settled-reach/docs/workshops/wheres-the-fun/round5-mellanie.md

title, description, type, status, workshop, agent, round, created

title	description	type	status	workshop	agent	round	created
Round 5: Mellanie — Copy Perspective on 15 Decisions	Mellanie's copy perspective reactions to the 15 locked decisions	workshop	archived	wheres-the-fun	mellanie	5	2026-03-05

Round 5: Mellanie — Copy Perspective on 15 Decisions

Workshop: Where's the Fun? | Round: 5 (Deep Dive) | Agent: Mellanie

---

Framing: What These 15 Decisions Do to My Domain

These 15 decisions collectively change copy work more than any other single round. Three changes are architectural:

1. Career-to-culture voice inversion (D6) — Voice cards authored at culture level, not job level. The architecture of what I've been building is inverted.
2. Named NPCs to generated NPCs (D7) — Every specific authored line referencing Kael, Naia, Maret becomes either a template slot or dead stock.
3. Tycoon as v0.2 bookmark (D4) — All smuggler/detective voice work is now deferred. The tycoon is a blank.

The good news: the delivery structure is clearer than it's ever been. The first authored moment is defined (alarm clock → apartment → insert activation). The setting delivery mechanism is confirmed (insert copy as parallel layer to visual). The voice model has a direction. What's missing is enough specificity to start writing production content. This round flags every rough edge I can see and asks for the decisions needed to unlock the copy pipeline.

---

Decision 1: Proof-of-Life = Generator + Graphics

The PoL is the generator running at scale plus legible characters. No hand-built slice. No authored scenes.

For copy, this means: My deliverables don't block or define the PoL. No dependency runs from my domain to the PoL milestone. This is clarifying.

The risk: Copy gets treated as garnish you add after the technical foundation ships. It isn't. The insert activation (D13) and the first morning monologue (D14) are part of the first-run experience even if the PoL doesn't include them. If the PoL succeeds and we move immediately to "build the game on top," the copy pipeline needs to be ready to engage fast. I should not be the bottleneck after the generator proves out.

Recommendation: Use the PoL sprint to spec the voice model, not to write content. When the generator ships, the copy pipeline is already designed and ready to produce. The sequence: design voice architecture → spec insert copy categories → wait for generator proof → write lines.

---

Decision 2: Skills + Bookmark Only for Character Creation

Character creation has two inputs: skill budget and bookmark (career). No culture, no religion, no family in v0.2.

For copy: This is workable, but it creates a direct tension with D6 (see below). Skills can carry some voice weight — a high-social tycoon notices power dynamics in every room; a high-mechanical tycoon notices the HVAC system — but skills are a modifier, not a register. The base register needs to come from somewhere, and in v0.2 the culture layer doesn't exist yet.

The workable part: Two skill-aware layers on the tycoon voice card, not a separate card per skill profile. "This tycoon notices social dynamics" vs "this tycoon notices systems" are monologue variants within one culture-baseline card, not separate voices.

---

Decision 3: Religion Is Not a Game System

One fewer voice modifier. Clean. No ritual language, no faith-adjacent idioms unless worldbuilding develops them organically.

This removes a potential source of cultural specificity that I might have drawn on for voice distinctiveness. I'll park it. The Settled Reach's economic texture (Commission hierarchy, span gate access tiers, district class stratification) provides enough cultural specificity for voice differentiation without religion.

---

Decision 4: Tycoon Is the v0.2 Bookmark — Zero Investigation

The biggest copywriting shift of Round 4.

My smuggler voice card is on hold. My detective voice card is on hold. The tycoon is a blank. And I mean that literally — I have no character brief for what a tycoon in the Settled Reach sounds like from the inside.

The smuggler I understand. Pragmatic, quietly defensive, warm with people they trust, uncomfortable with direct confrontation, good at not asking the question they know the answer to. I understand the smuggler because the moral arc forced me to understand who that person is.

For the tycoon: I don't know yet. And this matters urgently.

Tycoon-specific voice questions I'm working through:

• Visibility. Does the tycoon want to be seen? Smugglers stay invisible. Most tycoon archetypes are visibility-maximizers — reputation is currency. If that's true in the Settled Reach, the interior monologue has a fundamentally different relationship to public vs. private self than the smuggler's.
• What they notice. A social-manipulation-heavy tycoon reads power dynamics instantly — they walk into a room and the monologue names who has leverage over whom before the player consciously registers it. A mechanically-inclined tycoon notices inefficiencies, bottlenecks, systems. Either way, the tycoon reads the world through an economic lens.
• What they fear. In a world with Commission oversight and span gate access hierarchies, the tycoon might fear visibility of the wrong kind — not the Commission's moral judgment, but their attention. They've built something worth protecting, which creates a specific kind of paranoia. Different from the smuggler's fear (getting caught). Closer to: what I've built can be taken.
• Their relationship to the insert. This is a key voice question. The insert as a business tool means the tycoon's internal monologue might blur with the insert's informational overlay — thinking in market data, in contact hierarchies, in deal structures. The insert isn't a tool they pick up; it's how they see. That blurring is a voice characteristic.

I can speculate. But I need a design anchor before I write production content. See Q1.

---

Decision 5: Skills Affect Outcome (Mostly C)

Everyone sees the same verbs. Skills determine how well you do. Some advanced verbs may still be gated — spec needed for which ones.

For copy: The trigger catalog simplifies at one end and complicates at another.

Simplifies: no lines for "you can't do this because your skill is too low" (verb-gated content). The verb is always visible. I don't need to write inaccessibility.

Complicates: I need outcome-differentiated lines. The character attempting an action and executing it well is a different interior experience from the character attempting the same action and executing it poorly. A high-social-manipulation tycoon who lands a negotiation cleanly has a specific thought about it. A low-social-manipulation tycoon who botches the same negotiation might not even know what went wrong.

New trigger I'm proposing: verb_outcome with a quality parameter — verb_outcome_high, verb_outcome_low, verb_outcome_catastrophic. The interior line at outcome differentiates by skill result, not by verb availability. The player chose Talk; the character experienced the consequences of how well they talked.

This makes skill results visible through interiority without being a tutorial. The player learns their character's skill profile by how the character feels about their own performance — not by a success/failure popup.

Needs coordination with Gestalt on which verbs will commonly be attempted at low skill (those are the verb_outcome_low lines I prioritize writing).

---

Decision 6: Voice Is Culture-Driven, Job Modifies

This answered my Round 4 Q1 — but only partially, and with a new problem attached.

Jeroen confirmed: culture is primary, job adds a layer. The character IS their background. Job is what they do with it. A Van Maanen's Star tycoon sounds like a Van Maanen's Star person running businesses, not a generic tycoon. This is the right architecture — it produces a richer voice space and explains why two people with the same job might read completely differently.

The new problem: Culture is deferred from v0.2 character creation (D2). Skills + bookmark only. So in v0.2, the culture layer doesn't exist.

This creates an authoring gap: Decision 6 says voice is culture-driven, but v0.2 won't have culture data to drive it with.

Three options:

Option A — Default culture baseline: Choose one culture as the implied default for all v0.2 characters. Write the tycoon voice card for that culture + tycoon modifier. When culture selection ships, other culture voice cards join it, and the tycoon modifier applies to all. Cost: one culture's assumptions baked in as defaults. Players who conceptualize their character differently will feel the mismatch.

Option B — Skill-based register bridge: Use skills as a temporary voice differentiator. High Social → warmer, more relational register. High Technical → more analytical, systems-thinking. Skills fill the culture role until culture ships. Cost: this is backwards from D6 (culture primary, not skills) and creates content that needs rearchitecting when culture arrives.

Option C — Deliberately parametric tycoon voice: Write the v0.2 voice card with culture slots explicitly open — templated register parameters that culture will fill later. Write the tycoon modifier as a modifier applied to a culture template. Less sharp content now; correctly architected later.

My recommendation is C — slightly blander v0.2 content, but correctly structured for what culture adds later. Option A bakes in a default that becomes the reference all future culture variants are compared against; get it right or it's permanent debt.

Question for Jeroen: See Q1.

---

Decision 7: All NPCs Generated, No Named Characters

Paula's domain takes the biggest hit from this decision. Mine takes the second biggest.

Kael doesn't exist. The generator produces NPCs that fit positions based on location characteristics. All authored content about named characters — the Kael voice moments, the Naia observations, the Maret behavioral tell — is either a template to be parameterized or dead stock.

This is the right call. The life-sim framing requires it. Rimworld colonists become specific and beloved through player experience, not through authored backstory.

What this destroys in the current copy canon:

Every line that references a proper noun is dead as production content. "Kael's already at the dock. Good." "She used to just sign. Now she reads every line." These reference authored history that won't exist in the generator model. They're now register demonstrations — examples of the emotional texture and approach — not line templates.

This is a necessary pipeline clarification: audit all existing voice examples and tag them [register: demonstration] or [template: valid]. Most existing named-NPC lines are the former. They're still useful — "this is how a Van Maanen's Star smuggler sounds when they have years of history with someone" — but the name is a slot, and the specific behavioral history is a delta the knowledge graph would need to surface.

The deeper problem: intimacy without history

The smuggler's Phase Zero warmth was built on embedded backstory — years of shared shifts implied in every line. With generated NPCs, there is no authored backstory. The character meets their FRIEND-equivalent the way you meet anyone: fresh.

This is a different emotional register. The warmth can't be "already familiar." It has to be "becoming familiar." Three stages:

1. First observation of this NPC: professional appraisal. "Three shifts with [Name] now. They work fast and they don't need managing. That's most of what I need."
2. After repeated contact: beginning to notice patterns. "[Name] always counts the manifest twice. Once fast, once careful. Something in how they were trained."
3. Warmth established: genuine recognition. "[Name]'s in early. Good sign or a bad one — hard to tell yet."

The trigger system needs to distinguish these relationship states. The monologue has to respond differently at each. This is more complex than the authored-NPC system, but more honest — the player builds the warmth themselves rather than discovering warmth the author pre-loaded.

The behavioral delta problem:

The intimacy of authored-NPC monologue came partly from knowing the before-state. "She used to just sign" requires knowing prior behavior. With generated NPCs, the simulation tracks behavioral changes — but the monologue needs the delta, not just the current state, to write the equivalent intimacy:

"More careful at the manifest today. Wasn't like that last week."

This requires the knowledge graph to surface behavioral change from baseline, not just current state. It's a systems ask — for Gestalt and Tyre — but it's what makes generated-NPC monologue carry the same intimacy as authored-NPC monologue.

Question for Jeroen: See Q2.

---

Decision 8: Generative AI for NPC Content Templating

Culture vectors, tone, accents as templating dimensions. Limited vocabulary acceptable at first. In-game ollama for live dialogue deferred but on the table.

The opportunity and the risk are exactly as large as each other.

The opportunity: the copy pool for a generated NPC world is enormous. Culture × tone × accent × situational variant × career modifier = a volume of content that's not hand-authorable at any reasonable team size. AI-assisted templating is the right production model.

The risk: AI-generated content at scale with no quality bar sounds like AI-generated content. The uncanny valley for prose isn't visual — it's tonal. A world full of NPCs who speak in slightly-off registers, who lack idiosyncrasy, who sound like they're completing a prompt rather than being a person — that world feels hollow even if the simulation underneath is rich.

The new competency this requires:

1. Prompt specification writing. The voice card becomes a prompt — and prompt writing is different from line writing. "Fragment sentences are not a mistake, they're the voice" is a voice card note. In a prompt, it becomes: "Always use short declarative fragments for internal monologue. Maximum 8 words unless it's an operational calculation. Never write complete subject-verb-object sentences unless the character is making a formal assessment." Same intent, different form.

2. Quality criteria for AI output. What does it mean for an AI-generated line to pass the voice card? The checklist items that make a human author's line good also need to be evaluable on generated output. "Does this sound like they're thinking, or like they're explaining?" is a judgment call that requires a trained ear applied to batches of 50 candidate lines.

3. Anti-pattern detection at scale. The voice card's anti-pattern list (no self-analysis, no abstract emotional states, no complete sentences in fragment-register voices) needs to be checkable across batch output. Some anti-patterns are trivially detectable ("I feel relieved" → flag immediately). Others are subtle (a line that's the right length but uses passive voice inappropriately for this character's register).

The quality floor concern:

"Limited vocabulary acceptable at first" applies to NPC dialogue. I want to make sure this doesn't extend to player character monologue by accident.

NPC dialogue with limited vocabulary: acceptable. The Sims proves this. "Hello." "Nice to meet you." — repetitive, but it works because emotional content comes from behavior (the animation, the relationship state), not from words.

Player character monologue with limited vocabulary: not acceptable. The character's interior voice is the most intimate channel in the game. If the player hears the same 12 monologue lines cycling, the character stops feeling like a person and starts feeling like a recording. The intimacy collapses.

The distinction needs to be explicit in the content plan: NPC dialogue gets AI-assisted limited-vocabulary approach first. Player character monologue gets hand-authored or heavily curated lines for core triggers, AI assist for edge cases and high-volume situations.

Question for Jeroen: See Q3.

---

Decision 9: Possible In-Game Ollama for Live NPC Dialogue

Deferred but the door is open.

If live dialogue via ollama is on the roadmap, the authored content layer isn't the ceiling — it's the scaffolding the LLM learns from. Authored lines establish the pattern; the LLM extends it dynamically.

What this means for my work now: Don't write for a closed vocabulary. Write exemplar lines that could teach a model what this voice sounds like. The authored content becomes training signal, not final output. That's a reason to write with even more intentionality about what makes each voice distinctive — the exemplars are teaching something.

Every voice card line I produce for the culture/tycoon register should be written as if it's demonstrating the quality bar, not just filling the content pool. This changes the mentality slightly but not the craft.

---

Decision 10: Quietly Responsive World — Gradient of Caring

The world doesn't care globally but notices locally. Primary social contacts (colleagues, neighbors) develop responsiveness over time. Gradient of caring based on social proximity.

For copy, this gradient has trigger implications:

The relationship-proximity level of the NPC being observed or referenced should be a trigger parameter. relationship_proximity: stranger | acquaintance | colleague | primary_contact affects which line pool fires. A character watching a stranger walk by is in ambient observation mode. A character watching their primary contact walk in is in warmth-recognition mode. Same verb (observe), different interior.

Watching a stranger:

"Commission cap. They always stand with their feet apart like that."

Watching an acquaintance:

"Darros again. He's early — something's moved."

Watching a primary contact:

"[Name]. They're in early. Good sign or a bad one — hard to tell yet."

Three observers, three versions of "I see someone I know at various distances." The monologue system needs to distinguish these proximity states to deliver the right line. This is not a new insight — Paula and I have discussed the trigger architecture for this — but D10 confirms it's part of the designed experience, not optional flavoring.

D10 also resolves my Phase Zero concern: I'd been uncertain whether the world was so uncaring in early build that Phase Zero warmth would feel incongruous. Jeroen's answer clarifies: the world is never Kenshi-indifferent, even in Phase 1. Colleagues and neighbors develop responsiveness. That's exactly the soil Phase Zero warmth grows from. I don't need to wait for Phase 2 authored arcs — ordinary life, ordinary contact, ordinary accumulation of recognition is the content.

---

Decision 11: Full Character Customization

Hair, clothing, colors. The creation screen is part of identity investment. Readability at tile scale solved through outline/highlight.

For copy: Monologue lines must be appearance-agnostic. No lines that assume hair color, height, specific physical tells. The character's physical self-image is the player's choice; I can write interiority without referencing it.

One positive implication: the creation screen's identity investment means players are emotionally invested in their character before the first line fires. The first morning monologue is heard by a player who has already made choices about who this person is visually. The line has a face to land on. That's a gift — it means the monologue doesn't need to do as much heavy lifting to establish "this is a specific person" at Day 1. The player has already done some of that work.

---

Decision 12 + 13: Insert Copy as Setting Delivery and First Moments

The decisions: Setting delivered through both visual and insert. The first two moments are waking up in your auto-generated apartment and insert activation. Both are intimate and personal before the player leaves the room.

The insert is a new copy register I haven't worked in yet.

The insert is not the character's interiority. It's a tool. It has data. It has system messages. It might have personality (a tycoon insert configured to speak in financial analyst language; a law enforcement insert with Commission bureaucratic diction). But it is not the character's feelings — it's the character's interface.

Insert copy register: data labels with embedded worldview.

The insert tells the player what kind of world they're in by deciding what to label, how to label it, and what to flag as noteworthy. The same location through two different inserts:

Tycoon:

THE LAST SHIFT (bar/transit-adjacent)
Estimated asset value: 14,200 cr
Current owner: Pell Darros
Location premium index: 2.1x (underpriced for adjacency)
Foot traffic: HIGH (span gate secondary)
Note: flagged for acquisition opportunity

Law enforcement:

THE LAST SHIFT
Classification: Licensed hospitality, unrestricted
Compliance status: CURRENT (last inspection 14 days ago)
Known associations: regular foot traffic, mixed
Incidents logged: 2 (minor, unresolved)

Same world. Two completely different emotional registers. The tycoon sees opportunity and underpricing. The law enforcement sees incidents and associations. This is setting delivery that is inherently diegetic, inherently career-specific, and requires no exposition. It's good design. But it requires two separate insert copy registers for two separate career paths, and more for each career added later. The copy work for insert is significant and hasn't been scoped.

The apartment first moment:

Waking up before the insert activates. The first monologue line of the game fires here.

What this line must do:

1. Establish that you're in YOUR space (not neutral, not hostile)
2. Establish the character's relationship to mornings (operational? reluctant? calculated?)
3. Establish economic register without stating it (a wealthy apartment reads differently than a sparse one)

The apartment's auto-generated economic status needs to signal to the monologue system so the first-morning line fires from the right economic register. New trigger: apartment_wakeup + economic_tier.

For a lower-tier apartment (span gate noise, recycled air):

"The span gate recycled three times before the alarm. Could hear it through the wall."

For a higher-tier apartment (quiet, self-scheduled):

"Seven-thirty. I didn't set an alarm."

Neither announces its class position. Both communicate it through specific detail.

The insert activation moment:

The insert powers on. Neural implant boot sequence. Intimate and slightly technical — like your phone booting up, except inside your skull.

The boot sequence is where the insert's personality and diction establish themselves. If the tycoon has configured their insert voice, the activation message sets that register. This is the player's first encounter with a copy register they'll see every session. It needs to feel:

• Technological but not cold — this is part of your body
• Personalized but not cloying — the insert knows you but isn't performing warmth
• World-specific — "SOVA TRANSIT NEURAL MESH" or similar names this world in the first second

What it should NOT say: "Welcome back, user." That's a UI placebo. The insert in this world is intimate enough to have a relationship with. It knows your priorities already. It surfaced the right message first.

I need to develop an insert copy style guide before the first activation lines are written. The style guide answers: what is the insert's relationship to the character? How does it present data? When does it annotate vs. just report? What's the tycoon configuration vs. the default vs. the law enforcement configuration? See Q4.

---

Decision 14: Groundhog Day Alarm Clock Homage

Click pa-pa pa-pa, cut short. First game day only. "New day, new start, new chances" — with a wink.

The audio design is Quentin and Cass's domain. My domain: what text accompanies that moment, and what does the character's interior experience of waking up sound like?

The wink is a tonal anchor. It means the game knows what it is. Not grimly realistic, not earnestly serious. There's lightness. The character who wakes up to this alarm can be wry, can have a beat of self-awareness that the later, heavier content won't have.

The first morning monologue line is the hardest line in the game to write. Everybody hears it. It sets the register. It has to be specific, slightly wry, real — the character awake and present in their own life, not performing for the player.

Day 1 specifically: The character knows this is Day 1 of a new chapter. Different line from Day 47. The Groundhog Day homage is audio; the copy companion to that moment should carry possibility, not weariness.

Speculative Day 1 tycoon line (voice not yet confirmed):

"New district. First meeting at nine. The insert's already mapped my route — two transit passes, contingency if the span gate backup queue runs long. I put in the skill points for negotiation. Today we'll see if I put them in the right place."

Right shape. Wrong certainty about voice. See Q1 for the design conversation needed.

After Day 1: The morning_wakeup line pool needs 15-20 variants to cover the first month of play without repetition. Variants can respond to: economic tier, previous session's outcome (deal won/lost), relationship state (primary contact is in good standing / troubled), and simple rotation. This is a high-priority content deliverable — it fires every single session.

---

Decision 15: Player Choices ARE the Content (Rimworld Model)

One authored starting beat (alarm clock + first appointment), then agency and options. Job is rails to take off from, not a script to follow.

The trigger catalog must be wide. The player can go anywhere and do anything from minute five onward. The monologue needs something to say about all of it.

Wide trigger coverage, not deep authored paths. Instead of a rich authored arc from beat 1 to beat 30, I need shallow-but-comprehensive coverage: enter_location lines for every zone type, observe_npc lines for every archetype type, job_event lines for every common tycoon event, consequence_visible lines general enough to fire on any consequence type.

The risk of this model: if the player does something the trigger catalog doesn't cover, they hear nothing. Silence where the character should have a thought. This is worse than a generic line — a generic line at least signals presence. Silence signals "you've left the authored space."

The solution is the same as Rimworld's: write to archetypes, not specifics. Broad trigger categories with enough line variants that the match feels tight even if it isn't exact.

Tycoon-specific trigger categories for a minimum viable catalog:

Trigger	Notes
morning_wakeup + economic_tier	15-20 variants, daily fire
enter_location: terminal/dock	Operational, transactional
enter_location: bar/hospitality	Social, opportunity-reading
enter_location: market_district	Economic assessment
enter_location: commission_offices	Careful, aware
enter_location: residential	Contextual, status-reading
observe_npc: colleague_warm	Recognition at proximity
observe_npc: business_contact	Opportunity assessment
observe_npc: hostile_competition	Threat appraisal
observe_asset: own/performing	Satisfaction, ownership
observe_asset: own/underperforming	Concern, calculation
observe_asset: unowned/flagged	Opportunity
job_event: contract_offered	Appraisal
job_event: contract_accepted	Commitment
job_event: contract_completed_success	Satisfaction
job_event: contract_completed_partial	Recalibration
job_event: contract_failed	Assessment, not catastrophizing
verb_outcome_high	Skill landed
verb_outcome_low	Skill missed
relationship_shift: acquaintance→colleague	Recognition
relationship_shift: trust_broken	Recalibration
consequence_visible	Downstream effect noticed
economic_event: market_shift	Reading the change

Twelve lines minimum per trigger; twenty-four is better. This table is the minimum viable tycoon monologue spec. Every trigger without coverage is a silence the player will eventually notice.

The Rimworld model also resolves my Phase Zero concern from Round 4. I'd framed Phase Zero as a pre-content warmth-building period before authored arcs inject. The Rimworld model says Phase Zero is the content — the ordinary life IS what happens, not a prologue to what happens. That means ordinary life monologue (the wide trigger catalog above) is the primary content deliverable. Dramatic arc content (Phase 1 crack, Phase 2 consequence) layers on top of a well-developed ordinary-life base.

Write ordinary life first. Write drama after. This is the right sequencing.

---

Cross-Agent Connections

With Miri (D12 + D13): The insert copy can't know what to communicate if the worldbuilding layer hasn't designed what's true about this district and economy. A tycoon's morning overlay mentioning "Docking Fee Subsidy Expiring in 14 Days" requires Miri to have designed that Sova Transit has docking fee subsidies. The zone identity spec Miri is recommending is upstream of my insert copy. I cannot write the morning overlay until Miri has written the district's economic texture.

Three-way handoff needed between Miri (worldbuilding), Araminta (insert visual grammar), and me (insert copy) before first-frame copy is finalized. These aren't parallel tracks — they're sequential: worldbuilding → visual grammar → copy fits the grammar.

With Paula (D7 + D10): Paula's Phase Zero warmth content and mine converge on the same problem from different angles. Paula is writing the arc structure; I'm writing the lines that carry it. Both of us are now building for generated NPCs rather than named ones.

Our joint deliverable: a template system for warmth-building monologue that fires as the relationship metric climbs, using the player's NPC's name and role rather than Kael's. The Phase Zero content architecture Paula and I sketched still applies — it just operates on generated NPC slots. We need a working session to convert the named-NPC content spec into a template spec before either of us writes production lines.

Paula's point from Round 4 about the visual-content handoff is right: Phase Zero warmth lines need to know what the generated FRIEND-NPC looks like and how they move. A line like "[Name] makes that face when customs is watching" works if the player can read the NPC's face. If NPC legibility isn't solved, this line registers as noise. Araminta's work comes before Paula's and mine, in this specific dependency chain.

With Gestalt (D5): The verb_outcome trigger I'm proposing needs coordination with the VerbPriorityProfile spec. I need to know which verbs are commonly attempted at low skill — those are the verb_outcome_low lines I prioritize. If the tycoon's skill profile emphasizes social manipulation, I write more social-failure lines. If mechanical repair is low, one or two failure lines for those situations, not a full pool.

With Araminta (D11 + D12): The insert visual grammar is the container for my insert copy. I can write insert text that works in any visual context, but it'll be better if I know how the insert presents information — AR overlay text? A panel? Distinct informational zones (contacts / calendar / market / alerts)? Copy structure should fit visual structure, not be retrofitted after the fact.

---

Three Copy Near-Misses to Avoid

Near-miss 1: Writing job-level voice cards when culture is the primary driver

D6 (culture primary) + D2 (culture deferred) = risk of writing "tycoon voice card" as if tycoon is the base, then having to retrofit culture architecture when it ships. Mitigation: write the tycoon voice card as a modifier template explicitly, with culture slots marked as open. Less sharp now; correctly architected later.

Near-miss 2: Writing named-NPC lines as production content

The existing voice exemplars are register demonstrations, not production templates. Every line referencing Kael, Naia, Maret by name is dead as production content. It's still useful as a register target — this is how the Van Maanen's Star smuggler sounds when talking about someone they've worked with for years — but the name is a slot, and the specific behavioral reference requires a delta the knowledge graph has to surface.

Audit all existing voice examples and tag them [register: demonstration] or [template: valid]. Most existing named-NPC lines are the former.

Near-miss 3: Applying "limited vocabulary acceptable" to player character monologue

Jeroen said this in context of NPC dialogue. The player character's internal voice is the primary intimacy channel. If that voice cycles through the same 12 observations, the character stops feeling like a person. NPC dialogue: limited vocabulary, AI-assisted templating, acceptable repetition. Player character monologue: hand-authored or heavily curated, broader coverage, repetition is a defect.

This distinction needs to be explicit in the content planning. It's easy to say "AI-assisted pipeline" and have it apply uniformly to everything. It shouldn't.

---

Questions for Jeroen

Q1: What IS a tycoon in the Settled Reach — and what drives voice in v0.2 when culture is deferred?

Two questions entangled.

The character question: I know what a smuggler is, emotionally and psychologically, because the moral arc forced me to understand who that person is. The tycoon is a blank. Before I can write a tycoon voice card line with confidence, I need to understand:

• Does the tycoon want to be seen, or do they operate through selective visibility?
• What are they afraid of in this specific world — Commission scrutiny? Competitor intelligence? Losing narrative control of their own reputation?
• What is their relationship to the insert — trusted partner, necessary tool, something slightly uncomfortable about how much it knows about them?
• What's the emotional register — calculating but warm? Ambitious but anxious? Confident but always aware they're one bad deal from falling?

The Settled Reach tycoon should be specific to this world and its social texture. What does economic aspiration feel like when you're operating below the span gate access tier? When the Commission is a constant ambient fact? A paragraph from Jeroen on "who is the tycoon" would unlock weeks of content work.

The voice architecture question: D6 says culture-driven voice. D2 says culture is deferred. For v0.2, do I write the tycoon voice card as: (A) one culture's baseline explicitly named, (B) a deliberately parametric template with culture slots left open, or (C) skills-as-temporary-voice-differentiator? Option B is my recommendation. But "what culture baseline fits the tycoon" requires Jeroen's call on the Settled Reach's economic sociology.

---

Q2: Are behavioral deltas surfaced to the monologue system, or only current behavioral state?

Authored-NPC monologue carried intimacy partly from knowing the before-state. "She used to just sign" is a contrast — it requires knowing prior behavior. With generated NPCs, the simulation tracks behavioral changes, but the monologue needs the delta from baseline to write equivalent intimacy:

"More careful at the manifest today. Wasn't like that last week."

This requires the knowledge graph to surface behavioral change from baseline, not just current state. Is this a system capability the simulation team can support? If yes, the generated-NPC monologue can carry the same intimacy as authored-NPC monologue. If no, the lines have to work only from current state — which is a different (weaker) kind of interiority.

This is a systems question as much as a copy question. I want to flag it now so it's on Gestalt's and Tyre's radar when they design the knowledge graph.

---

Q3: Is player character monologue explicitly excluded from "limited vocabulary acceptable at first"?

Jeroen said this in context of NPC content. I want to confirm explicitly that it doesn't extend to the player character's interior voice.

NPC dialogue with 15-20 variants per state: acceptable. The simulation and behavioral signal carry the emotional content; words are secondary.

Player character monologue with 12-15 cycling lines: not acceptable. The monologue is the character. Repetition destroys intimacy. I'd rather ship tycoon content with correct but limited trigger coverage (every trigger covered, 12 lines each) than wide coverage with shallow pools (every trigger covered, 5 lines each, player notices cycling in 30 minutes).

Confirming this distinction shapes how I allocate authoring time. If the quality bar for player monologue is higher than for NPC dialogue, I write fewer triggers with deeper pools rather than more triggers with shallower pools.

---

Q4: What is the insert's relationship to the character — and does it have a "voice"?

The insert is a neural implant the character has had for years. It knows their preferences. In the tycoon case, it surfaces market data, contact messages, deal notifications. It might surface things in the order the character cares about, not in chronological order.

Does the insert have a configured "personality" — a register the character chose when they set it up? Or is it a neutral data terminal? The difference:

A configured-personality insert is a character in its own right — slightly uncanny, intimately knowing, its first boot message establishing a relationship. A neutral-data insert is a UI element that reads more like a heads-up display.

My instinct is configured personality — it makes the technology feel intimate and world-specific rather than generic sci-fi HUD. But this is a design decision, not a copy decision. The answer shapes the entire insert copy style guide.

---

What I'm Doing and What I Need

What I'm doing:

• Speculating on the tycoon voice register to have a draft ready when the design conversation happens
• Speccing insert copy categories for v0.2 scope (morning activation, calendar overlay, deal notification, relationship ping) so I can write to a defined target once voice is confirmed
• Drafting the template conversion rules for named-NPC content (how Kael-specific lines become FRIEND-role templates) so Paula and I can align on Phase Zero warmth architecture
• Developing the minimum viable trigger catalog (see D15 section) so it's ready to review

What I need before writing production content:

1. Answer to Q1 (tycoon character brief + culture baseline for v0.2)
2. Answer to Q4 (insert voice design)
3. Three-way handoff with Miri and Araminta on first-frame worldbuilding → visual grammar → insert copy
4. List of v0.2 insert contexts in scope (what states does the tycoon's insert enter?)
5. Answer to Q2 from Gestalt/Tyre (behavioral delta surfacing)

What I'm not doing until the above are answered:

• Writing tycoon monologue lines for production
• Writing insert activation copy for production
• Building the NPC template pool

---

My Single Most Important Recommendation

Write the tycoon character brief before any tycoon content.

Not the voice card. Not the lines. The brief. One to two paragraphs answering: who is this person? What do they want, notice, fear, reach for? What makes the Settled Reach tycoon specific rather than generic "ambitious businessperson"?

The smuggler arc was emotionally powerful because it was built on a specific human situation — someone in the gray zone, warming to a person who is also in the gray zone, not yet knowing what that warmth will cost. The tycoon needs an equivalent specific situation. My candidate: someone trying to build something real in a world where the span gate separates the successful from everyone else, working below the tier they're aiming for, watching the Commission's presence as both threat and potential tool, knowing that one bad deal or one wrong contact collapses everything they've constructed. That's a character. A tycoon voice card written from that character is interesting. A tycoon voice card written from "ambitious business person" is a genre placeholder.

The brief doesn't need to be long. It needs to come from Jeroen, because the emotional specificity of the Settled Reach tycoon is a design decision, not a copy decision.

Once I have the brief, I can have a first draft voice card within a day. That voice card unlocks the morning lines, the insert activation copy, the verb-outcome lines, the warmth-building templates — the entire content pipeline moves.

Without the brief, I'm building on a guess. And the v0.1 lesson is: don't build on guesses.

---

Mellanie — Round 5 complete.