Several statute and .gov pages block automated fetching and were verified via Internet Archive snapshots of the same URLs.
- **VA home loan and GI Bill:** [VA home loans](https://www.va.gov/housing-assistance/home-loans/) (zero down, no PMI, nearly 90% no-down) and [purchase loan](https://www.va.gov/housing-assistance/home-loans/loan-types/purchase-loan/); [no loan limit for full entitlement](https://www.va.gov/housing-assistance/home-loans/loan-limits/) since Jan 1, 2020 ([Blue Water Navy Act, PL 116-23](https://benefits.va.gov/benefits/blue-water-navy.asp)); [funding fee schedule and disability exemption](https://www.va.gov/housing-assistance/home-loans/funding-fee-and-closing-costs/) (2.15% first-use at 0% down, waived at any 10%+ service-connected rating; the ~$8,600 on a $400,000 loan is my arithmetic). GI Bill: [Post-9/11 rates AY2025-2026](https://www.va.gov/education/benefit-rates/post-9-11-gi-bill-rates/) ($29,920.95 private/foreign cap, $1,000 books, E-5-with-dependents MHA, 36 months); [Post-9/11 overview](https://www.va.gov/education/about-gi-bill-benefits/post-9-11/) and [transferability](https://www.va.gov/education/transfer-post-9-11-gi-bill-benefits/) (6 years served + 4 more); the $100,000+ total is a widely used third-party estimate, not a VA figure.
- **100% P&T family and access benefits:** [CHAMPVA](https://www.va.gov/health-care/family-caregiver-benefits/champva/) (spouse/children not eligible for Tricare; $50 deductible, $3,000 household cap per [CHAMPVA care](https://www.va.gov/resources/champva-care/)); [Chapter 35 DEA](https://www.va.gov/education/survivor-dependent-benefits/dependents-education-assistance/) and its [$1,574/month full-time rate](https://www.va.gov/family-and-caregiver-benefits/education-and-careers/dependents-education-assistance/rates/); [commissary/exchange/MWR access for any service-connected rating since Jan 1, 2020](https://news.va.gov/67974/commissary-military-service-exchange-mwr-access-extended-veterans-beginning-january/) (Purple Heart and Disabled Veterans Equal Access Act 2018), [commissary cost-plus-5% pricing](https://corp.commissaries.com/our-agency/about-deca) and [~25% savings](https://corp.commissaries.com/rewards-and-savings/patron-savings), [exchange sales-tax immunity](https://www.aafes.com/exchange-stores/faq/); [Space-A travel at permanent 100% P&T](https://news.va.gov/60642/eligibility-disabled-veterans-space-available-flights/); [Special Monthly Compensation above the 100% rate](https://www.va.gov/disability/compensation-rates/special-monthly-compensation-rates/).
- **Versions and prices** were checked on 2026-07-18 and will drift; where a current doc contradicts this post, trust the doc.

*This post is informational, not career, financial, tax, benefits or security-clearance advice. Mentions of third-party companies, products and government programs are nominative fair use; no affiliation or endorsement is implied, and nothing here describes any non-public system. Salary figures are survey averages or point-in-time postings, not offers; military pay, VA rates and state benefit rules carry detailed eligibility conditions summarized here only in outline, so verify your own situation with DFAS, VA and your state's veterans agency.*


---

Canonical HTML: https://jwatte.com/blog/becoming-defense-data-platform-engineer/
RSS: https://jwatte.com/feed.xml
JSON Feed: https://jwatte.com/feed.json
Hero image: https://jwatte.com/images/becoming-defense-data-platform-engineer.webp