Fixed — (Seen)
- A range slider dragged to the top could not be moved again. Drag the low handle all the way up and both handles pin. The two range inputs overlap and only their thumbs take pointer events, so every later grab landed on whichever was painted on top — hi, by DOM order. hi cannot go up, and the input handler clamps it straight back down to lo, while lo, the one handle that could escape, sat underneath. Tonnage, PV and BV were all trapped; the bottom end always recovered, because there hi on top is correct. Present since the sliders were pasted from the prototype. Fixed by z-order alone: when the thumbs coincide at the top of the step grid, lo is raised. Deliberately not a forced gap between the handles, which would have made “only 100-ton ‘Mechs” un-selectable. Verified live at the three values that trapped — 100 t, 69 PV, 3610 BV. Live.
- The tonnage “no value” control vanished while its filter kept filtering. It shared a visibility gate with the tonnage slider, so refusing the slider hid the control with its state still live: pick Infantry, set the control to ONLY, then add Aerospace Craft, and the results stayed narrow to units with no recorded tonnage behind a control that was no longer on screen — no way to see it was on & no way to clear it. The two are now independent: only the slider is withheld. The tonnage range still clears on refusal. Live.
- The refusal message named a cause the user could see was wrong. It’d say: “1–100 t is too wide a span to slide” — but 1–100 is a span the slider handles for BattleMechs. What actually fails is the step count: a type with fractional tonnages drops the step to 0.1, so 1–100 needs ~990 increments against a limit of 500. Read as a bug in the site rather than a property of the selection. Verbiage now states what the gate actually compared: “Too many steps between 1 and 2,400,000 for the selected unit types to show a slider (23,999,990 steps of 0.1 t; the limit is 500).” Two cases were swept into the same wrong sentence gets sorted (sorta): every unit sharing a single tonnage, and no unit having a recorded tonnage — neither of which narrowing can help, so neither talks about it. Live.
Fixed — (Seen)
- Free-text search returned zero results for every query. Typing a unit name into either search box produced “No units found” on the site. Caused (by me) the previous day: a new tonnage-refusal message read two fields that are absent when no era is selected, threw a TypeError inside the page’s refresh, and stopped it before the results were assigned. The page looked normal, so the failure read as a data problem rather than a crash. Every check run that day had an era in the URL, which is why it survived verification. Live. (And I’ve added a positive verification to catch this in the future.)
- The search index intermittently failed to load at all. Hosting’s cache was replaying an empty 304 for the manifest.json—the file the search reads first, from which every filename is derived—so an affected visitor got a message of “Could not load the unit index.” Two of twelve requests, measured by probe. Same defect as the broken unit art I uncovered a few days earlier, but on a second path. Exposure tracks how often a file is revalidated, and the manifest’s 30-second lifetime makes it the most revalidated file on the site. Additional positive verification added to the path. Live.
- A unit page went from about 3.24 MB to 828 KB. (::facepalm::) Two causes, both measured by probe and stats. ForcePack and sourcebook cover thumbnails were being served as full-size PNGs—908 KB and 975 KB to fill a 32×44 box—that are now 4.4 KB and 4.6 KB. The Tabler icon font shipped its entire set, 449 KB, for the 36 standard glyphs the site draws; it is now a 7 KB subset. There’s a new guard that reads the shipped font and fails if an icon has no glyph. Across the catalogue, 319.9 MB of superseded originals were reclaimed. Live.
- The force builder no longer throws on load. A coverage hint evaluated coverage.totals in an x-text while its neighboring x-show checked that coverage existed. Alpine evaluates x-text regardless of x-show. Identified by a new test rule rather than by a user. Lucky. Live.