Timeline

The problem

A chronology of one company is a Wikipedia article. The question nobody can answer in a single view is the comparative one — when GPT-4 shipped, what did Anthropic have in market, and how long did the others take to respond?

What I owned

• Product concept, data model and implementation end to end
• A zod-backed content schema where each category is its own JSON file, free to extend a shared minimal contract
• The race grid: CSS Grid against a quarterly axis, with weighted events and empty stretches folded into a labelled break
• Editor-style pan and zoom — wheel, drag, pinch and keyboard — written directly rather than taken from a library
• A separate phone view, because eight columns of quarters cannot be compared on a 360px screen
• Researching, dating and sourcing every event, and the script that verifies the citations still resolve

Technical challenges

• Event density: the field produced more in 2023–2025 than in the twenty years before, so a year-per-row axis collapses into an unreadable pile
• Making a horizontally scrolling grid usable with a mouse, where a wheel does nothing and the scrollbar is the only way across
• Keeping a pinned column pinned — padding on the scrolling content silently unsticks it, because it pins to the scrollport rather than the padded box

Engineering decisions

• Organisations as rows rather than columns: eight columns put five on screen, which defeats the comparison the product exists for
• Quarters rather than years, with a weight on every event so the default view carries the significant ones and a control reveals the rest
• A source URL required on every event by the schema, day-precise dates even though the view groups by quarter, and tests for dangling references, duplicate ids and future dates
• No pan-and-zoom library — the carousel ones fight a free-scrolling grid and the gesture ones were more than one handler needed

Technology