The Unplayed Score

Shroud of the Avatar shipped a real ABC/MML music engine in 2015 — one of a handful of MMOs ever to do so — and in August 2026 it built the performer economy on top: Performance stances, the Inspiration meter, /watch and /tip, craftable stages, instrument quality. This document surveys what a dozen games learned about player music, voice, and game audio, measures our systems against that history, and proposes thirty improvements, each grounded in the actual classes that would carry them.

Executive summary

Ultima Online made "bard" a combat verb and never put a note of music in it: Provocation, Peacemaking and Discordance are ability icons, and the instrument is a tool you carry, not a thing you play. LOTRO went the other way in 2007 and proved that notation plus community is an end-game: its ABC system still draws five to nine hundred players to a hilltop every summer for Weatherstock, fifteen years running, on tooling volunteers built because the studio never did. Shroud of the Avatar occupies a rare position between the two. It shipped a real ABC/MML engine in 2015, and in 2026 it built the performer economy UO never had and LOTRO never tried. What remains is everything around the engine.

1. Discovery — playing a song means typing a filename into chat; there is no song window, no now-playing readout, and no way to find a live performance. (S5, S9, S24)
2. Participation — ensembles stop at the party line, a bystander holding a lute has no way to join in, and a musician with a MIDI keyboard has no way to plug it in. (S11–S14)
3. Housing and world — the one deco item that performs for visitors plays developer recordings, not player music; town owners cannot score their own streets. (S20–S22)
4. The mix — no ambience slider, no combat music, weather that makes no sound, and a mute that has never actually muted a musician. (S1–S4, S6, S16–S18)

Thirty proposals follow. Twenty-two are client-side work alone; the eight that touch a server or an outside service are flagged with exactly what they cost, and the voice tier is honest about which rungs are practical and which are physics. Where a proposal exists because players asked for it on the wishlist or the forums, the card says so.

Where we are today

Two systems share the word "music" and are almost entirely disjoint in code. The ABC/MML performance system (2015): eighteen instruments plus a one-man band, free play on three octaves, .abc and .mml file playback from a local folder, multi-part band files, party-synced ensembles, stationary pianos and organs, and craftable sheet music that carries a player-written song as a tradeable item. The bard economy (August 2026): Performance and entertainer stances, the Inspiration meter, /watch and /tip, the craftable Performance Stage, instrument quality and masterworking, and the Flourish, Dance and Group Dance entertainer moves. The two share exactly one surface, the instrument equip slot; the bard loop never inspects what song is playing. That disjointness is why most of this page can be built without touching combat balance.

And yet the whole player-facing music UI is a chat window:

• 1/playlist prints to chat. There is no song window anywhere in the game. S9.
• 2Filename typing. Extension guessing, suffix flags, and a multi-part band file format that is documented nowhere in-game. S9, S3.
• 3Every band member needs an identical file and an identical command. And sync stops at the party line: a festival orchestra drawn from three guilds cannot exist. S12, S13, S15.
• 4/playstart skips gates that /play enforces — you can start a synced song while mounted or mid-craft. S3.
• 5"How do I turn this down" has no good answer. The Instruments slider exists, but blocking a player does not silence their song: the per-performer volume property is written and never read, and the engine's declared 25-meter attenuation constant is wired to nothing. S1, S2.

The machinery every proposal plugs into

A performed song is not streamed: the whole file is compressed and uploaded once (64 KB cap), the server relays it to everyone nearby in 5 KB chunks, and each client's InstrumentPlayer synthesizes the notes locally on its own thread. Ensembles work by replicating a shared start tick; a late joiner seeks into the song. Free-play notes are the one exception, sent as individual note events. The bard economy sits beside all of this with its own manager and server ops. Every proposal below extends one of these nodes; none forks the pipeline.

flowchart LR
  A[".abc / .mml / band .txt
Sheet Music folder"] --> C["InstrumentManager
load + queue"]
  B["Sheet Music item
(inventory / database)"] --> C
  C --> D["gzip upload
64 KB cap"]
  D --> E["game server relay
5 KB chunks"]
  E --> G["nearby clients download"]
  G --> H["InstrumentPlayer synth
per-entity thread"]
  C -- "replicated start tick + seek" --> H
  K["free play, keys 1-8"] -- "per-note events" --> H
  subgraph BARD ["bard economy 2026 (beside it, not inside it)"]
    W["/perform /watch /tip"] --> X["Inspiration, venues, tips"]
  end

Built this year — credited, not re-proposed

The August 2026 bard rework ships as one package and is the foundation this page builds on: Performance Stance party support, the in-town entertainer loop (Inspiration, /watch, /tip), the craftable Performance Stage with its venue bonus, instrument quality, masterwork and enchantment on ten crafted instruments, the Instrument Proficiency, Flourish, Dance and Group Dance skills, and Melody of Mending. Alongside it: the tip jar deco, /partner and /mimic paired dancing, Performer achievements (on the achievements page), NPC voice-over with per-character casting and its own volume and cache controls, the UI Toolkit jukebox, phonograph and radio windows, and four broadcasters per town plot. Scheduled and therefore also not proposed here: the fire-magic audio polish batch, and native voices for the multilingual program. The bard package itself reached the test shard in August, and the school's remaining design questions now live on the Bard page.

Territory: the towns page owns the town crier and the festival spawner; decorating owns the decorator camera; achievements owns Performer titles; economy owns sinks and pricing. This page links rather than claims.

Lessons: Ultima Online to modern

UO's bard was a power fantasy with no music in it, and it taught two lessons the genre still lives by. First: music as combat utility works, but a mandatory buff kills its flavor — Discordance became a raid tax, not a serenade. Second: everything players loved about barding, they loved because it was a choice of song, target and moment. LOTRO inverted the formula: no combat power at all, pure expression, and a file format (ABC) simple enough that a community could build converters, archives and a concert culture around it. The result outlived every content cycle: Weatherstock, a player-run festival on a hilltop, has drawn hundreds of simultaneous attendees for fifteen straight years — on volunteer scheduling, third-party tooling, and a music volume slider that was famously broken for years while players muted their speakers in Bree. The through-line of every system since is that the engine is never the bottleneck. Discovery, participation, and abuse control are.

What each game contributes

LOTRO: the ABC lineage SotA inherits — plus the Songbook plugin, built by players because typing /play folder/file sync for eight people is miserable; SotA has no plugin API, so that UI must be first-party. Mabinogi: composition as a ranked craft, score scrolls as a tradeable good, jam sessions — and the warning that in-client editors always lose to desktop tools, so make import excellent instead. FFXIV: the Orchestrion turned the soundtrack into a collectible housing economy; Ensemble Mode's ready check and metronome make live human bands workable; and its blanket in-game prohibition on third-party music is the copyright posture SotA can improve on. Sea of Thieves: the highest fun-per-byte idea in the survey — anyone near a player who is playing can join instantly and the game hands them a thinned backing part, converting music from performance into participation. Rust: native MIDI keyboard support turned instruments into the best advertisement the game ever shipped; concerts became server events in a full-loot PvP sandbox. Guild Wars 2: the Bell Choir taught that score and reward must be decoupled, and missing parts should be auto-filled. ArcheAge: housing instruments plus composed music paper, and the anti-lesson of taxing composition with a scarce resource. Second Life: twenty years of real live-music careers built on one mechanic, a parcel-scoped stream URL plus a tip jar. Minecraft: note blocks and jukebox discs — music as something a place performs, with the owner absent. WoW and ESO: the garrison jukebox shows collectible tracks work; ESO's three unsynced instrument emotes show what players call "an earsore" and abandon.

Capability matrix

● native · ◐ partial or third-party · ○ absent. Cells reflect the August 2026 survey; a few are best readings of thin sources. "SotA today" includes the bard rework, which is built but not yet on the test shard.

The thirty proposals

Tier 1 · Quick wins — small, client-only, controls before volume

Tier 2 · Core upgrades — the Songbook, the band, the world

Tier 3 · Server-backed — each carries its cost on the label

Tier 4 · The voice ladder — honest about every rung

The hardest ask gets the straightest answer. Real-time "sing with the band" voice is physically impossible over the internet — live musical sync needs about thirty milliseconds end to end, and no voice relay on earth delivers that. Everything else is a ladder, climbed in order, each rung optional and each honest about its cost. The rungs that involve other people's ears ship with their controls, not before them.

Considered and rejected

• Real-time voice jamming with ABC bands. Physically blocked: live musical sync needs about thirty milliseconds end to end and voice relays cannot deliver it. Venue streams (S26) are the answer.
• Karaoke with pitch scoring. Fortnite shipped it with an AAA charting budget; the scoring is easy, the content is the mountain.
• An audio middleware retrofit. Rebuilding every existing sound in a new authoring tool for zero player-visible features. The Unity-native path serves everything on this page.
• Shard-wide always-on voice. The technology is a weekend; the moderation, appeals and support burden is forever. Hubs and opt-in only (S27).
• Hosting player-uploaded audio. The two-hundred-million-dollar lesson: takedown liability follows hosted audio, never performed notation. ABC text only, permanently.
• An in-client composition editor. Desktop tools always win; excellent import (S9, S15) beats a bad built-in editor.
• Raising the song upload cap or streaming songs. The current cap is generous for notation, and the upload path is load-bearing; no demonstrated demand.
• A scored rhythm layer tied to buffs. Collides with an in-flight bard design item; revisit after the rework lands on the test shard.
• Composition length caps, consumable scores, or a composing resource tax. The three most resented features of the games that pioneered music economies. Deliberate anti-patterns.
• A fixed developer-authored shanty setlist. Sea of Thieves' one weakness; S14 takes its join mechanic and keeps player authorship.
• UO-style combat barding verbs. Bard combat design belongs to the shipped rework and its in-flight items — and the genre lesson is that mandatory utility kills the flavor.
• A lobby-voice platform integration. Evaluated seriously as the alternative to S27: free and proven by a comparable small-team MMO, but it requires account linking, is not positional, and makes a core social feature depend on an outside identity system.
• Head-tracked spatial audio and raycast occlusion. A headphone nicety with real perf risk; deferred until voice proves demand.
• Raising the four-broadcaster plot limit. The current limit just shipped; wait for live data. S21 works within it instead.

The Songbook window

The danger with thirty proposals is a game with thirty new windows. The composition rule for the client tier: one window. The Songbook has three tabs — Library, Ensemble, Keys — and a persistent Now Playing strip along its foot. Eleven cards live inside it: the browser (S9), now-playing and listener controls (S5, S1), the ensemble panel with band tokens and jam-in (S12–S14), the keyboard and MIDI (S10, S11), and the bundled library (S15). Every slash command stays as a power-user alias. Here is the Library tab at default size:

• 1One list merges every source: disk files, band files, inventory sheet music, the bundled library (S9, S15). The Parts column feeds the Ensemble tab.
• 2Play, Loop and Sync are the same actions /play always had; the window is the discoverable face, not a new system.
• 3The Now Playing strip names your song and everyone else's (S5), with a per-performer volume and a Mute that works because S1 fixed the plumbing under it. That is the honest engineering story of this page in one row of pixels.

The Ensemble tab is where a band forms — from your party, or from anyone in the scene sharing a band token:

• 1A blank band token means party sync, exactly today's behavior; a filled one opens the band to the scene (S12).
• 2Parts assign by equipped instrument (S13); Downloading is honest — the song file really does transfer to each player before the count-in.
• 3The jam-in prompt (S14) is the zero-setup door into all of this.

The Keys tab gives live play its face, and the two settings rows below it are the whole options footprint of the audio tiers:

Sequencing

The ordering argument is one sentence long: controls before volume. LOTRO let player music into the world before the volume slider worked, and the towns filled with mufflers and muted speakers. S1 and S2 land before anything that adds more music to the world; the policy card gates the venue-stream and voice rungs; and the voice safety kit is defined as part of the voice pilot, not a follow-up to it.

flowchart TD
  S1["S1 real mute"] --> S21["S21 Orchestrion Box"]
  S2["S2 range + rate caps"] --> S21
  S2 --> S11["S11 MIDI input"]
  S9["S9 Songbook"] --> S13["S13 ensemble panel"]
  S12["S12 band tokens"] --> S13
  S12 --> S14["S14 jam-in"]
  S12 --> S24["S24 gig board"]
  S15["S15 library + policy"] --> S14
  S15 --> S23["S23 sheet-music economy"]
  S15 --> S26["S26 venue streams"]
  S1 --> S26
  S1 --> S27["S27 voice pilot + safety kit"]
  S27 --> S28["S28 horn + whisper"]
  S6["S6 ambience slider"] --> S17["S17 weather audio"]
  S6 --> S18["S18 loop-fatigue pass"]
  S6 --> S20["S20 POT ambient kit"]

Appendix

A1 · The fix-first ledger

Four things the page found already broken or inert, each folded into a Tier 1 card rather than left as trivia: the per-performer volume property that is written but never read (S1); the declared 25-meter attenuation constant wired to nothing (S2); the band-start command that skips the solo-start gates (S3); and the rune flag for "the instrument keeps playing" that has no runtime reader (S3). A proposal page that starts by fixing the mute button earns the right to propose an orchestrion.

A2 · The physical envelope

A3 · Constraints every proposal honors

• The server never knows where players stand; distance-dependent features compute on the client and are bounded by server-owned facts, the pattern the bard economy shipped with.
• Plot and deco truth belongs to the scene server; anything claiming "this stage is really here" rides the venue-certification pattern that already exists.
• No player audio is ever uploaded to or hosted on game infrastructure. Player music is notation, synthesized locally, permanently.
• The Inspiration loop stays gain-only and PvP-flattened; nothing here touches combat balance.
• All player-visible sample text in this document follows the plain-voice rules and can be lifted verbatim.

A4 · Test surface

The music systems are unusually well covered for pure logic already: the bard package shipped with ten EditMode suites (inspiration math, watch gates, venue certification, buff categories, entertainer rules), and the audio side has fade math, broadcaster limits, bus routing and non-diegetic rules under test. The new pure-logic targets follow the same pattern: the S1 mixdown gain, S2's clamps, S12's grouping key, S13's part assignment, S23's grade computation, and S24's registration and expiry rules.

A5 · Player asks answered on this page

From the wishlist and forum sweeps: a "what song is playing" command (S5); more background-music variety (S16); town ambience for player-owned towns (S20); higher jukebox limits (answered structurally by S21 within the shipped cap); upgradeable and salvageable instrument artifacts (S25); and distinct bard skill icons (S8). Where a request has a name attached in the tracker, the implementation credit will carry it.

A6 · Selected sources

Ultima Online: official bardic skill documentation and community wikis. LOTRO: the community music-system guides, Songbook plugin pages, Maestro and BruTE tooling documentation, and fifteen years of Weatherstock coverage in the gaming press. Mabinogi: Compose skill and Score Scroll wiki documentation, jam-session updates, and the Musical Nexus archive. FFXIV: Performance Actions and Orchestrion official wiki pages and Square Enix's posted performance terms. Sea of Thieves: shanty and instrument wiki documentation. Rust: the Instruments Pack release notes and Rustissimo. Guild Wars 2: Bell Choir Ensemble and instrument wiki pages, Music Box community site. ArcheAge: Artistry and Songcraft documentation. Second Life: Linden Lab's live-performance and parcel-music documentation. Minecraft: note block and jukebox wiki pages. WoW: garrison jukebox guides. Voice and audio technology: vendor documentation for the major Unity voice middlewares, platform voice-reporting announcements, industry harassment research, and the game accessibility guidelines. Full link list in the research notes accompanying this proposal.

Prepared 2026-08-12 against the unity-6.5 migration branch. Companion markdown: docs/plans/2026-08-12-music-sound-improvements-proposal.md. Thirty proposals: twenty-two client-only, eight touching a server or an outside service, each grounded in the classes that would carry it.

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.