The 60+ page report reconstructs how Direct File was built with USDS/18F help, ran a successful pilot, and was killed despite strong operational metrics — completion rates, satisfaction scores, and sub-hour median filing times. It concludes the shutdown was a policy decision driven by a decade of paid-preparation industry lobbying and a new administration treating the pilot as discretionary, not a response to any technical or budgetary failure.
By submitting the Vinton report to HN with the framing 'Life and Death of Direct File,' the submitter highlights that this is a story about a functioning public service being deliberately terminated, not a technology failure worth studying on its engineering merits alone.
The editorial argues Direct File executed the modern govtech playbook cleanly: small cross-functional pods, iterative rollout, real user research, plain-language UX, open-source components, and phased state expansion — with no prime contractor, no waterfall, and no big-bang launch. It frames the project as the intended vindication of the USDS model born from the Healthcare.gov trauma, making the shutdown a challenge to the assumption that better delivery practices are sufficient for civic tech t
The Berkeley iSchool just released Vinton's report on the life and death of IRS Direct File — the free, government-built tax filing tool that ran a pilot in 2024, expanded to 25 states for 2025, and was effectively shut down shortly after. The 60-plus-page report reconstructs how the product was built inside the IRS with help from USDS and 18F, what users actually experienced, and how the political machinery around it moved.
The operational numbers were not the problem. Direct File processed hundreds of thousands of returns in its first two seasons with a completion rate and satisfaction score that would embarrass most commercial fintech. Median filing time was under an hour. Refunds landed on schedule. The system integrated with state filing where states cooperated, and it handled the ugly parts of the U.S. tax code — EITC, CTC, ACA reconciliation — without pushing users into upsells.
The report is blunt about how the shutdown played out. It wasn't a rollback triggered by outages, security incidents, or budget overruns. It was a policy decision executed on top of a working system, driven by a decade of sustained lobbying from the paid-preparation industry and accelerated by a new administration that treated the pilot as a discretionary line item rather than infrastructure.
Most engineering post-mortems ask "what broke?" This one asks "what does it mean when nothing broke and the thing died anyway?" That is a much harder question, and it's the one civic tech has been dodging for two decades.
The Direct File team did roughly everything the modern-govtech playbook says to do: small cross-functional pods, iterative rollout, real user research, plain-language UX, open-source components, phased state expansion. They avoided the classic failure modes — no billion-dollar prime contractor, no waterfall, no big-bang launch. Healthcare.gov 2013 was the trauma that produced USDS; Direct File was supposed to be the proof that the lessons stuck. Technically, they did stick.
What the report documents — and what practitioners should sit with — is that the surrounding system was never rebuilt to match. Intuit spent an estimated $90M+ on federal lobbying over the last decade, most of it aimed at keeping the IRS out of the filing-software business. The Free File Alliance, the industry-run "free" program Direct File was meant to replace, spent years steering eligible users toward paid upgrades; ProPublica's reporting in 2019 documented Intuit deliberately hiding its truly-free product from search engines. Against that backdrop, a product with a 90% satisfaction score is not safe. It's a target.
The comparison to other public-option software is instructive. The UK's GOV.UK platform survived multiple governments because it embedded itself as shared infrastructure across dozens of agencies — killing it would break too many things. Estonia's X-Road is load-bearing for the entire state. Direct File was designed as a standalone consumer product, which meant it had exactly one dependency graph pointing at it: taxpayers, who don't have a lobbyist.
Community reaction on HN and among civic-tech alumni has been unusually pointed. Several ex-USDS engineers have publicly said they're done with tours of duty; a common refrain is that shipping the code is the easy 20% and the other 80% — building a political moat around the product — was never staffed. The report echoes that: the team had product managers and researchers and engineers, but no one whose full-time job was making the tool structurally difficult to remove.
If you work on anything that touches government, regulated industries, or platform-level infrastructure that someone else profits from being bad, treat this as a working case study. A few concrete implications.
First, usage metrics are not a moat. They're a talking point. If your product's existence depends on an annual appropriation or a single executive sponsor, ship-quality doesn't save you. The Direct File team had the metrics and lost anyway. Design for structural entrenchment from day one — integrations, data dependencies, downstream systems that would break — not just for user satisfaction.
Second, open-sourcing components matters more than open-sourcing the product. Direct File's underlying fact-graph engine was open-sourced; that code will outlive the pilot and can be picked up by states (New York and California have both indicated interest in building on it). If you're building public-interest software inside an institution you don't control, the durable artifact is the reusable component, not the branded application on top of it.
Third, audit trails and public metrics are a defense mechanism. The reason we have this post-mortem at all — the reason the shutdown is politically expensive rather than invisible — is that the team published usage data, satisfaction scores, and cost figures throughout. That transparency didn't save the product, but it's what makes the counterfactual undeniable. If you're inside a large org shipping something a rival unit wants dead, instrument aggressively and publish externally.
The Direct File codebase isn't going anywhere; several states are already forking pieces of it, and there's an active push to make the fact-graph engine a shared civic library. The larger question the report leaves open is whether the USDS/18F model — parachute in strong engineers, ship modern software inside legacy agencies — can survive without the political architecture to protect what gets built. The next generation of civic tech will need to spend as much design effort on institutional durability as it does on user experience, or it'll keep producing beautifully-built products with a two-year shelf life. That's a harder problem than shipping the software, and it's the one the field has to solve next.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.