Available Hire Me
← All Writing Career

Going Independent — What a 21-Month Public-Sector Contract Taught Me About Shipping Alone

Leaving a 21-month DWP Digital contract to work independently: what transfers from national-scale delivery, what does not, and how GiltIQ shipped in 14 weeks.

My contract with DWP Digital ended in September 2026 after 21 months. I now work independently through Samuel Jackson Ltd, with product work delivered under the Trinity Logic brand. Nothing dramatic happened: the programme reached a natural boundary, and I had already decided that the next thing I wanted to build was my own.

This is a note on what transferred, what did not, and what I would tell anyone making the same move. It is not a resignation story. It is a delivery story with a different org chart.

What the public-sector contract actually taught

The New Style Jobseeker’s Allowance platform is a mission-critical service at national scale. The technical shape will be familiar to anyone who has worked in government digital: Spring Boot microservices, Kafka for asynchronous workflows, MongoDB for claimant state, Hexagonal Architecture to keep domain logic away from infrastructure, and an unusually strong emphasis on idempotency, fault tolerance, and recoverability.

The parts that changed how I work were not the frameworks.

Correctness is a governance property, not a code property. When a service manages payments, “the retry might duplicate a payment” is not a bug to schedule; it is a design constraint. Idempotency keys, deduplication, and replay safety have to be in the first design, not retrofitted after a load test. That discipline is now the default I bring to commercial work.

Auditability is a feature. A figure that cannot be traced to its inputs is not a figure. In government work, that is enforced by review and evidence. In commercial work it is usually optional, which makes it more valuable, not less.

Multi-supplier delivery is an architecture problem. Clear service ownership, explicit contracts, and disciplined integration boundaries matter more than any internal code quality metric. A beautiful service with an ambiguous contract is an outage waiting for a programme boundary.

Testing is the specification when the organisation is large. Contract tests, integration tests, and a build that fails loudly are how a distributed team keeps a shared understanding. I left with a strong preference for test suites that encode invariants rather than mirror implementation.

What does not transfer: the platform team

The biggest change is not technical. On a programme, there is a platform team, a security team, a delivery manager, and an operations function. Independently, those functions are either automated, bought, or done by me.

That is mostly liberating and occasionally tedious. The parts I kept:

  • A build loop I own. I moved my daily Java work to a terminal-first setup — Neovim, jdtls, Maven, and a scripted inner loop. The point was not aesthetics; it was removing a dependency on an IDE’s hidden build model. I wrote about the measurements and the traps separately.
  • Boring infrastructure. Where a managed service does the job, use it. The interesting decisions should be in the domain, not in the operational plumbing.
  • One source of truth for configuration. The same instinct that makes a large programme work — explicit contracts, versioned data, no shadow configuration — makes a one-person product maintainable.

The parts I had to add: commercial scoping, a cost model, and the discipline to stop building before the next feature.

GiltIQ as the proving ground

GiltIQ is the product I built after the contract: a UK gilt analytics and portfolio accounting platform for private investors. It prices all 104 in-issue gilts after tax on every business day, publishes a daily report, and folds an event-sourced trade log into positions, taxable-income projections, and performance.

It shipped in fourteen weeks as 1,309 commits, with 470 Java source files and 443 test files. Almost one test file per source file was not a vanity metric; it was the only way to move quickly in a domain where a silent arithmetic error is invisible until someone makes a decision on it.

Three decisions did most of the work:

The method travels with the number. Every published figure carries a calculation version, and every figure can be opened to its inputs. That is the auditability habit from public-sector delivery applied to a retail product. It also makes regression testing possible: a change in methodology is a diff, not a rumour.

The read path was designed for verification. S3 is the system of record; an embedded DuckDB store serves reads. The store is rebuildable, so a corrupted read model is an operational inconvenience rather than a data loss event.

Guard tests police the invariants. Some tests read templates and stylesheets verbatim and fail if a label, convention, or method note disappears. That sounds brittle. It is the cheapest way to keep a one-person codebase honest when the reviewer is also the author.

What I would tell someone considering the same move

Decide what “independent” is for. If it is a bridge between contracts, optimise for availability and keep the product small. If it is a product business, accept that the engineering is the easy half. I wanted both: a product I own and the freedom to take contract and consulting work. That shapes the schedule, the runway, and the kind of client I say yes to.

Build the evaluation habit early. Whether the deliverable is a gilt yield or an LLM answer, you need a set of cases that define “correct” before you start optimising. Without it, every change is a vibe check and every regression is discovered by a user.

Keep the operating surface small. One repository, one pipeline, one deployment path, one place where secrets live. Solo delivery fails through operational sprawl more often than through technical difficulty.

Write the note while the decision is fresh. The method note, the ADR, the “why this and not that” paragraph. Future-you is a different person with less context, and future-client will ask.

Do not confuse motion with progress. A 1,309-commit product is not 1,309 units of value. The useful measure is whether a user can do something they could not do before, and whether the platform can prove it.

What is next

I am in early discussions with a UK housebuilder about an enterprise-wide AI platform capability, and I am available for contract, consulting, and architecture review engagements through Samuel Jackson Ltd. The gilt platform continues to run and publish.

The move from a 21-month programme to a one-person product did not change what I care about: systems that are correct, traceable, and operable by the people who inherit them. It changed the size of the team enforcing it.

If you are weighing up a similar move — or you need someone who has done both — get in touch.

Samuel Jackson

Samuel Jackson

Senior Java Back End Developer & Contractor

Senior Java Back End Developer — Betfair Exchange API specialist, Spring Boot, AWS, and event-driven architecture. 25+ years delivering high-performance systems across betting, finance, energy, retail, and government. Available for Java contracting.