The Company You Keep

SotA's party system is better built than its reputation: a dedicated social server, a twelve-member roster with leader roles, three loot rules with a roll threshold and a self-service opt-out, a party XP bonus, party chat, instance matchmaking that follows the leader, and a Looking For More window that finished its move to UI Toolkit this summer. And yet a party here still has no name, no memory past a twenty minute database timeout, no dot on the compass, no ready check, and no answer to the question a small population asks every night: who else is out there, and how do we find each other? This document surveys what the genre and its fiction learned about adventuring parties, measures our system against it, and proposes thirty-four improvements, each grounded in the actual classes that would carry them.

Executive summary

Ultima Online ran for thirty-one months without a party system at all, and it worked, because in a use-based, classless sandbox the rewards were already individual. When Publish 5 finally retrofitted parties in April 2000 it shipped a small, exact list: a roster, a chat channel, consent flags for looting, shared fame, and dots on the map. The genre then spent twenty-five years elaborating on that list, and the elaborations sort cleanly by population. Queues, role gates, and level penalties assume a deep pool of strangers; boards, mentor pairs, NPC companions, and named companies work with whoever is actually online. Shroud of the Avatar sits today almost exactly where UO sat in early 2000: the wire protocol is complete, the loot rules exist, a party finder is live and freshly rebuilt in UI Toolkit, and the classless flat XP split already lets a ten-year veteran and a week-old newcomer hunt side by side with no penalty. What is missing is everything that makes a party feel like a company rather than a shared damage ledger.

The gaps cluster in three places, and they are exactly the three promises on this page's tile:

1. Finding each other. The Looking For More system exists and is orphaned: nothing routes a player to it, listings have no activity tags, no time, no tavern presence, and there is no open-party or call-for-help path across the game's three multiplayer modes. The genre's answer at this population is boards and beacons, never queues.
2. Fairness you can see. XP splits evenly with a party bonus, but the twenty meter share radius is invisible and players discover it by silently earning nothing. Loot has three rules where the genre settled on five, and no receipt anywhere says what was shared and how.
3. Being a party. No name, no banner, no history; state dies on a twenty minute timeout. No target marks, ready checks, pins, or compass dots. Beyond standing in the same scene, there is almost nothing a party does as a party.

Thirty-four proposals follow. Eleven are small client-only changes, several of them plain bug fixes. Twelve are larger client work, all riding one existing extension point (the party member property bag) with no new network operations. Eleven need real server work, and each of those carries its exact cost in the open: which request group it rides, whether it touches the database, and what the migrations ledger owes.

The party today

Credit first, because there is a lot to credit. The party stack finished its UI Toolkit migration this summer: the HUD member column, the Social window's Party tab, the Looking For More window, the invitation dialog, the players-in-scene list, and the teleport picker are all rebuilt, with the party leader's star on nameplates and four EditMode test suites pinning the tricky logic. The loot roll opt-out that Tirrag asked for on the forum shipped in July, an instance-relabelling XP exploit was closed on both halves, and the Summon Party scroll, broken for years, casts again. The system underneath is a real one: parties live on their own social server with a twelve-member cap, leader roles, a sixty second invite timeout, automatic leader succession by seniority, and a matchmaking loop that walks the whole party into the leader's instance.

Here is what a player actually sees, and where it stops short. Four surfaces, none of them aware of the others:

• 1Out-of-scene members go dark. Cora's row shows a town name where her health bars were. Vitals only replicate inside the scene; the member property bag could carry them everywhere. Proposal S12.
• 2No party presence in the world. No compass dots, no minimap marks; the overworld marker is the last NGUI holdout in the party stack. UO had party map dots in the year 2000. Proposals S13, S14, S15.
• 3The finder is invisible. The Looking For More window works today and nothing in the game ever points a player at it. No slash command, no badge, no board metaphor, no tags. Proposals S10, S16.
• 4Three loot rules, one consent flag. Free-for-all, Roll, Master Looter, and the roll opt-out. No round-robin, no need-before-greed, and the rules live in a settings tab nobody reads aloud. Proposals S20, S26.
• 5Coordination verbs missing. No ready check, no countdown, no target marks, no pins. The canon has been settled since June 2006. Proposals S4, S5, S15.
• 6The invisible fairness rule. XP shares only inside twenty meters of the kill, and nothing anywhere shows it. Proposals S2, S3, S25.
• 7Leader-only everything. Only the leader invites; everyone else asks the leader to type. Proposal S24.
• 8No identity, no memory. The party has no name, and every trace of it expires from the database twenty minutes after the last member logs off. Proposals S19, S31, S32.
• 9Nothing to do together. No content anywhere answers party size with party challenge. The forum has asked for champion-spawn style escalation by name. Proposals S33, S34.

The machinery every proposal plugs into

Party state lives on the social server, one authoritative object per party on its own fiber, reached through the group server's request router. The game server never holds a roster: when a kill pays out, the master client sends the actor list and the server splits evenly, applies the party bonus terms, and fans out. The client's PartyManager owns everything the player sees, and per-member data travels in exactly one extension point: the member property bag, four keys today, published by each member and rebroadcast to the rest of the party unread by the server. Almost every proposal below is either a new bag key, a new message on the existing party chat control channel, or a new consumer of data already arriving.

flowchart LR
  A["PartyManager\n(client)"] --> B["GroupClientPeer\nrequest router"]
  B --> C["Party object\nsocial server, own fiber"]
  C --> D[("PartyMember +\nPartyStorage\nMongo, TTL 20 min")]
  C --> E["IPartyEvent fan-out\nto every member"]
  subgraph XP["the XP lane (separate server)"]
    F["AwardXPForDead\nmaster client"] --> G["ChangeCharacterExperience\nForCombat + actor list"]
    G --> H["GameImpl split loop\neven split + bonus terms"]
  end
  style C stroke:#c9a55a,stroke-width:2px
  style D stroke:#c9a55a,stroke-width:2px

Lessons: Ultima Online to modern

UO launched in September 1997 with no party system at all. Grouping meant standing near friends and trusting them, and the game's early flagging rules actively punished cooperation: healing a criminal made you a criminal, and helping the wrong person got you cut down by town guards. Guilds arrived in 1998 and carried group play for two years. Then Publish 5, in April 2000, retrofitted the entire party stack in one publish: a ten-member roster with leader-only adds, party chat with per-member tells, status bars built for fast support play, shared fame and karma at full value for everyone, dots on the radar map, and the most SotA-shaped idea in the whole survey, a per-player consent flag reading simply party can loot me. No XP math, no roles, no matchmaking. That minimal list carried the game for a decade.

UO's loot rights are the best-aged idea here: corpse access accrues from damage points, indirect contribution counts (a tamer's pet credits the tamer), and anyone within thirty percent of the top contributor loots. It took until Publish 74, in 2012, for healers and tanks to count equally, twelve years to learn that support play is contribution. And in 2024, UO's New Legacy season turned group identity itself into a reward, reserving the largest houses for guilds. Twenty-seven years on, the ancestor is still arguing that the group is the endgame.

What each game contributes

EverQuest: group XP as a strict mathematical win, and the social machinery of camps, pullers, and camp checks that a single server's reputation enforced without code. Also the cautionary trinity: three of fourteen classes mandatory for everything. City of Heroes: sidekicking, then Super Sidekicking in 2009, the whole team auto-set to the mission owner's level. In a shrinking population this is the difference between forty people online and forty people you can group with; SotA's classless design gets most of this for free and should say so out loud. WoW: the full arc, from LFG channel to Dungeon Finder to Raid Finder, then the slow regret, and the two features from 2006 nobody ever regretted: ready checks and raid marks. Its 2014 premade Group Finder quietly re-added the manual board it had replaced. FFXIV: Party Finder listings, post-duty commendations feeding a titled mentor rank, and Trusts, story NPCs as party members, the proof that NPC fill can be a narrative reward instead of a crutch. Guild Wars 2: no kill stealing, per-player nodes, everyone-tags credit, and the mentor tag with zero mechanics that people still wear. OSRS: when you cannot afford matchmaking, ship a board and a channel and let the community do the rest. Monster Hunter World: the SOS flare, converting solo players to co-op at the exact moment difficulty appears. Deep Rock Galactic and Sea of Thieves: rituals as retention, a salute, a shanty, a round of grog; the crew is the content. Baldur's Gate 3 and the classic CRPGs: the camp as social anchor, and the tavern, from Gilgamesh's in 1981 onward, as the named place where parties are formed. SotA has player-run taverns doing nothing mechanical, which reads less like a gap and more like an invitation.

And the fiction is unanimous. The Fellowship was chosen, named, and commissioned before it walked a mile. Kurosawa's samurai are recruited one by one for distinct competences, and Heihachi paints the company's banner: six circles and a triangle, flying over the village. Critical Role's Vox Machina named itself on stream, first draft rejected, and built an empire on the name. Roughly two thirds of tabletop campaigns begin in a tavern. What fiction says a party is, and what almost no MMO has ever shipped, is a named, persistent, heraldic thing with a founding story: a name, a motto, a sigil, a banner, a chronicle of deeds, and a table where it drinks. That is Tier 3 of this document.

Capability matrix

● native · ◔ partial · ○ absent. *UO has no XP: it shared fame and karma at full value, and its use-based skills sidestep the level-gap problem entirely, the same structural position SotA holds. †Within a party, SotA's even split already pays healers and tanks; credit for helping strangers is the life-magic page's open question (its S20), cited here rather than re-proposed. The queue row stays empty by design: the PvP page's ruling stands, below a few hundred concurrent players a clock beats a queue. True open-party exemplars, Warhammer Online 2008 and Monster Hunter's SOS flare, are cited on their cards.

The thirty-four proposals

Numbering runs S1 to S34 in build-order tiers. Tier 1 is client-only and small; Tier 2 is client work of real size, all of it riding the existing member property bag and party chat channel with no new network operations; Tier 3 needs the server, and every card there states its exact cost. Sample names in the mockups belong to real players whose forum requests shaped this page.

Tier 1 · Quick wins, client-only

Tier 2 · Core upgrades

Everything in this tier rides one of two existing rails: the per-member property bag (four keys today, published by each member, rebroadcast unread by the server) or the party chat control channel. No new network operations anywhere in the tier.

Tier 3 · Server-backed, each with its exact cost

Ground rules for this tier, stated once. A new request added to the existing party interface means a new opcode, a new entry in the existing party dispatch, and a regenerated RPC table, but no new interface wiring. A brand new interface group means manual dispatcher wiring in the group server's request router, the one step the code generator does not do. Any new server file must be registered in both server project files. Any durable database change owes the migrations ledger an entry in the same commit. Every card below names which of these it pays.

Considered and rejected

• Auto-matchmaking and dungeon queues. Standing law from the PvP page: below a few hundred concurrent players, a clock beats a queue. The board (S16, S27) and the flare (S18) answer the same need without the empty-queue failure mode.
• Role queues or trinity tags on the finder. Classless game; activity tags carry the intent without the lie.
• A party focus feed, or any cross-member focus grants. Closed on the focus page: the wire was deliberately severed years ago and stays severed.
• Click-to-cast party frames and mouseover healing. Closed on the life-magic page: healing happens in the world, not a grid. S8's row menu is management-only forever.
• Cross-party healing credit. Inside a party, the even split already pays support; credit for helping strangers is the life-magic page's own open vote, cited, not owned.
• Two-way level or skill sync. There is no server authority surface for capping effective skill, and nothing to cure: SotA has no level-gap penalty. The mentor pair (S22, S29) captures the value without the machinery.
• A party stash or shared escrow container. An item-escrow and duplication surface against a twenty minute party lifetime; the ledger (S21) delivers the fairness transparency without new ownership machinery.
• Party votes on decisions. At duo and trio scale, a chat message beats a ballot.
• A raid tier or alliances above twelve. The cap is uncontested at current concurrency, and band tokens already cover large cross-party ensembles; band beats party is design of record.
• Cross-school combo abilities. Party-combat verbs are arriving card by card on the skill pages (the shield wall, the traveler's chorus, the split heal); a combo system is an XL combat rework with no owner.
• Quest sharing and story sync. Belongs to the quests page, whose tile is live and voted; this page wires quest coordination through pins and callouts only.
• Master-loot councils. Fourteen years of genre drama; existing Master Looter stays as-is, joined by round-robin and need-before-greed.
• Teleport-to-banner rally. The server knows no positions; shipping it client-trust would be shipping an exploit. The banner stays a marker.
• Anything that requires a bard, or a fixed party size at a door. Bard-page law and the anti-pattern list; encounters scale, never gate.

The Company Window

The danger with thirty-four proposals is a screen with thirty-four widgets. The composition rule: everything lands in exactly three places. One window, the Company Window, four views under one header, replacing the scattered Party tab, the orphaned finder window, and the options fragments on three tabs. One satellite, the existing HUD column, which gains only glanceable glyphs and never casts. One world layer, the compass marks, pins, and the tavern board. Here is the window at its default size, wired to the real feature set:

Sequencing

flowchart TD
  S1["S1 plumbing repair"] --> T1["Tier 1 QoL wave"]
  S10["S10 surface the finder"] --> S16["S16 Adventurer's Board"]
  S16 --> S17["S17 open party"]
  S16 --> S18["S18 signal arrow"]
  S16 --> S27["S27 persistent board"]
  S5["S5 target marks"] --> S15["S15 pins + pings"]
  S13["S13 scene markers"] --> S15
  S13 --> S32b["S32 field banner render"]
  S12["S12 bag-key pattern"] --> S19["S19 named parties"]
  S12 --> S22["S22 mentor's mark"]
  S19 --> S30["S30 announce the company"]
  S19 --> S31["S31 Company charter"]
  S21["S21 ledger"] --> S28["S28 commendations"]
  S28 -.optional gate.-> S29["S29 mentor's kicker"]
  S22 --> S29
  S31 --> S32["S32 banner"]
  S11["S11 toast"] --> S23["S23 tavern muster"]
  S16 --> S23
  EB["embodied-bots epic"] -.gates.-> S34["S34 hired blades"]
  S34 -.serves.-> S33["S33 the muster"]
  A18["achievements announce card"] -.gates.-> S30

Appendix

A1 · Why Tiers 1 and 2 cost the server nothing

A2 · Constraints every proposal honors

• The game server holds no party roster and no player positions; nothing here asks it to learn either. Server-side additions are structural (a permission check, a bonus term, a payload field), never spatial.
• The member bag's existing keys are load-bearing and never renamed; new keys append. New replicated view properties, where ever needed, go last in channel order.
• The gold-share list and the roll list are deliberately separate builds and stay separate; the roll eligibility builder returns null, never an empty array.
• Rows and frames never cast; the life-magic and focus page rejections are treated as law throughout.
• Server work follows the house rules: new ops on existing groups where possible, dispatcher wiring named when unavoidable (S31 only), dual project registration for new files, a migrations ledger entry in the same commit as any durable change, RPC table regeneration on any op change.
• All sample strings in the mockups follow the plain-voice rules and can lift verbatim into the game.

A3 · Test surface

Four party suites exist and get extended rather than duplicated: instance ownership, loot-roll opt-out, HUD model, and finder conversion. The pure-logic pieces here, the rotation model, the flare and pin lifecycles, the receipt formatter, the commendation rate limits, the mentor predicates, are all EditMode-testable with no scene objects. The known gap this page inherits: the social server's party lifecycle itself has no EditMode coverage and relies on the server-side bot tests; S31's company work should bring its own bot-client lifecycle suite from day one.

A4 · Requester credits

Tirrag asked for the loot-roll opt-out that shipped in July and anchors this page's consent thread. The champion-spawn revamp thread asked for parties entering together, escalating waves, and a boss with a real payout: that is S33, directly. The high-health world boss thread feeds the same card. Akandriel's request for organizable scenario play is answered by the persistent board (S27) and the muster (S33), deliberately without a queue. Mondravis and Rinaldi, whose reports sharpened the taming and XP-fairness work this page builds on, appear in its mockups. Thank you for writing it all down.

A5 · Selected sources

UO: the party system and Publish 5, 18, and 74 notes on uo.com and UOGuide; champion spawn and New Legacy coverage. EverQuest: the Project 1999 grouping guide and mercenary documentation. City of Heroes: sidekick and Issue 16 documentation. WoW: patch archaeology for 1.11, 3.3, 6.0.2, 8.2.5, and 10.2.5, and the Dungeon Finder debate coverage. FFXIV: Party Finder, Trust, mentor, and commendation documentation. GW2: dynamic events and the Enhanced Squad UI announcement. Also: the DDO social panel, OSRS boards, Monster Hunter SOS documentation, Deep Rock Galactic and Sea of Thieves community histories, BG3 co-op guides, Wizardry and Bard's Tale histories, and on the fiction side, the Council of Elrond, Seven Samurai, and the Critical Role archives. Full URL list in the research notes accompanying this proposal.

Prepared 15 August 2026 against the unity-6.5 migration branch. Companion markdown: docs/plans/2026-08-15-party-grouping-improvements-proposal.md. Thirty-four proposals: eleven quick wins, twelve core upgrades on existing rails, eleven server-backed with their costs in the open.

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.