Home / Articles / The bridge, followed to the end / Versions / v1.1.0
The bridge, followed to the end: what changed in v1.1.0
From v1.0.0 (2026-10-07, 9e71c7d67) to v1.1.0 (2026-10-08, a071264ce), paragraph by paragraph.
18 paragraphs added, 0 removed, 0 changed in place, 39 unchanged. About 1,198 words added and 0 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.
5 unchanged paragraphs
!trail bridge
9 unchanged paragraphs, under The journalism is getting the date right, Newsworthy on five days, needed on fifty-seven
[figure collage-front-pages.webp] The Courier on the five days the bridge was news, each front page one projection of the story graph with the tracker beside it, and the fifty-seven days in between that readers needed it.
22 unchanged paragraphs, under Three readers, three graphs, one story, With accurate information, and without it, The buyers who do not read the paper…
Version 0.2: what it cost, and what it keeps earning
The first version of the vault showed who was paid. The question it left open is the one any publisher would ask next: what did it cost, and does it make money? So the vault now has a second build script, tools/economics.py, and nine new views, and because the vault is versioned like code, its *What's new* page records the change.
[figure collage-views.webp] The vault, view by view: twenty of its twenty-two views, the nine new in version 0.2 outlined.
Costs and profit. The fixed cost is mostly the reporting, and it comes first: 50 hours of the reporter's time and 7.5 of the editor's, £2,040 with a legal read, set-up and the platform. The variable costs, payments, alerts, compute and hosting, follow use and come to 9.3% of revenue. During the closure the Courier makes £678, the infrastructure share covers its costs with £20 to spare, and the reporter is under water: her share is about £19 an hour against a cost basis of £30. The story breaks even on 20 November, day forty, and the scale chart says why the town mattered: below about 1,500 followers it would not have paid for its reporting, and above that every follower is margin.
[figure el-economics-charts.webp] Revenue against cost, day by day, crossing on 20 November; and the same story with more or fewer followers, where fixed costs stay put and variable costs follow use.
Two years on. The story stopped being news on 4 December. It kept selling. A council post-incident review, loss adjusters handling business-interruption claims, the county's inspections of eight bridges of the same design, a second fund ahead of the contractor's results, a procurement watchdog, a university research group, an insurer and two other towns with the same bridge all buy the record of what was known, when, and from whom. Over two years the bridge earns £4,378 and £1,181 of profit. After the reopening, 96% of its revenue comes from institutions and agents rather than the town: the revenue moves from the local reader to the players who want the data. The reporter's effective rate rises from £20 an hour at the reopening to £25.54, and Priya's three photographs have earned her £328.
[figure el-projection-chart.webp] Revenue by month over two years, by who paid: the town in October and November, institutions and agents after that, against a cost that is almost all in the first two months.
The next stories, and the payment as a signal. The bridge is the first story in the Courier's graph, not the last. The vault plays out eleven more over two years: a school catchment change, the footbridge resurfacing, the county's sister-bridge inspections, a business rates review, a bus timetable, a wharf flood-defence overrun, a contractor's pattern across three councils, a car park, winter floods, a school meals contract and night closures on the bypass. Each starts with the council, the procurement portal, the routes and the contributors already in the graph, which saves 157 of 331 reporting hours. And each is sized by what people paid for in its first week, from readers and from institutions asking ahead: six are investigated, five get a tracker, and the footbridge resurfacing, which few people needed, gets a brief, three hours and a £73 loss that keeps the graph complete. That is the argument made concrete: a bridge nobody cares about earns no payments and gets no investigation, and the first payments are what tell the newsroom where to dig. By the second year 41% of revenue comes from stories more than three months old, the twelve together earn £26,523 and £13,969 of profit, and across all of them the reporter earns £53 an hour.
[figure el-stories-chart.webp] Revenue by month across twelve stories: stories in their first three months in blue, the long tail of older stories in orange, which grows with every story added.
Trust, earned. Every story with a date in it adds a row to a record: the official estimate, the Courier's range and the outcome. Across eight of them the outcome lands inside the Courier's range seven times, and once a day outside it; the official estimates are out by 34 days on average. The record also keeps what each source got right, including the engineer who was interviewed for nothing and whose range held, and the anonymous workers whose record attaches to the claim, not to a name. A trust index built from the record scales the take-up of the stories that follow, so credibility shows up in the accounts.
The council and the accountants. Dev's £1,200 of relief looks different from the council's side. A council wants to refund correctly. With the tracker it sized the budget from the graph on 16 October rather than topping it up on 25 November, paid 44 of the 52 eligible businesses instead of 18 and none it was not meant for, and spent half the staff time, for £152 of targeting bought from the Courier. On the business's side, an accountant agent reads the same claims, keeps the takings and the van log on the business's side, and pays pence per question: for Dev's café the closure costs £1,530 net with the tracker, against £5,540 without it.
[figure el-council.webp] The relief scheme without and with the tracker: when the budget was set, who heard about it, who was paid, and what it cost the council to process.
Ask the graph, and the API. Two views show what the vault is to a machine. *Ask the graph* simulates a model query: pick a question and a date, and the answer is composed from the claims live that day, with citations and the context a model would be given, 583 tokens against 45,000 for searching six pages, about a tenth of the cost and right. No model is called, which is the point: the hard part is the graph. *The API* is a Swagger-style explorer for an OpenAPI description in the vault, seventeen endpoints, each a GET whose response is a file already in the vault, with a price per call where there is one. The thirty data files are listed as assets in their own right, with downloads. Readers' graphs have no endpoint: they live on readers' devices.
[figure el-ask.webp] A simulated model query: when will it reopen, asked on 16 October, answered from the three live claims with citations, and the context the model would be given.
[figure el-api.webp] The story as an API: every endpoint a GET, every response a file in the vault.
2 unchanged paragraphs, under What the simulation is for
*Revised on 8 October 2026 for the vault's version 0.2, from a second voice note asking for the costs, the profitability over time, the next stories, the trust, the council, an accountant agent, simulated model queries and an API.*
1 unchanged paragraph