The Well-Run Forge

SotA's crafting UI finished its move to UI Toolkit this month: the station window with its three panes, the recipe book, and the enchant and masterwork effect picker, all sharing one recipe card. Underneath sits one of the deeper crafting simulations in the genre. This document surveys what the rest of the genre learned about crafting interfaces, measures our window against it, and proposes twenty-two improvements, most of them zero-server-cost, each grounded in the actual classes that would carry them. Per-school balance is out of scope; each crafting tree gets its own deep dive later.

Executive summary

Ultima Online shipped tactile, simulation-first crafting in 1997 and then spent twenty-five years retrofitting the interface. Publish 14 (2001) alone added the unified crafting gump, your exact success percentage shown before committing, backpack material pooling, Make Last and Make Number — and UO players are still asking for "Make Last Number" two decades later. The genre then built the modern canon on top: crafting straight from storage (GW2, ESO, WoW, New World), the recipe book as a progression artifact (FFXIV), one-click subcomponent crafting (GW2), and player order economies (UO's Bulk Order Deeds, ESO's writs, WoW Dragonflight's crafting orders). SotA inherited UO's put-the-ingredients-on-the-table simulation, and almost none of the retrofits.

The sharpest fact this review found: the client already knows nearly everything the UI refuses to say. All 3,283 recipes with every field load client-side at startup; the player's known-recipe set syncs once at login; the success-chance formula is byte-identical shared code, and the recipe cards already run it locally every frame. Yet the station window waits on a server round-trip per table change to print the same number, the recipe book hides percentages away from stations, search matches recipe names only, and the fresh UITK port dropped item tooltips from both crafting grids.

Measured against the canon, our gaps cluster in three places:

1. Round-trips the client does not need — the summary waits on the server, the book hides the odds, discovery stays silent, all on data that is already client-resident.
2. Repetition without memory — no Craft Again, no Make Max, no queue, quantity resets, no favorites, no record of what you have crafted; every batch beyond fifty is manual re-pressing.
3. The recipe graph is invisible — search cannot see ingredients or results, nothing links "created by" to "used in", there is no materials breakdown, and materials are locked to the backpack (the lot-supply feature, in flight, is the first crack in that wall).

Twenty-two proposals follow. Seventeen are pure client work with zero new server traffic. Four need server support, each flagged with its cost. One is a directional design for the crafting-orders economy.

Where we are today

The crafting family finished its UI Toolkit conversion this month: the station window, the recipe book, and the enchant and masterwork effect picker. One recipe card renders in both the station list and the book; ingredient rows already tint three ways (carried, lot supply, short) with live "have/need" counts; the Only Craftable filter and the new craft-from-lot-storage toggle are in. Enchant, masterwork, socket, pattern, and brew need no buttons of their own: they are recipes — put the item and reagents on the table and Craft routes the action. This document builds on all of that rather than re-proposing any of it.

• 1Name-only filter. The Filter box matches the recipe's name. Searching "iron ore" (an ingredient) or "longsword" (a result) finds nothing. Proposal S8.
• 2No tooltips anywhere. Neither item grid showed hover info (the port dropped the wiring), and hovering a recipe row showed only its name. Fixed: proposals S1 and S2 both shipped, and recipe hovers now carry the result, the tier, the school and every ingredient with counts.
• 3One fixed sort, no pinning, no counts. Finding your six daily recipes in a 58-card list is scrolling. Proposals S6, S7, S15.
• 4Silent shortfalls. Double-clicking a recipe you lack materials for aborts with a chat line; the window itself never says what is missing. Proposal S5.
• 5The round-trip summary. Every table change is a server conversation before "84% success chance" appears — using math the recipe cards beside it already run locally. Proposal S10.
• 6Repetition tax. A batch is capped at fifty with no queue behind it. Most of this is fixed: S3 shipped, so the quantity is remembered per recipe and an empty table offers to repeat your last craft, and S4 shipped the Max button. What is left is the queue. Proposal S11.
• 7Three-station gating. Salvage, Repair, and even the Recipes button only exist at blacksmith, tailor, and carpentry stations; the book has no in-window entry at a smelter or kitchen. Proposal S16.
• 8The lot tint is the seed. The teal "(+2 lot)" counts come from the in-flight craft-from-lot-storage feature; bank and deeper storage stay invisible. Proposal S20 extends that feature, not this document.

The machinery every proposal plugs into

One session loop owns crafting: interacting with a station creates a server-side provisional bag for the table, then a watcher hashes the table contents and asks the server what they make on every change; the answer drives the summary text, the craft gates, and the Salvage and Repair enables. Pressing Craft sends one craft request per batch item, re-gathering ingredients from the backpack between items. The recipe catalog, the player's known-recipe list, the skill bonuses, and the success-chance formula all live client-side already; the UI controller exposes a clean view-surface seam that every proposal below extends. None of them forks the shared client/server crafting logic, and none touches the server's craft transaction itself.

flowchart LR
  A["Use a station\nprovisional table bag"] --> B["Drop ingredients\nequip tool"]
  B --> C["Table watcher\nhash on every change"]
  C -->|"today: round trip per change"| D["(Server recipe\nresolution)"]
  D --> E["Summary: recipe name\nsuccess + exceptional %"]
  E --> F["Craft pressed"]
  F --> G["(Server craft transaction\nauthoritative, per batch item)"]
  G --> H["Progress bar\nresult + chat line"]
  H --> C
  style D stroke:#a33d2e,stroke-width:2px
  style E stroke:#a33d2e,stroke-width:2px

Lessons: Ultima Online to modern

UO 1997 was SotA's diagnosis in miniature: a rich simulation — skill curves, material tiers, exceptional procs, maker's marks, a real player economy — behind a bare interface of tool-in-backpack menus, hidden odds, and materials destroyed on failure. Publish 14 (2001) changed no formulas and is remembered as the crafting update: one unified gump, your percentage chance shown before committing, backpack pooling ("no materials will need to be targeted"), Make Last, Make Number, and docked Repair and Smelt verbs. Bulk Order Deeds arrived in the same publish and gave crafting a reason to be opened daily. The salvage bag (2007) turned the most repetitive action in the game into one context-menu click. Imbuing (2009) converted the runic-tool slot machine into a spent budget with shown odds. Every step was interface-and-loop work on an unchanged simulation.

What each game contributes

Ultima Online: the template — the gump before the math, pool materials, show the number, give crafting a goal loop, bulk verbs, and ship the batch primitive completely (UO never did; "Make Last Number" is a twenty-year-old open request). Star Wars Galaxies: the closest kin to SotA's player-economy DNA — schematics as knowledge, experimentation as directed enhancement, factories as the proof that deep crafting and mass production can coexist, practice mode as the market-safe trainer. FFXIV: the crafting log with first-craft checkmarks makes breadth itself a goal; right-click a material to find its recipes; Teamcraft's existence is the strongest evidence that multi-level crafting lists belong in the client. WoW Dragonflight: the expected-quality bar is the gold standard of pre-commit transparency, and crafting orders solved "find me a crafter" after nineteen years. ESO: daily writs and an improvement dial that reaches 100% if you pay for it; the craft bag is the mechanic to copy and the paywall not to. GW2: the best crafting panel shipped — storage-fed counts, inline subcomponent crafting, discovery that tells you when you are warm. New World: storage-shed crafting with a spatial cost, and bulk salvage with the lock rail UO's salvage bag lacked.

Capability matrix

● native · ◔ partial or third-party · ○ absent. "SotA today" = the UITK crafting windows on the migration branch, 2026-07. WoW = Dragonflight and later.

The twenty-two proposals

Tier 1 · Quick wins — client-only, zero new server traffic

Tier 2 · Core upgrades

Tier 3 · Server-backed

Each of these lands with its data-migration ledger row, its shared-code mirror sync, and a tester task, one server change per commit.

Tier 4 · Directional

Considered and rejected

• ESO-style auto-routing craft bag — rejected in the inventory deep dive for economy distortion; reaffirmed here. The lot-supply feature and S20 attack the same pain from the retrieval side.
• Trial craft (FFXIV Trial Synthesis) — FFXIV rehearses a twelve-action rotation; a SotA craft is one roll whose odds S9 and S10 already display before you commit. A no-material rehearsal would rehearse nothing.
• Spend-to-guarantee and directed enhancement (WoW Concentration, ESO tempers, SWG experimentation, UO imbuing) — the right answers to masterwork and enchant variance, and pure per-tree balance; catalogued for the tree dives. UO's cash-shop guarantee item is the monetization line never to cross.
• Decorrelating the success and exceptional rolls — today one roll serves both thresholds, so exceptional results correlate with very low rolls; changing that is a balance decision inside the protected craft transaction. Documented for the tree dives.
• Player-to-player orders before NPC writs — escrow, scam surface, and commission plumbing on top of an unproven loop. S22 v2.
• A server-side gather op for storage pull — the lot-supply feature rejected it for safety reasons; that rejection stands for S20.
• Raising the ten-item table cap — S10 and S11 make the table a transient staging area; the cap stops mattering.
• Raising the per-request batch cap — S11's queue makes the cap invisible to players; raising it becomes per-tree balance.
• SWG-style factories on lots — the ideal end state for batch crafting and a natural player-housing fit, but XL on top of XL with market-flood risk. Revisit after S18 and S22.
• Auto-crafting intermediates without a click — surprise material consumption; S12 keeps the player in the loop, which is the documented lesson from GW2's own version.
• A dedicated shopping-list window — the drill-in covers it without a fourth pane.
• Waiting on list virtualization — the quick wins mitigate with debounce and caching; virtualization itself belongs to the inventory program, coordinated, not duplicated.

The composed window

Composition rules that keep twenty-two features from becoming twenty-two widgets: the Filter box is the query surface (S8 extends its matching; no second box appears); one options surface (a gear in the chrome) hosts every preference tail; tooltips and the summary block absorb every pure-information feature (S2, S5, S9, S10, S13, S14, S19); the queue lives inside the existing progress block (S11); the shopping list is a drill-in of the selected card (S12), never a fourth pane; favorites are a star on the card plus a category entry (S7). Net new widgets across the whole catalog: three — the Sort dropdown, the Max button, and the options gear.

• 1The chrome gear is the one options surface: quantity memory, can-make counts, skill tint, queue confirm, tooltips. S3, S15, S6, S11, S1.
• 2Favorites live in the existing category dropdown and as the card star; the Sort dropdown is new widget one. S7, S6.
• 3The same Filter box now matches "hilt" against ingredient and result names. S8.
• 4The Materials selector is the lot-storage toggle grown up; until S20 it renders exactly as today's checkbox. S20.
• 5The selected card's drill-in is the shopping list; "Craft this next" jumps to the subcomponent. S12.
• 6The summary block absorbs the instant match, the record line, the shortfall, and the discovery whisper. S10, S19, S5, S14.
• 7Max is new widget two; the queue lines live in the progress block that already swaps in during a craft. S4, S11.
• 8Tooltips return on both grids, carrying the created-by and used-in links. S1, S13.
• 9The Show dropdown finally knows what a blacksmith cares about. S16.

Zone anatomy

chrome        title + options gear (the ONE options surface) + pin + close
left pane     query row (category + sort) / filter row (THE query surface)
              source row (materials selector + only-craftable)
              card roster: favorites first, drill-in lives on the selected card
center pane   station vignette / table grid (empty state hosts Craft Again)
              capacity + equipped tool
              summary: recipe, odds, record, shortfall, discovery
              stepper + Max + Craft / Salvage Repair Recipes | Take All
              progress block: bar + queue lines (swaps with the controls)
right pane    Show dropdown (station-aware) / backpack grid, tooltips on

The recipe book, evolved in place

Same book, four upgrades: search that sees ingredients and results (S8), a Favorites chapter first in the contents (S7), success chances away from stations with their qualifier (S9), and reference links that page the book (S13). Cards stay draggable for teaching and vendor listing.

Sequencing

flowchart TD
  P1["Phase 1: S1 S3 S4 S5 S7 S9"] --> P2["Phase 2: S2 S6 S8 S13 S15"]
  RIDX["Reverse recipe index"] --> S8x["S8"] & S12x["S12"] & S13x["S13"]
  S10x["S10 local resolver"] --> S11x["S11 queue"] & S14x["S14 discovery"] & S15x["S15 counts"]
  S8x --> S12x
  S11x --> S17x["S17 bulk salvage"]
  S7x["S7 favorites"] --> S18x["S18 per-character state"]
  LOT["Lot-supply feature (in flight)"] --> S20x["S20 bank pull"]
  S18x --> S22x["S22 orders board"]
  S12x -.-> S22x

Appendix

A1 · Zero-RPC proof

Every display datum in the seventeen client-only proposals is already client-resident: the full recipe catalog with ingredients, results, tiers, difficulties, schools, discoverability and fuel flags (loaded once at startup); the known-recipe set with teachability (synced at login); every skill level and crafting bonus stat (local character state); the success, exceptional, and recovery formulas (shared code, the server's own math); recursive backpack ingredient counts plus the lot-supply advisory counts; the result-to-recipes reverse lookup; and the ingredient-hash function that names any hypothetical table's recipe. Only actions transact. The one genuinely server-side display datum in this document is S19's crafted-count, which is why S19 sits in Tier 3.

A2 · Constraints honored

Shared crafting logic extended, never forked; the server craft transaction untouched by every item here; no new item-mutation paths anywhere (S17 and S20 sequence existing operations); every bulk action calls the inventory program's bulk-protection predicate; per-card computeds cache against item signatures because cards re-bind on any grid change; search debounces; the one recipe card keeps rendering identically in station and book; new reads ride a regenerated RPC table; every database change lands with its migrations-ledger row in the same commit; and every player-visible sample string in this document follows the plain-voice rules.

A3 · Test surface

Pure-logic targets for the EditMode suite: the S10 resolver (parity replay of recorded server validation answers across the full recipe database), the S11 queue state machine, S12's cycle-safe graph walk, S8's matcher and index, S6's comparators, and S14's partial-match counter. The recipe card and book page models already established the pattern, and the existing thirty-plus crafting test files stay green throughout.

A4 · Sources

UO: UOGuide (Publish 14, Bulk Order Deeds, runic tools, Salvage Bag, imbuing), the official UO wiki, and the long-running "Make Last Number" forum request. FFXIV: crafting log, Trial Synthesis (7.4), the Endwalker HQ-material removal, Teamcraft. WoW: the Dragonflight professions overhaul, The War Within's Concentration, Warband-bank crafting orders, and the fifteen-year addon canon. ESO: craft bag, trait research, writs. GW2: crafting, Material Storage, and the Discovery panel. New World: settlement storage, gear-score bands, bulk salvage. EQ2: tradeskill events, Work and Rush Orders. Valheim (the CraftFromContainers mod as a shipped spec) and Enshrouded's Magic Chest. SWG: schematics, experimentation, factories, practice mode. Vintage Story: the Survival Handbook.

Written against the unity-6.5 branch, 2026-07-26. Per-school deep dives (blacksmithing, tailoring, alchemy, cooking, agriculture and the rest) follow separately. Comments and votes below reach the team directly.

General comments

No account needed. Every proposal above also has its own Discussion box for talking about that one alone; this thread is for the plan as a whole. Comments are plain text, held to basic decency, and may be trimmed.