GAJRA Earth
Nothing rolls off into nothing.
The front of the watch holds roughly the last month, because a page nobody can scroll is a page nobody reads. Everything older moves here and stays, word for word, with its sources intact.
The value is in the sequence.
Any single morning's entry is a handful of links. What is worth having is the run of them: which bodies moved and which stayed silent, how long a proposal took to become law, whose objection was answered and whose was not, and how the language changed while that happened. None of that is visible in a snapshot, and it cannot be reconstructed later from memory or from search, because the pages being cited get edited and the search results reshuffle.
So the rule is that entries move and never disappear. If this project's argument is that the record of how a decision was made should be open to the people it lands on, the watch has to hold itself to that first.
By month
July 2026: the watch begins. 1 dated entry.
Each month page carries a keyword filter, so you can pull out every data centre item, or everything from the Pacific, or everything tagged against one of the open questions, without reading the whole month.
Keywords
Every entry is tagged from a fixed vocabulary held in data/watch-keywords.json, grouped as region, who acted, and what it was about. The list is fixed on purpose: keywords invented fresh each morning drift within weeks, and an archive where the same thing has six names is not searchable, it just looks like it is. New terms get added to that file deliberately, and the run that adds one says so.
Two of them are not topics but open questions this project is carrying: Q9 who the many are and how they are actually asked, and Q10 the accounts nobody can sign, meaning any case where a non-human interest changed a decision. Both are set out on the questions page.
Three copies, on purpose
The record exists in three forms, so that losing one does not lose the archive.
These pages. Readable by a person with a browser and no tools, which is the form that matters most and the one least likely to survive a redesign on its own.
A machine-readable file. Every item ever logged, one per line, with its date, scan time, source URL, question tags and keywords: data/watch.jsonl. Append-only. If the site is rebuilt in another shape, the whole watch can be re-rendered from this file without asking anyone to remember what was there.
The commit history. Every change to every page, including the corrections, is in the public repository. That is the layer that shows the edits, not just the result.
Commit history alone was the original plan and it was not good enough: it is a record only for people who use git, and it can be rewritten by anyone with push access. The pages and the data file are the parts a stranger can check.
What is not here
Defence and military items are excluded by standing rule, and where something notable was dropped on that basis the day's entry says so. Items that turned out to be older than they looked are dropped rather than logged as current, and the reason goes in the entry. Neither of those leaves a silent gap.
Every line is a pointer to something published elsewhere. If a source disappears, the entry stays with its dead link rather than being quietly removed, because the fact that it was reported is itself part of the record.