# Lost Ark GPD chart — methodology What every progression system costs, and what it buys, on one scale: **gold per 1% damage**. First build covers **supports**. ## The damage axis Damage stacks multiplicatively, so scores are logs and add up (the house model, shared with the accessory and astrogem calculators): ``` D = 100 * ln(multiplier) ~= percent damage, exactly additive ``` A support does not deal the damage; it hands damage to the party. So a support step is scored by what it adds to a damage dealer, above a support wearing nothing — the accessory calculator's model, ported in `model/support.js`: ``` Q = 100 * ln( ap * brand * identity ) ``` | Channel | What it is | Reached by | |---|---|---| | `ap` | the ally attack-power buff: the support gives away `0.22 * (1 + allyAtkEnh)` of its own base attack power | **honing, karma**, earring weapon power, ring ally-atk lines | | `brand` | 10% damage scaled by brand power | neck stigma lines, ark grid, karma ranks | | `identity` | Serenade + Major Chord + T-skill, one shared bracket, diluted by the dealer's own additional damage | ring ally-damage lines, ark grid, spec | Only `ap` touches the support's own gear. That is the whole reason honing and karma reach the party at all. ## The gold axis ``` gold per 1% damage = gold / (damage% * partySize) ``` `partySize` defaults to 3 — a support's 1% lands on every dealer in the party. This matches the astrogem calculator's `SUPPORT_GPD_MULTIPLIER`, so numbers here and there are on the same scale. ## The support's own stats Assembled the way bebkok's support buff calculator does it (sheet `1le-LqVr9l4dXxBDlPaSMNpf6tDVfvfsFE_QRIAmVONE`, tab "Sup buff calc v3.81", cells DN20..DP29). Flat sources are summed first and the percentage pool multiplies the lot — a flat weapon-power line is amplified by every weapon-power percent you own: ``` total WP = (weapon piece + accessory flats + ark grid core + feast + bracelet) * (1 + earrings% + karma% + ark grid core%) total MS = (five armour pieces + accessories + roster + level + food) * (1 + skin% + stronghold%) basic AP = sqrt(total WP * total MS / 6) * (1 + stone% + gem%) ``` | Weapon power | Default | |---|---| | accessory Weapon Power+ lines | 2,400 flat | | ark grid weapon core (Ancient) | 3,900 flat and 2.94% (0.75 + 1.50 + 0.23×3) | | feast | 2,400 flat | | earrings, two high lines | 6.00% | | Karmic Enlightenment | 0.10% per level | Percentages are **one additive pool**, so each further karma level lands on a bigger denominator and is worth slightly less: 0.01382% party damage at level 21→22 against 0.01375% at 29→30. `tools/verify.js` reproduces bebkok's own saved character exactly — total weapon power 226,085.218, total main stat 687,158.89, basic attack power 184,566 — and fails loudly if the model drifts. One divergence from the accessory calculator, taken deliberately: bebkok puts an attack-power percentage (ability stone 1.5% plus damage gems 13.2%) on the support's *own* base attack power, which scales the buff it hands out. The accessory calculator leaves that out. This tool follows bebkok, which raises every ap-channel step by roughly a tenth. A second deliberate divergence, ruled on 2026-08-16: the ark grid packer taxes any core under 17 points at a flat -3% / -6% / -9% of total damage (14-16 / 10-13 / 0-9). The game's real loss, read off bebkok's "Ark Grid Cores" tab and run through this model's buff math, is smaller and lumpier — about 1.0-1.2% for the Brave order cores (Serenade-scoped 17-options on Shizu's bard; Seraphic AP builds run ~1.6-1.8%), 0.5-0.9% for the chaos cores with the Echoing Brand damage-taken rider priced in. The Buckshot meter core uses the meter rulings on record — Buckshot is 20% of meter generation and meter converts to Serenade uptime at half effectiveness, both fixed by Shizu 2026-08-16 (the factor is the accessory calculator's documented bar-step approximation) — which puts its 17-option at 0.16 and its 14-option (the 500-meter counter proc) at ~0.3: the one core whose cliff sits at 14, not 17. Everywhere else it is a cliff at 17, nearly flat below. Shizu keeps the harsh flat tax on purpose: 17+ points per core is treated as table stakes, not a trade the optimizer may sell. The tax lives in the packer's objective only; reported damage assumes the standard away-team of 17-point cores it enforces. ## Where the game data comes from `tools/fetch-game-data.py` pulls Maxroll's planner feed (`assets-ng.maxroll.gg/laplanner`) — the tables their upgrade calculator itself runs on, tracking the July 2026 patch — and bakes the small derived files under `data/`. Big downloads are cached under `tools/.cache/` and not committed. **Cross-check that runs on every fetch:** the `itemLevel` table reproduces bebkok's Serca gear sheet exactly (chest 57,614 at +0 and 111,477 at +25; weapon 124,793 and 241,367; gloves 86,421 and 167,216). That is the same baseline the bracelet calculator uses, from an independent source. ## Honing — T4 Upper (1675), normal track Two rows: **all five armour pieces moving together**, and **the weapon on its own**. Steps run +11→+12 up to +24→+25. Rates are 0.01% units. A tap lands at ``` p = base + min(fails * failBonus, failMax) + juice ``` Juice is Lava's Breath (weapon) or Glacier's Breath (armour), `rate` each up to `juiceMax`. Every failure banks artisan energy; once the success rates already spent add up to the artisan threshold (215.00%), the next tap is guaranteed. "Average scenario, optimal" means expected gold under the cheapest constant breath count, solved per level rather than assumed. It matters: on armour at +11→+14 and +16→+17 the optimum buys **no** breath, because at 150g the juice costs more than the taps it saves. Material prices are editable and each material has an on/off switch, matching the upgrade calculator — off means the player has it bound and pays nothing. Defaults: | Material | Unit | Default | |---|---|---| | Destiny Shard Pouch S / M / L | 1,000 / 2,000 / 3,000 shards | 0, off | | Superior Abidos Fusion Material | each | 125 | | Destiny Crystallized Destruction Stone | per 100 | 1,800 | | Destiny Crystallized Guardian Stone | per 100 | 30 | | Great Destiny Leapstone | each | 25 | | Lava's Breath | each | 300 | | Glacier's Breath | each | 150 | The step's damage is measured against a self-consistent character: the rest of the gear sits at the step's starting level, and only the track under test moves. ## Karma Only **Karmic Enlightenment** counts for a support: **Weapon Power %, +0.10 per level**, 2.10% at level 21 rising to 3.00% at level 30. It feeds the same `ap` channel as weapon honing, through `wpPct`. Ranks 1–5 cap at level 21; rank 6 carries 21→30. Evolution's rank bonus also grants brand power (+1% per rank, 6% at rank 6), which is already inside the accessory calculator's brand base. Karmic Evolution (Max HP) and Karmic Leap (Ultimate Awakening Damage) are worth nothing to a support's party contribution and are not charted. ## Character lookup The lookup places one character on every ladder. Nothing is guessed: a system the pull cannot read keeps its row in the gear list and says why. **Two sources, because neither carries a whole character.** | | astrogem Worker | bracelet Worker | |---|---|---| | sign-in | required | none | | accessory grinding lines | yes | no | | accessory main stat | no | yes | | ark grid gems, skill gems | yes | gem levels only | | bracelet | the trait pair only | the raw payload | | honing, karma, stone | yes | yes, per piece | Both are asked at once and the answers merge, the richer field winning. **The bracelet is scored by the bracelet calculator, not by this page.** The raw payload is decoded with `Bracelet.decodeBibleBracelet` and scored with `Subrank.braceletScore` — the same call `tools/build-bracelet-rows-dps.js` used to cut the ladder, so the letter here and the letter there cannot disagree. The rung is then the last one whose band the character has reached. Before 2026-08-26 the bracelet was placed by the SUM OF ITS TWO COMBAT TRAITS against each rung's `hit.stats`, and that field is the rung's *example* bracelet, not a threshold. Every DPS rung shows 100/100 or better, so a 115/83 bracelet failed all ten and read "worse than C+" while the calculator had it at A-. The three effect lines — most of what a bracelet is worth — were never read. **Accessories are placed through the ladder's own damage table.** A piece is read as a configuration — the two primary tiers, the flat roll (`atk-` and `wpn-` kept apart on the DPS axis), and the main-stat quintile — and looked up in `data/accessory-scores.json` (support) or `data/accessory-configs-dps.json` (DPS). Each rung's `to` label is read back into the same configuration and looked up the same way, so the two are on one scale. Rings and earrings are placed on the WEAKER of the pair, because that is the piece the next step buys. A main stat the pull did not carry is taken as the middle quintile and the row says so. **The gear list's accessory rungs are the readable projection, not the chart's frontier.** The chart ranks accessories on the cost frontier — the cheapest configuration at every damage level — which is right for pricing and unreadable as advice: it wanders through flat rolls and main-stat quintiles, so the next step off a high-main-stat piece can be a LOW-main-stat one. The gear list fixes the main stat at its top quintile and drops the flat roll, leaving one plain primary-line step per rung, and the prices are the same lattice deltas the chart's rows are built from, so a rung here is still comparable to a honing or bracelet rung. Everything under the market's price floor is the ladder's base rather than a rung, which is the chart's own "nothing worth pricing" rule. Your own piece is a rung on that ladder. Both prices are measured against the piece actually worn — flat roll and main stat included — so a good piece is neither charged twice nor credited to a rung nobody bought. **The ability stone comes from the astrogem pull, never from node counts alone.** The bracelet Worker sends `stoneNodes`: every engraving's node count, the malus included, sorted high to low. The top two are therefore not reliably the two combat engravings — a 9/6 stone whose malus sits at 7 reads `[9,7,6]`. That array is used only when nothing else carries the stone, and the row says it is a guess. **Freshness is a reading like any other, and it is shown.** Both workers cache, independently, and neither auto-refetches a stale record — that is the bracelet worker's own written policy, because background churn is the upstream load these tools promised not to generate. So the lookup does two things instead. The older record never overwrites the newer one: honing, karma and the stone are read from whichever pull is fresher, and a field the astrogem pull read exactly is only displaced by a strictly newer bracelet pull. And the gear list says how old the reading is — on the row, and once at the top with a re-pull link. This was not academic. White's bracelet-worker record was 15 days old and had the character at +20 weapon and +19 armour; a re-pull returned +22 and +21. The stale record was also winning over a fresher astrogem pull, so the page had no way back to the truth. A re-pull cannot beat lostark.bible's own snapshot either, so when THAT date is old the header says so and stops promising a fix. The support bracelet rungs still run about one band hot — the ladder's own damage came from this repo's simplified support model rather than from `jointScore` (docs/RESUME.md, open work #1). The character's letter is right; the rung it lands on inherits that skew, and the gear list says so on the row. ## Open questions 1. **The quality block.** Maxroll carries a second per-honing-level stat block beside the honing recipe, and their item tooltip adds it to the ilvl base. On that basis a +25 chest reads 142,267 main stat rather than 111,477 and every gain rises about 67%. Default is off (`useQualityBlock: false`), matching bebkok and the bracelet calculator. Waiting on a reading from a real +25 piece. 2. **Karma level costs.** The data lists 1 Destiny Stone + 900 gold per attempt with a `prob` falling from 20% at level 21 to 0.25% at 29 and **0** at 30, plus a `care` field at about half of `prob`. A 0% success at the last level cannot be a plain roll, so `care` is presumably the pity path. Unresolved.