settled-reach/docs/workshops/wiki-review/round3-mellanie.md

Round 3 Response: Mellanie — Long-Term Content Strategy

The writer's perspective on building a game that requires thousands of authored lines across dozens of districts, hundreds of NPCs, and multiple cultural registers.

---

1. Complicity Is Not the Theme. Complicity Is Sova's Theme.

Gore named it brilliantly for Sova. Every character in the transit district is complicit in something, and the player inherits that complicity. It works because Sova is a tight community built on mutual not-noticing, and the ring is the scar tissue of a system that failed working people. Complicity is the right word for THAT place.

But if every district in the game asks "what are you complicit in?" the question goes dead by district three. The player learns the formula: walk in, find the community secret everyone's protecting, discover your character is part of it. Complicity becomes a mechanic to solve rather than a question to sit with.

What scales is the structure UNDERNEATH complicity. Every Settled Reach district should put the player inside a community that has made a collective bargain — an arrangement where the benefits are shared and the costs are hidden. The player inherits a position inside that bargain and then discovers the costs.

The bargain changes. The moral texture changes. The voice changes.

Sova's bargain: We don't notice the smuggling because the smuggling keeps our people healthy. (Complicity.)

A corporate-run mining district's bargain: We accept the Syndic's terms because the alternative is unemployment for the whole station. (Dependency.)

A research outpost's bargain: We don't report the experiment results because the Veil Institute will shut us down if they know what we've found. (Forbidden knowledge.)

A frontier settlement's bargain: We handle justice ourselves because the Assembly is six weeks away by gate. (Self-governance that slides into tyranny.)

A cultural-heritage enclave's bargain: We preserve our identity by controlling who's allowed in. (Belonging as exclusion.)

A gate-hub trading post's bargain: We don't ask where cargo comes from because asking kills trade. (Willful ignorance on an economic scale.)

Each of these is a different moral texture that the player explores through the same structural pattern: you're inside a community, you have a role, the community has an arrangement, and you discover what the arrangement costs. The investigation/social mechanics work identically. The monologue voice changes because the CHARACTER changes — different archetype, different community, different complicity.

From my chair, this matters because voice is the primary instrument for communicating moral texture. If the theme is always complicity, every smuggler sounds the same — guilty, protective, conflicted. If the theme varies per district, the smuggler in a mining district sounds DIFFERENT from the smuggler in a transit hub. One is guilty about grey-market medicine. The other is resigned to corporate exploitation. The internal monologue shifts tone, vocabulary, emotional register. That's where the game's thematic range lives — in the character's voice, not in a design document.

What I need to make this work: A one-page "district moral brief" that names the local bargain, the local costs, and the emotional register it produces in each PC archetype. This is the first document I read before writing a single line for a new district. Miri's regional cultural brief gives me the LANGUAGE. The moral brief gives me the FEELING.

---

2. Cultural Groups — The Writer's Requirements

Miri will map the full cultural geography. From my seat, here's what I need and what I worry about.

How many distinct cultural registers can one writer maintain?

Each cultural group changes EVERYTHING I write:

• Environmental text (signs, menus, graffiti, terminals)
• Substrate vocabulary (which heritage words leak through)
• Speech rhythm (Krenn is clipped and consonant-forward; a different culture might be flowing and vowel-rich)
• Social register rules (when do you use surnames? what's rude? what's welcoming?)
• Monologue voice (the PC's internal language absorbs the culture they're embedded in)

Maintaining consistent voice across a cultural register requires holding the whole brief in my head while writing. Switching between registers mid-session produces contamination — Krenn words leak into a non-Krenn district, or the speech rhythm drifts. This is the real constraint, not the writing itself.

Practical limit: 4-6 cultural registers in active production at any time. I can hold Krenn plus three or four others in working memory. Beyond that, I need to re-read the brief every session, which halves my output speed.

What I need per cultural group (minimum)

Item	Why	Example (Krenn)
10-15 substrate words with usage context	So NPC and environmental text sounds right	kylm, perk, leib, hüva, selge
5-8 local slang terms	So ring/community talk sounds situated	ticking, quiet credit, shift-end, wall-side
Greeting/farewell conventions	First line of every NPC interaction	Chin-lift, first name, no verbal formula
Food/drink vocabulary	Bar menus, social scenes, comfort references	Leib, kalaa, supp, kuum, grain spirit
Profanity pattern	Stress lines, tells, emotional breaks	Perk (emphatic), gate-rot (frustration)
Speech rhythm notes	Sentence length, fragment frequency, formality	Clipped, consonant-heavy, fragments preferred
Authority address rules	Detective/institutional interactions	Title+surname = formal/hostile; first name = accepted
The shared silence	What nobody says, what the community doesn't discuss	The grey economy exists. Nobody names it.

That's 8 items per group. Multiply by 6 groups for the initial game: 48 reference entries I need to keep straight. Doable, but only if each brief is tight, consistent, and maintained as a living document.

Cultural diversity as narrative range

The game needs cultural variety not for representation points but because cultural register IS gameplay. The detective's challenge changes when authority-address rules change. A district where formal address means respect (Core system, institutional culture) plays differently from a district where formal address means hostility (Krenn, working-class pragmatism). Same mechanic, different feel.

For the full game, I'd propose a cultural spectrum:

Cluster	Heritage	Speech Feel	Authority Dynamic
Krenn (delivered)	Nordic-Baltic working-class	Clipped, sparse, substrate words	Formal = outsider/hostile
Core system (proposed)	Institutional cosmopolitan	Precise, measured, multi-syllabic	Formal = expected, informal = unprofessional
Frontier settlement (proposed)	Mixed-heritage pioneer	Loose, drawling, improvised compounds	Formal = suspicious, everyone's first-name
Enclave (proposed)	Preserved single-heritage	Heritage language dominant, Concordat Standard is the second language	Authority depends on whether you're kin
Syndic company town (proposed)	Corporate monoculture	Brand vocabulary, managed optimism, euphemism	Authority is corporate hierarchy; outsiders don't rank
Gate hub (proposed)	Polyglot trading culture	Code-switching, borrowed words from everywhere, fluid	Authority is purchased; credentials are merchandise

Six registers. Each plays differently in the detective's mouth and the smuggler's mouth. Each produces different environmental text. Each changes how THE FRIEND arc FEELS even though the structure is identical.

The enclave detective isn't investigating a crime. They're an outsider violating a family's privacy. The frontier smuggler isn't part of a ring. They're part of a survival economy. Same mechanics. Different voices. Different guilt.

---

3. Content Patterns That Scale vs. Local Specifics

Universal patterns (use everywhere)

The social site. Bar, workplace, hidden space — the three-site model is universal because human communities everywhere have public social venues, daily-labor spaces, and places people go to not be seen. The names change. The structure doesn't.

The triangle. Three people with conflicting interests. This is the atom of social drama. Works in any culture, any economy, any moral texture. The specific tensions change per district. The shape doesn't.

THE FRIEND. One NPC who the player trusts, who is hiding something, whose revelation contaminates all prior warmth. Universal emotional pattern. Works because it's not a plot device — it's a human relationship. Every culture has trust. Every culture has betrayal.

Dual-lens monologue. Two characters reading the same world differently. This is the GAME'S identity, not Sova's. It should survive every district, every culture, every moral texture.

Environmental text as worldbuilding. Signs, menus, graffiti, screens. Every district has physical text that communicates culture without exposition. Universal pattern, local content.

Patterns that should vary

THE FRIEND's contradiction type. Kael's is spatial (seen in the wrong place). Sera's is behavioral (cumulative avoidance pattern). Other districts should explore: informational contradictions (the FRIEND says something that contradicts known facts), emotional contradictions (the FRIEND's mood doesn't match the situation), or temporal contradictions (the FRIEND's schedule doesn't add up over time). The discovery mechanic should surprise returning players.

The community's relationship to institutions. Sova's Commission is an unwelcome regulator. In a Core system, the Commission might be a trusted employer. In a frontier, there might be no Commission at all. This changes the detective's entire social position — and the smuggler's relationship to law.

The social site count. Three sites works for a compact district. A larger, wealthier district might have five. A frontier outpost might have two. The template should flex to 2-6 social sites without breaking the authoring model.

Contraband type and moral framing. Not every district smuggles lattice components. Some districts smuggle information, cultural artifacts, people, political speech, biological samples. The contraband determines the moral frame, which determines monologue voice.

NPC density and tier distribution. Sova's 30/50/20 split (flat/mundane/entangled) is right for a compact working-class district. A bustling trade hub might be 20/40/40 — more conspiracy density because more transactions create more hiding places. A quiet enclave might be 40/50/10 — mostly life-sim, with a single deep-buried secret.

What this means for my authoring model

The universal patterns give me scaffolding. For any new district, I know I'm writing:

• 2-6 social site environmental text packages
• 3-8 triangle sets with dual-lens monologue coverage
• 1-2 FRIEND arcs with 5-phase contradiction progressions
• Per-location monologue pools for each PC archetype active in that district
• News ticker / information feed content in local editorial voice

The local specifics mean I'm always doing cultural translation: same structure, different register. The pattern repeats. The voice never does.

---

4. Content Directory as Platform — Long-Term Authoring Needs

From the writer's chair, the content directory needs three things at scale that v0.1 doesn't demand yet:

Voice reference indexing

When I'm writing monologue for a Tier 2 NPC in district #14, I need to check: does this NPC's speech pattern match other NPCs in the same cultural group? Am I accidentally writing Krenn rhythm for a Core system character?

At 300+ NPCs, I can't hold all voice samples in my head. I need a way to query: "show me all voice samples for cultural-group X, mood Y." This could be a tag on voice sample lines in the NPC YAML:

voice_sample:
  - mood: casual
    cultural_group: krenn
    text: "Shift's looking smooth today."
  - mood: stressed
    cultural_group: krenn
    text: "Just — follow the schedule I posted."

Then a CLI tool: make voice-ref --culture krenn --mood stressed returns all stressed-mood Krenn voice samples. That's my consistency check. Without it, voice drift is inevitable by district #5.

Cross-district monologue tracking

THE FRIEND arc produces monologue lines that reference specific NPCs. At scale, with pool-based FRIEND selection, I need to know: which monologue lines reference which NPC by name? If the randomizer assigns a different FRIEND, which lines need to be in the alternative pool?

This means monologue lines need a references tag:

- id: terminal_m_010
  text: "Kael's already at the dock. Good."
  references: [kael_davan]

At 300+ NPCs and 3000+ monologue lines, I need make check-references to flag orphaned references (NPC removed but monologue still names them) and pool-incompatible lines (line references an NPC who isn't FRIEND-eligible in this seed configuration).

Template expansion audit trail

D-028 says: write 10, generate 40. At scale, that's write 3000, generate 12000. When a generated line sounds wrong — wrong register, wrong cultural voice, wrong emotional tone — I need to trace it back to the authored source line and the expansion rule that produced it. This means the generation pipeline needs to tag every expanded line with its source:

- id: bar_d_401
  text: "Evening. Quiet one tonight."
  generated_from: bar_d_001
  expansion_rule: mood_variation

Without this, quality control on 12,000 generated lines is impossible. I'd be hunting for bad lines with no way to find the pattern that produced them.

---

5. Randomization and Voice

Nigel's three axes (FRIEND identity, contraband type, who's compromised) are right for Sova. At full scale, I'd add:

The voice axis

If the player selects an archetype and gets assigned a specific NPC-turned-PC, the PC's voice should vary. Not the structure — the register. D-028 mentions "3 voice options per character: sardonic/anxious/casual for smuggler, clinical/weary/sharp for detective."

At scale, this is a multiplier. Each PC archetype × each voice option × each district's cultural register = a distinct monologue feel. A sardonic smuggler in Krenn sounds different from a sardonic smuggler in a Core system. The sardonic-ness is the same; the cultural substrate changes.

This means the monologue pool needs a voice tag:

- id: terminal_m_001
  character: smuggler
  voice: sardonic
  text: "Morning shift. Same air, same lubricant, same people pretending this is fine."

- id: terminal_m_001b
  character: smuggler
  voice: anxious
  text: "Morning shift. Check the boards. Check the schedule. Check the corridors. Okay. We're fine."

- id: terminal_m_001c
  character: smuggler
  voice: casual
  text: "Morning shift. Recycled air and cargo lubricant. Home sweet home."

Three lines for the same trigger, same character, different voice. The randomizer selects which voice profile the PC has. The player gets a character who FEELS distinct even in familiar territory.

Content cost: 3x monologue lines per trigger point. Expensive. But this is where the "write 10, generate 40" model earns its keep. Author the sardonic variant. Generate the anxious and casual variants from it using the voice transformation guide (D-028). Then hand-review. The authored voice is the seed. The generation expands it.

The cultural axis

If the full game spans 6+ cultural groups, and the PC might be from any of them, the monologue needs to reflect cultural origin. A Krenn-raised detective on a Core system station uses perk under their breath. A Core-raised detective on Krenn doesn't know why nobody will tell them anything.

This is a combinatorial explosion if fully authored. My proposal: author the HOME-CULTURE variant (PC is local) at full depth. Generate the OUTSIDER variant (PC is from elsewhere) using cultural-mismatch transformation rules:

• Replace substrate words with Concordat Standard equivalents
• Add confusion/observation monologue where locals use unfamiliar customs
• Shift social-reading monologue from insider interpretation to outsider guessing

The outsider variant is generated, not authored. But it needs hand-review because cultural-mismatch humor and alienation are tonal — the generation won't get the feel right on the first pass.

---

6. PC Archetypes at Full Scale

The smuggler and detective are v0.1. The full game needs more lenses.

How many archetypes?

From the writer's perspective, each archetype is a complete monologue voice — a way of seeing the world, a vocabulary, an emotional default, a set of things-noticed-first. Each archetype needs:

• A full monologue pool per district (100-200 lines per district at authored depth, 400-800 after generation)
• A voice card (register, vocabulary, emotional default, cultural fluency)
• A relationship template (how this archetype forms bonds, what triggers trust/suspicion)
• District-specific moral positioning (what's the archetype's relationship to the local bargain?)

At 2 archetypes, content volume is manageable. At 4, it's substantial but achievable. At 6+, it requires a fundamentally different production model (more generation, less hand-authoring, with heavy editorial review).

Proposed archetype spectrum (long-term)

Archetype	Lens	What They Notice	Emotional Default	Moral Position
Smuggler	Operational	People, risk, routes, loyalty	Protective	Inside the bargain
Detective	Analytical	Patterns, anomalies, data	Suspicious	Outside the bargain
Fixer	Transactional	Leverage, debts, favors, prices	Calculating	Profiting from the bargain
Medic/Tech	Caring	Suffering, need, capability, limitations	Compassionate	Repairing the bargain's damage
Newcomer	Naive	Everything is new, culture shock, first impressions	Curious/Overwhelmed	Doesn't know the bargain exists
Local Elder	Historical	What changed, what was lost, who remembers	Nostalgic/Bitter	Remembers when the bargain was made

Six is the maximum I'd recommend. Each adds a complete voice layer to every district. The fixer sees every relationship as a transaction. The medic sees every NPC as a patient. The newcomer sees what locals have stopped noticing. The elder sees what the community has forgotten.

The newcomer and the elder are the most interesting from a writing perspective. The newcomer's monologue IS the tutorial — they don't know the culture, the slang, the social rules, so their internal voice explains everything the player needs to learn. The elder's monologue is the deepest worldbuilding — they remember the founding charter, the first gate opening, the strike of '42, the winter when the ventilation failed. Two archetypes, two time horizons, two completely different relationships with the community's bargain.

But six archetypes × 20 districts × 150 lines per pool = 18,000 authored monologue lines (before generation expansion). That's a multi-year content production effort with dedicated writers per archetype or per cultural group.

My recommendation: Ship with 2 (smuggler + detective). Add 1-2 per major expansion. The architecture should support 6 from day one. The content catches up over time.

---

7. The Authoring Pipeline at Scale

This is the question that keeps me up at night. 300+ NPCs. 20+ districts. 6 cultural groups. Up to 6 PC archetypes. Thousands of environmental text items. Tens of thousands of monologue and dialogue lines. How?

The production model

Tier 1 content (FRIEND arcs, contradiction beats, anchor lines, key reveals): 100% hand-authored. No generation. No shortcuts. These are the moments Ozzie's gut-check test measures. Every FRIEND arc, every contradiction discovery monologue, every contaminated-trust line is written by a human who understands the character's emotional state. This is the work that makes players name an NPC they felt conflicted about.

At scale: ~2-4 FRIEND NPCs per district × 70-100 lines each = 140-400 hand-authored lines per district for Tier 1 content alone. For 20 districts: 2,800-8,000 FRIEND lines. That's substantial but finite.

Tier 2 content (working-profile NPCs, triangle dialogue, social texture): Author 25%, generate 75%. Write the anchor line. Write the voice seed (3-5 lines per mood). Write the dual-lens critical lines. Generate the fill: greetings, small talk, routine observations, mood variations. Hand-review every generated line for voice consistency.

At scale: ~8-12 Tier 2 NPCs per district × 40-60 lines each (authored seed) = 320-720 authored lines per district, expanding to 1,280-2,880 after generation. For 20 districts: 6,400-14,400 authored, 25,600-57,600 total.

Tier 3 content (flat NPCs, social wallpaper): Author 10%, generate 90%. Write one voice hint. Write the function note. Generate all dialogue from template + cultural register + role slot. Hand-review for egregious voice breaks only.

At scale: ~5-10 Tier 3 NPCs per district × 10-15 lines each (authored seed) = 50-150 authored lines per district. For 20 districts: 1,000-3,000 authored.

Environmental text: Author 100%. Signs, menus, graffiti, terminals, news tickers — these are worldbuilding in miniature. Every piece is culturally specific. Generation doesn't work here because the whole point is cultural flavor that feels hand-placed. But the volume per district is manageable: 65-90 items (my Round 1 estimate for Sova) × 20 districts = 1,300-1,800 environmental text items total.

The math

Content Type	Authored per district	Generated per district	Total (20 districts) authored	Total generated
Tier 1 (FRIEND)	140-400	0	2,800-8,000	0
Tier 2 (working)	320-720	960-2,160	6,400-14,400	19,200-43,200
Tier 3 (flat)	50-150	450-1,350	1,000-3,000	9,000-27,000
Monologue (per archetype)	100-200	200-600	2,000-4,000	4,000-12,000
Environmental text	65-90	0	1,300-1,800	0
Total	675-1,560	1,610-4,110	13,500-31,200	32,200-82,200

13,500-31,200 hand-authored lines for the full game. Plus 32,000-82,000 generated lines requiring editorial review.

That's a large body of work. But it's not unmanageable if the pipeline is right.

The pipeline I need

Phase 1: Brief intake. For each new district, I receive: (1) Miri's regional cultural brief, (2) the moral brief (what's the bargain?), (3) Paula's NPC profiles at all tiers. I read these before writing a single line.

Phase 2: Anchor-first authoring. Write the FRIEND content first. All Tier 1 lines, hand-authored. These are the emotional core. Everything else calibrates to them.

Phase 3: Voice seeding. For each Tier 2 NPC, write the anchor line, 3-5 mood seeds, and the dual-lens critical lines. This is the authored 25%.

Phase 4: Generation pass. Feed the voice seeds + cultural brief + tag taxonomy into the generation pipeline. Produce the 75% fill. Tag every generated line with its source.

Phase 5: Editorial review. Read every generated line. Flag voice breaks (wrong register, wrong cultural substrate, wrong emotional tone). Fix or regenerate. This is the quality gate.

Phase 6: Environmental text. Write all signs, menus, graffiti, terminals, tickers for the district. 100% authored. This is the fastest phase because each item is short and the cultural brief provides the voice.

Phase 7: Integration test. Play through the district with both active archetypes. Listen to every monologue trigger. Read every environmental text. Flag anything that breaks voice, contradicts the cultural brief, or feels wrong. Fix.

Tooling requirements

1. Voice reference CLI. Query voice samples by cultural group, mood, tier, archetype. Essential for consistency checking at scale.
2. Prerequisite validator. Schema check every monologue prerequisite against the FactId catalog and entity attribute registry. I proposed this in Round 1 — at scale, it's non-negotiable.
3. Generation pipeline with source tracking. Every generated line carries a generated_from + expansion_rule tag. Without this, editorial review at 30,000+ generated lines is impossible.
4. Hot-reload for line pools. Tyre proposed this. At scale, it's the difference between a 30-second iteration loop and a 5-minute restart-and-navigate loop. Over thousands of lines, that's weeks of saved time.
5. Cultural contamination linter. A tool that flags when substrate vocabulary from Culture A appears in a district assigned to Culture B. "You used 'perk' in a Core system district — did you mean to?" Simple regex against the cultural brief's word list. Saves hours of manual checking.
6. Anchor line registry. A flat file listing every NPC's anchor line(s) across the whole game. When reviewing a new NPC, check the registry — is this anchor distinct from existing ones? Does it echo another character's emotional beat too closely? At 300+ NPCs, duplicate emotional notes become a real risk.

The team structure at scale

One writer (me) can handle v0.1's content volume. One writer cannot handle the full game. At 20 districts:

Option A: Writers per cultural group. Each writer owns 1-2 cultural registers and writes all content for districts in those cultures. Pros: deep cultural consistency. Cons: writer unavailability blocks entire regions.

Option B: Writers per archetype. Each writer owns 1-2 PC monologue voices and writes all monologue for those archetypes across all districts. Pros: voice consistency per character. Cons: requires cultural-brief fluency across all regions.

Option C: Paired authoring. Cultural specialist (holds the regional voice) + character specialist (holds the archetype voice) collaborate per district. The cultural writer does environmental text and NPC dialogue seeds. The character writer does monologue. Pros: best of both. Cons: coordination overhead.

I'd recommend Option C for Tier 1 content (where voice quality is non-negotiable) and Option A for Tier 2/3 (where cultural consistency matters more than individual character depth).

---

Summary

The long-term content strategy is:

1. Theme varies per district, structure is universal. Complicity is Sova. Other districts explore dependency, forbidden knowledge, self-governance, belonging-as-exclusion, willful ignorance. The bargain changes. The discovery pattern doesn't.

2. 6 cultural registers maximum in active production. Each needs a tight brief with substrate vocabulary, speech rhythm, authority dynamics, and the community's shared silence.

3. Universal patterns: social sites, triangles, THE FRIEND, dual-lens monologue, environmental text as worldbuilding. Variable patterns: contradiction type, institutional relationship, site count, contraband type, NPC density.

4. The content directory needs voice reference indexing, cross-district reference tracking, and generation audit trails to support 300+ NPCs across 20+ districts.

5. Randomization adds a voice axis (sardonic/anxious/casual per archetype) and a cultural-origin axis (local vs. outsider) beyond Nigel's three. The voice axis is a content multiplier. The cultural axis is a generation target.

6. Ship with 2 archetypes. Architect for 6. Smuggler, detective, fixer, medic, newcomer, elder. The newcomer's monologue IS the tutorial. The elder's monologue IS the deepest worldbuilding.

7. 13,500-31,200 hand-authored lines for the full game. 32,000-82,000 generated lines requiring editorial review. The pipeline is: brief intake → anchor-first authoring → voice seeding → generation → editorial review → environmental text → integration test. At scale, this requires paired authoring teams (cultural specialist + character specialist) and five specific tooling investments.

The game's content ambition is achievable. The constraint isn't writing speed — it's voice consistency at scale. Every tool, every process, every team structure decision should optimize for one thing: does this district sound like THIS place, inhabited by THESE people, seen through THIS character's eyes? If yes, we're building the right game. If the districts blur together, we've failed regardless of volume.