Release 1.1.0

run the tests before the release touches anything, so a failure leaves the version and the changelog as they were

keep the test run inside the project's own directory: it used to walk the module corpus, call other projects' snapshots obsolete and finish non-zero on a green run

skip a checkout the registry does not know about instead of stopping the whole build on it, and say in the log how many were skipped

release the code on its own tag: a version, a changelog section and release notes, separate from the dated data release that carries the site

clear the footer of eighteen links labelled empty that led back to the front page, and of a resource group whose three links already stood under Learning

build the site with a newer toolchain, so a dated document that declares its own address counts as an entry again. The monthly ecosystem digest reaches the front page instead of an empty block

keep test-directory samples off the site. They are fixtures for other people's documentation generators, not documentation of the module. A README among them is still taken.

carry what a module declares next to the measured pages: the units it ships and what it depends on. A rating computed later from a stored record can then be reproduced.

refuse to write a corpus record when a registry it reads is missing or does not parse. A stored record would otherwise read as though no module declared anything.

note whether a page shows the output of its example: a worked example carries both the code and what it prints, a bare snippet only the code

list the selected files separated by NUL and skip the ones whose name holds a control character: one registry entry carries a newline inside its own name, which tore a path in two and lost the file without a word

name every file the parser could not read, instead of sending the whole complaint to nowhere

measure the corpus while parsing it: markup census with the module reach of every construct, per-page facts, unparsed files and the settings the numbers were taken under

take documentation out of module sources: a source file gets a page when its pod says something the README does not

declare the packages the repository's own scripts import, and restore the dependency tree after a site build detaches it

create the release before recording the snapshot, and let a failed snapshot push not fail the run

let a second data release on the same day get its own tag instead of failing

build a page for every module in the registry, including the 139 that ship no documentation

clean stray markdown and windows backslashes out of link targets on module pages

publish a data release weekly, not only on demand

link module names in the change report to their pages, for modules that have one

build and publish a dated data release from CI — site archive, offline bundle and server image, with the version diff as release notes

resolve a snapshot name against the snapshots directory whichever way it is given

fetch module sources over https — the git protocol they list has been switched off since 2022

stop the refresh when more than a twentieth of the module sources have no checkout at all; a checkout that merely failed to update is reported, not fatal

credit the tools in the site template footer, where the rest of the footer lives

record a dated snapshot of module versions on every data refresh, and compare two snapshots into release notes

stop storing generated corpus trees in the repository; a fast rebuild now says so when they are missing

document the offline bundle in the mcp server readme

bump the @podlite toolchain to the v0.0.77 release

treat any url scheme as external, so links like irc: no longer get a section prefix

strip markdown left inside a link target; the parser now hands over the target itself

quote the module page url so names with spaces keep their full path

skip modules whose name carries control characters

reduce pages size up to 2-3 times

refactor components to use =include feature