About
One archive of what Hastings' councils and other public bodies publish: decisions, money and opportunities, joined up and put into plain English, with every claim linked to its source.
Why
Committee packs run to hundreds of pages. Grant awards sit in databases nobody browses. Consultations close before the people they affect hear about them. The local press that once translated all of this has largely collapsed. Open Hastings is being built to fill that gap. It is a public record residents can actually use, rather than a newspaper or a campaign.
What's here now
The archive index: committee meetings and their papers, planning applications and open consultations, collected continuously from the councils' own publishing systems; the money going in (published grants and contracts) and, on the spending page, the money going out, from the borough council's own monthly payments files. Everything links to the original document.
How the summaries are checked
Where you see a plain-English summary, it has been through checks designed to stop a wrong claim about a council decision from being published:
- The document is split into numbered passages, and every sentence we write has to cite at least one of them. If a sentence can't be anchored to a passage, we don't write it.
- Automated checks then reject the draft if any sentence lacks a citation, cites a passage that doesn't exist, editorialises, or claims something was decided when the document only proposes it. A rejected draft is rewritten once and then held back.
- A second, independent pass reads each sentence against only the passages it cited, and is asked to find anything the sentence claims that those passages don't say. Whatever it can't fully support is held back for a person to review.
- The words that decide the meaning (approved, refused, proposed, applied for, awarded) are not written by the model. They come from fixed templates chosen by what kind of document it is. "Proposed" and "approved" are different facts, and confusing the two is the mistake we most want to avoid.
- If a document changes after we've summarised it, we stop publishing the summary rather than leave it describing a version that no longer exists.
Which papers are summarised first. Only a few papers can be summarised each day, so they wait in a queue with a fixed order: papers for meetings in the last two months or still to come before older ones; among those, Cabinet, Full Council and Lead Member decisions first, then minutes, then other papers; borough papers before county ones; and the newest meeting first. The same rule applies to every paper, whatever it is about.
We treat "the record doesn't say" as a proper answer rather than a failure. Plenty of documents don't support a summary, and saying so is more useful than guessing.
Where we use AI
We use it in three places. The summaries above are written by a language model and then checked in the ways described. Asking the record uses a model only to choose between passages the councils wrote, and it never writes a sentence of its own. And for a few county papers, a model helps decide whether a street they name is the Hastings street. Everything else (the topic pages, the radar, how money is counted, how streets and organisations are joined up, and where the rest of the lights go on the map) is plain rules, applied the same way every time, and each page states the rule that made it.
Asking the record
Tap the search button at the top of any page (or press / on a keyboard) and type a question in plain words. Three things then happen, in this order:
- Find. An ordinary word search over every passage in the archive, from committee reports, minutes and the borough's papers, picks the 20 passages that share the most words with your question.
- Judge. A language model run by TypeSafe is asked one yes-or-no question about each of those passages, and gives a probability. The question, word for word, is: "Does this passage directly answer the question?" It counts as yes when the passage itself states the answer or a specific fact, figure, date or decision that answers it, and no when it is only about the same subject or shares words with the question.
- Show. Passages the model puts at 50% or above are shown as answers, best first, quoted exactly and linked to the council's document, with the probability beside each. If none gets there, the reply says the record doesn't seem to answer your question, and the two nearest passages are labelled as the closest, not as answers.
We measured this before switching it on: on real passages from this archive, the passage that answered came top for every answerable question we tried, including ones where plain word-matching ranked a non-answer first, and a question the record can't answer scored no higher than 7%. That is a small test, and we re-run it whenever the question or the 50% line changes.
Who represents you, and when meetings are. These two kinds of question are not answered from passages. "Who is my councillor for Central St Leonards?" is answered from the councils' own member registers, and "When is the next Full Council?" from their meeting calendars, with no model involved. A council paper can mention a name for many reasons (a candidate in an election notice, say), so it is the wrong place to look for who holds a seat.
Searching the text. The search box matches titles in your browser. To find a word inside the papers themselves ("Hastings Pier", "bin collection"), choose "Search the text of council papers": an ordinary word search on our server over the same passages, the phrase as you typed it first, then papers using some of its words. No model is involved.
What it sends. Searching titles happens in your browser and sends nothing. Searching the text sends your words to this site's server, which searches and forgets them. When you choose to ask, your question goes to this site's server and, with the passages, to TypeSafe so that its model can judge them. Our server doesn't log your question or save it, although a recent question may be answered again from a small memory cache that is cleared whenever the server restarts. There are no accounts, so nothing links a question to you.
Is it the Hastings street?
The map lights a street when council papers name it. A borough council paper is about Hastings, so it always counts. County council papers name streets all over East Sussex, and plenty of towns have a High Street or a Station Parade, so a county paper counts when the sentence naming the street also names Hastings or St Leonards. For the other county sentences, a language model run by TypeSafe is asked one yes-or-no question and gives a probability. The question, word for word, is: "Does this sentence refer to the street named X in Hastings, East Sussex (the town, including St Leonards-on-Sea), rather than a street of the same name somewhere else?" It is shown only the street's name, the sentence, the paper's title and its committee. It counts as yes when the sentence places the street in Hastings or names Hastings places beside it, and no when the street is in another town, when a school, surgery or station is named after a street elsewhere, or when nothing in what it is shown places the street in Hastings.
A sentence is included only when the model puts it at 75% or above, and the map says so under the sentence, with the probability. The model writes nothing: the map still shows the council's own sentence. We measured this on 27 real county mentions before switching it on. Every sentence about a street somewhere else scored 63% or less, so none of them gets past the 75% line. The line is set high on purpose. A sentence from a county paper rarely says where its street is, the model is often unsure about these, and a light in the wrong place is worse than a missing one, so most real Hastings mentions that don't name the town still stay off the map. We re-run the test whenever the question or the line changes.
The same question also checks the county sentences that do name Hastings or St Leonards, because naming the town isn't proof. One county agenda gave directions "from Eastbourne and Hastings" in the same sentence as the High Street in Lewes. If the model puts one of these sentences below 17%, the map leaves it off. In our test the Lewes sentence scored 14%, and the lowest real Hastings sentence scored 19%, so the line is set between them. That is a narrow gap, and we'll widen the test as more papers come in.
Corrections
When we get something wrong, the correction is made in place, with a note saying what changed, and logged here.
Who
Built and maintained in Hastings by the person behind Swim Hastings and Live Hastings. Nobody funds it, it sells nothing, and how it works is documented on this page.
Sources & licensing
Committee papers link to East Sussex County Council's and Hastings Borough Council's own committee systems (the borough's are collected through the web service its committee platform publishes for machines); consultation listings link to hastings.gov.uk. Documents remain © their publishing authorities; this site indexes and links rather than mirrors. Public-sector information is used in the spirit and, where applicable, the terms of the Open Government Licence v3.0.
The rules
- Every entry shows its source and the date we last confirmed it.
- How every summary and label is made is explained on this page.
- If collection breaks, the site says so at the top of the pages it affects instead of showing old data as though it were current.
- We report what public bodies publish and decide. A consultation entry tells you how to respond, and never suggests what to say.
- No accounts, no tracking, no ads.