A map of what municipalities publish
Muniscan starts with a bounded question: what public digital systems can we observe from the links Peruvian municipalities declare on gob.pe?
The first run collected and classified those links for 1,794 municipalities. It separated municipal domains from central government platforms, published the raw snapshot, and documented the scoring method. The result was useful, but still static. One successful census says nothing about whether the instrument can run again.
The useful failure
The next scans returned much smaller municipal sets. Publishing either result would have made hundreds of municipalities appear to vanish overnight.
They had not vanished. The upstream directory had simply produced an incomplete discovery set.
Muniscan’s completeness guard blocked both releases. That was the right failure. More importantly, it revealed the wrong mental model: the current directory response was being treated as the whole known universe.
Discovery is an observation. It is not memory.
Giving the pipeline memory
The repair made recurrence stateful. Each scan now unions current discovery with the latest published census, then fetches every municipality again. A restored municipality can return fresh data, or an explicit HTTP error, but it cannot silently disappear because one discovery run was weak.
The next complete run scored 1,871 municipalities:
| Snapshot | Municipalities | Change |
|---|---|---|
| 2026-08-08 | 1,794 | first census |
| 2026-08-21 | 1,871 | +78 added, −1 explicit removal |
The second snapshot passed the same completeness gate without a force flag. That demonstrates one successful recurrence. It does not yet demonstrate stable monthly operation.
What the score means
The score summarizes visible links and hostname classifications in one frozen snapshot. It helps inspect the shape of a municipality’s declared digital surface.
It does not establish ownership, uptime, accessibility, security, legal compliance, service quality, or institutional performance. We also published a stratified 100-domain HTTP sample, but that remains a point-in-time observation, not an availability monitor.
That boundary matters more than the leaderboard. A public measurement is only useful when readers can tell what would falsify it and what it cannot support.
The public instrument
The live atlas now has its own domain and visual identity. It is where anyone can search the census, inspect the latest snapshot, download the data, and follow the method.
The source remains open:
The next test is intentionally boring: keep running it. Enough complete cycles would let us distinguish a recurring instrument from one successful rerun. Until then, the honest result is smaller and more useful: the census can recur, its failures are visible, and incomplete data cannot quietly rewrite the map.