max / makenotwork

Seal the hand-written stylesheets, and forward-fix the quasi pins Two halves of the conversion's cost were already sealed and the third was not. frontend_globals ratchets the script surface and DEAD_VOCABULARY_HIGH_WATER ratchets the generated one; the hand-written sheets, which are 322KB of the 345KB this server ships by hand, had no number. So "we converted a screen" and "we removed its cost" were the same claim, and only the first had happened: four conversion batches landed and the sheet did not move. tests/frontend_payload.rs seals it in whole kibibytes, at 337. Two-sided: a rise wants an argument in the commit that raises it, and a fall must lower the seal or the next batch grows back into the room. Kibibytes rather than bytes so ordinary editing is free and a deleted rule block is not. no-js.css was in neither guard. It is hand-authored CSS served to real visitors, so it joins HAND_WRITTEN_CSS, and it was outside the cache fingerprint the same way wizard.css and media-player.css once were, so a change to it served stale to anyone holding a cached copy. Both fixed, and the_two_lists_agree holds the seal's list and build.rs's together. The pins are why none of this could run. This server required quasi ^0.5 against a published 0.8.0 and could not resolve at all, so the repo did not build; makeover-layout ^0.23.0 then collided with quasi 0.8's ^0.26.1 through the links guard. Bumped to quasi 0.8 and makeover-layout 0.26.1 / makeover-build 0.37.0. No code changes were needed, which is what three un-forward-fixed minors cost to close. DEAD_VOCABULARY_HIGH_WATER 20 to 23, and the argument is the one the constant already documents twice: track-entry, track-slot and track-tick are emitted by quasi-webview from a description and are invisible to a scanner reading this repo. They had never been weighed here because the pin could not resolve.