Random options reveal the quality of every roll, the damage formula page reads like a proof, and its calculator mirrors the game's own screen.
On a piece with random options, every stat now unfolds into five tiers, each with its own range and its real odds: the best one lands only 6% of the time when the stat rolls, against 32% for the most common. The damage formula page reorganises around its proof — a Measured, Datamined or Untested badge on every term, the sealed validation table moved up to second position, and a « Scope & honesty » section owned at the end of the page. Its calculator adopts the semantics of the game's own screen, its labels come from the official text of all thirteen languages, and its twenty-two boss presets are generated from the site's data. On the fixes side, a hero's element follows the weapon of its slot on the hub cards, ten of Durak's stat columns are realigned on readings taken in game, the account menu opens at full width even when the rail is collapsed, and the Timespace mechanic badges fit on a single chip in every language. Under the hood, images already at the right size no longer go through the optimiser, and a detector reports event-loop stalls.
详情(7)
- 💎 Random options — roll quality, tier by tier — A piece with random options announced the odds of rolling each stat and the highest value reachable. Side by side, the two numbers read as a promise: the ceiling looked as frequent as the stat itself. Every stat now unfolds into five tiers, each with its own range and its absolute probability — the one that combines the odds of rolling the stat with the tier's share. The game numbers them against intuition: tier 1 is both the strongest and the rarest, landing only 6% of the time when the stat rolls, against 32% for the most common one. Since the weight profile is identical across every engraving pool in the game, the legend is derived from the data and named once at the top of the panel. 152 pieces carry it.
- 🧮 Damage formula — the page reads like a proof — The page reorganises around what it can demonstrate. A banner opens on « The damage formula, measured » and its four key figures; every term and every constant carries a badge — Measured, Datamined or Untested; the sealed validation table moves up to second position, with its eight damage values predicted before being filmed, all landing within ±0.014%. Element, burst and stacks are grouped together, the FAQ becomes accordions, a table of contents follows the scroll, and the old « Limits » section becomes a « Scope & honesty » one, owned at the end of the page. The defense term is refined along the way: K is 5,128.
- 🎯 Damage calculator — fields mirror the game's own screen — Attack fields now follow the semantics of the game's own screen, with no reinterpretation, and their labels are copied from the official text in all thirteen languages. The flow reads in three steps — hero, target, skills — under a results banner that follows the scroll, where a skill pins on click. The twenty-two boss presets are generated from the site's data, at the latest world level for field bosses and at the top tier for dungeons and raids, and they fill in Defense too. Measured values that contradict the client tables are annotated as such.
- ⚔️ Team Comps — a hero's element follows its weapon — On the hub cards, a team built as mono-element could show four different elements. The data was never lost: the choice is recorded on the comp, and the detail view rendered it correctly. A hero's element depends on the weapon variant of its slot, not on its character page — yet two surfaces were still reading the per-character column, which only describes the default variant. The rule now lives in a single module that all three surfaces consume, and the element leaves the payloads keyed by slug.
- 🗡️ 「Corrupted」 Barbarian Chieftain Durak — stats aligned with the game — Durak is the only replay boss whose scaling is computed server-side: its client tables carry nothing to rebuild it from. It was therefore approximated by the health-point ratio applied uniformly, which over-estimated attack by 26 to 28% and defense by 40 to 43%, and left eight columns wrong. The tier profile of a sibling boss, named explicitly rather than guessed, is now replicated and carried by the scale measured in game: ten columns regain their value across the three difficulties. Independent control, the profile having never been fitted on them: the critical damage resistance produced lands within 0.2% of the three values read in combat.
- ⚡ Images — served without going through the optimiser — The image optimiser runs inside the process that serves the pages: while it encodes, it serves nothing else. Companion portraits are 512 × 512 for about twenty kilobytes and were being requested at 3840 wide for a 200-pixel render — paying two and a half times the encoding time to produce a file larger than the original. Assets already at the right size are now served as they are: on the Companions page, 933 possible conversions drop to 21. A guard records every exception with the measurement that justifies it.
- 🦅 Hawk Pass — the season box gets its name, and leads to its season — The season 4 reward box of the Hawk Pass had no readable name: it fell through to the generic fallback and published its technical key in title case across the 24 weapon pages of the pool. Its parent row is discarded by the item name deduplication — the only row in the game in that situation — so the naming cascade had no path left. A fourth path now reads the title from the Hawk Pass table row that references the box, in all thirteen languages, and future seasons are covered without coming back to it. The source was also a dead end: the same season identifier now links out to the reward tab of the season concerned. A box left unnamed now raises a build warning instead of surfacing on a page.