Fixtyerpothole claims,made sense of.

A motorist with a bent wheel and a repair bill has to answer several questions before they can claim a penny: who manages that stretch of road, what evidence is needed, whether the pothole had already been reported, what paperwork to assemble and how to submit it. Each answer lives in a different place. Most people give up, or hand a third of any payout to a claims company.

Fixtyer pulls those tasks into one journey while keeping an important boundary intact: the customer reviews, submits and manages their own claim. The product prepares documents and explains the process. It does not represent anyone, and it does not promise an outcome.

That framing set the design problems. Make an unfamiliar, stressful process understandable without demanding registration first. Turn photographs and receipts into clear next actions. Present AI help honestly, as software, not an expert. Make a very large public library discoverable next to a private application. And make payment, background generation and file delivery reliable across separate services.

The identity is deliberately practical: a strong blue, orange for every call to action, and diagonal blue-and-orange stripes borrowed from the roadside. Syne for display, DM Sans for reading. Pale surfaces, rounded cards and clear status colours keep the denser application screens legible. Everything is expressed as Tailwind theme tokens and shared components, so the public site and the logged-in application share one language while serving different jobs.

Much of the real design work is in the in-between states: upload feedback, evidence gaps, analysis in progress, an uncertain authority, a payment that is still processing, a session that needs refreshing. Optimistic uploads, manual vehicle entry when the registration lookup fails, guarded payment-modal closure and clear wizard errors are what make the service usable when an external service is slow.

Mac is the persona for evidence review, and the site says plainly that Mac is software, not a person, and what Mac cannot do. The voice is direct British English: what you have supplied, what is missing, what to do next. “You’ve made a start, but there are gaps to fill.” A plain-English summary sits beside evidence categories, priority actions and photo guidance, rather than an unexplained score. Keyboard focus, labelled controls and interaction-level accessibility fixes are implemented work, not a certified conformance claim.

The paid artefact is a substantial part of the product. A successful £35 payment, through embedded Stripe payment components with server-side handling, triggers background generation, file storage and a pack-ready email. The pack itself is built with pdf-lib and embedded fonts: cover, next steps and evidence checklist for the customer; a formal claim letter, draft witness statement and schedule of facts; location information, defect and damage photographs, financial evidence and any prior-report material.

The most useful engineering decision is the boundary between supplied facts and generated interpretation. Shared fact helpers use only what the claimant provided and the document categories that exist. They do not invent driver status, sensory detail, missing dates or a submission event. The draft witness statement is built from structured damage facts rather than an AI-rewritten narrative, and pre-signing checks ask the customer to verify every page.

Payment and generation share an idempotency key so the webhook and the confirmation request converge on one background run; unique payment identifiers and generation checks guard against duplicates. An £89 Court Pack exists in the product definition but is blocked from purchase until it is built. There are no subscriptions, and the platform never submits or negotiates a claim on the customer’s behalf.

Around the application sits a large public library: a page for each of 153 English highway authorities combining published maintenance information, Department for Transport ratings, source material and practical claim instructions; guides on evidence, responsibility, timelines and rejected claims; a searchable ratings dataset with CSV and JSON exports; sample pack material; and a newsroom whose press releases are shared between web pages and generated, branded PDFs. It is structured data engineering and editorial work together, not a set of place-name landing pages.

The site is React and Vite, which search engines historically read badly. Rather than move to a request-time framework, a server entry renders every public route at build time into page-specific HTML that hydrates in the browser. Serving that correctly from a static host took 185 managed rewrites and 19 header rules across 219 routes, so deep URLs get their own page rather than the homepage shell, and a missing page is a genuine 404. Release tooling checks page identity, canonicals, indexing directives, prerender markers and the deployed commit before a release is finalised.

In September, Search Console showed an intent problem: council pages led with maintenance research when visitors wanted to know how to claim. A limited pilot on Dudley, Bristol and West Northamptonshire put claim-focused titles, the checked official submission route, a road-responsibility caveat, the evidence checklist and submission steps ahead of the research. Keep the useful material; put the reader’s task first. The pilot is being measured against a later comparable window before it is rolled out further.

What is evidenced: a complete self-service document-preparation platform; a substantial council research and guidance library; public React routes served as page-specific HTML; a journey that gives value before registration; AI feedback inside a structured workflow; factual review boundaries built into generated documents; and continued development driven by search and operational evidence.

What we do not claim: compensation recovered, success rates, conversion or revenue growth, guaranteed authority identification, certified security or accessibility, or passed Core Web Vitals. Search traffic is an early baseline: 73 clicks from 9,243 impressions over the summer, almost all from the UK. The main client bundle is still large and is an identified improvement area. Next: measure the September council pilot, keep the official submission routes checked, shrink the bundle, strengthen editorial authorship and build the Court Pack before enabling it.