Case file 01 · Data · Journalism
Under Fire
Six years of rocket, missile and drone alerts across Israel, turned into an interactive data story: timelines, heatmaps, salvo arcs across the map, and records. Just the data, no commentary.
- Year
- 2026
- Role
- Solo: data pipeline, design, build
- Read
- 4 min
Context
A public-interest data project covering six years of rocket, missile and drone alerts across Israel. It was built originally on RocketAlert.live's dataset, which runs from January 2020. The framing was a hard rule from the start: no blame, no policy, no commentary. Just the patterns, and a clear note of where every number comes from.
The problem
It started during the Iran war in March. I wanted to know whether there was any pattern in when and where Iran was sending its missiles. Alerts are published one event at a time: a timestamp, a place, a siren. At that level you can see that something happened, but not the rhythm of it, where it concentrated, or how different wars played out over six years. Making that readable was the main goal: pull the patterns out and make them easy to see.
The harder problem was keeping it current. The first version was a snapshot. It was built from a one-off 54MB download, so it started going out of date the day it went up, and updating it meant re-running everything by hand. What it needed was a live source of alerts the site could check on its own. I found one more or less by luck: a public feed from the same source as the original dataset. The feed has a gap, though. It says where a siren sounded but not who fired, so that part has to be estimated, and labelled as an estimate.
What I built
A hub of pages that each load on their own: a six-year timeline you can filter by who fired, monthly fronts, a daily calendar, a 24-hour clock of when attacks happen, records, a month-by-month time-lapse map, a side-by-side comparison of the wars, a story map that scrolls through six chapters, a minute-by-minute replay of October 7, a live feed of the last 14 days, and a map of the thirty Home Front Command regions you can hover over. Python boils the full record down to sixteen small summary files, about 680KB in total, before anything reaches the browser. Behind that, a scheduled job runs every 30 minutes. It pulls new alerts from the feed, removes duplicates, works out who most likely fired each one, rebuilds every summary and saves the result. So a site with no server behind it still keeps itself up to date.
Decisions & iterations
This was my first big project, so I wanted it to be something I could be proud of, and it went through a lot of versions. The first build was one long scrolling page that loaded every chart at once. That was fine while there were only about four charts. Splitting it into separate pages was the most important structural decision in the project, and it made every chart after that cheap to add. The salvo animation, which sends missile arcs across the screen, draws a scaled-down number of them. October 7 alone had nearly 4,000 alerts, and drawing every one would freeze the browser, so the label carries the true count.
Working out who fired needed a rethink too. My first method looked at each alert on its own and guessed from where the siren sounded. Checked against the older alerts that already had a label, it agreed only 67% of the time. So I changed it to look at alerts in groups. Sirens within 15 minutes of each other count as one attack, and days with hundreds of sirens across the centre of the country count as Iranian barrages. That agreed 93.1% of the time.
I also deleted the fake terminal that made you press “Run” before each chart. On a site whose whole argument is that the data is real, pretend code was costing credibility. The most recent change is the biggest visual one. I redesigned the whole site as a split-flap information board, the kind where the numbers flip into place, and in my view it's now much easier to find your way around.
Outcome
Live at under-fire.org and updating itself. The record runs from January 2020 to today and has passed 160,000 alerts. That includes 17,548 I added to close a gap in the spring, with 235 duplicate rows removed along the way. Before launch I cut the weight of the site's images by 98%, from 5.9MB to 115KB, so it loads quickly. The totals you see on the site aren't typed into the pages. They load from the same data file the 30-minute job updates, so they can't quietly go out of date. That matters on a site that's asking to be trusted.



