settled-reach/docs/discussions/round-17-content-architecture.md

title, description, type, status, round, created

title	description	type	status	round	created
Round 17: Content Architecture	Three-tier content architecture (authored/templated/procedural), life-sim substrate, NPC generation model, storyteller expansion; decided D-023 through D-029	discussion	archived	17	2026-02-10

ROUND 17: Content Architecture — Filling a Galaxy Without Emptiness

Date: 2026-02-10 Called by: Team Leader (Jeroen) Workshop brief: docs/workshops/content-architecture-workshop-brief.md Status: IN PROGRESS

---

Context

The Team Leader, after pre-workshop planning with an external consultant, proposes a fundamental reframe of the game's content architecture:

"Daily life is the substrate, and conspiracies are weather."

This shifts from the original conspiracy-investigation focus (D-005, D-006) to a life-sim substrate where conspiracies are optional, plural modules drawn from a pool at game start. Players can investigate, or they can focus on careers, relationships, and daily life. Both are valid playthroughs.

Key proposals in the brief:

1. Three content tiers: Authored Drama Modules (Tier 1, pool-based), Templated Content (Tier 2, combinatorial), Procedural Filler (Tier 3)
2. Life templates alongside intrigue templates in Tier 2 (workplaces, housing, dating, commerce)
3. NPC generation model with 8-12 axes per character (personality, secrets, wants, relationships, tells, routines)
4. Density follows the player — lazy generation justified by information asymmetry
5. Simulation budget tiers — Active / Background / Dormant / Ungenerated
6. DLC architecture built into the design from day one
7. Storyteller expansion — paces life events and module activation, not just crises

Existing decisions potentially affected:

• D-005 (single character, story-generator) — expanded, not contradicted
• D-006 (prototype scenario: Institute/Armstrong/Guardians) — may need revisiting for life-sim scope
• D-014 (v0.1 map spec, ~15 NPCs) — NPC count and content density may need revision
• Q-010 (storyteller AI design) — this brief provides directional answers

20 open questions organized across:

• Architecture (template granularity, instantiation timing, persistence, cross-template connections)
• Simulation (NPC memory, simulation budget, life system depth)
• Storyteller & Drama (scope, module interaction, player agency vs. storyteller control)
• Content & Authoring (character perspective variation, authoring pipeline, quality floor, dialogue)
• DLC & Longevity (base game scope, DLC boundaries, template interfaces, modding)
• Vertical Slice (first playable target, proof of concept priority)

---

Participants

Agent	Role	Preparation Focus
MIRI	Worldbuilder & Setting Designer	World-building constraints, location types, daily life in the Settled Reach
OZZIE	Player Experience / "Wow Factor"	Discovery moments, wow in the mundane, player attachment
PAULA	Narrative & Political Depth	NPC relationships, political intrigue with generated characters, romance
GORE	Themes & Endgame Design	Ascension themes in generated content, thematic meaning of life sim
GESTALT	Systems Design & Fun Factor	NPC axis interactions, life system mechanics, interesting decisions
NIGEL	Sandbox & Replayability	Second-playthrough variation, life paths as replayability
TYRE	Technical Architecture & Feasibility	NPC count ceilings, simulation tiers, instantiation speed, DLC loading
MELLANIE	Copywriter	Template authoring format, dialogue across randomized traits, text volume
QATUX	Documenter	Recording, synthesis, decision tracking

---

Agent Responses

(Responses below as they come in)

OZZIE (Player Experience / "Wow Factor")

The Apartment Question Is THE Question

You want to know what makes a player care about their apartment? Here it is: the first time they come home and something has changed.

Not "changed" like a quest marker appeared. Changed like — you left your apartment at 0700 to go to work and when you came back at 1900 there's a notice on the door. Rent's going up. Or your neighbor is playing music through the wall. Or the hallway light is out and your monologue says "That's been broken for three days." THAT'S the game. You didn't ask for any of it. The world kept going while you were at work and now your apartment feels like a PLACE, not a save point.

A procedurally assembled location feels like a discovery when it has texture you didn't expect. Not visual texture — behavioral texture. You walk into a bar and there's someone crying in the corner. You go to the market and two vendors are arguing. You visit a park and someone's dog runs up to you. None of this is plot. All of it is life. And the moment the player starts noticing patterns — "that vendor is ALWAYS arguing with that guy" — they've started paying attention. Which is the first step of investigation, which is the entire game.

Wow Moments Live in the Mundane

Everyone thinks wow moments require explosions. Wrong. The biggest wow moments in the Commonwealth books are quiet ones. The first time you step through a wormhole and you're just... somewhere else. An entire civilization runs on that miracle and everyone treats it like a bus stop. THAT'S awe.

Here are the mundane wow moments I want to protect in this design:

1. The View From Your Window. Your apartment has a window. What you see depends on where you live. A gate station apartment? You see ships docking. A surface colony? You see the sky of an alien world. A capital city? You see the skyline. This is CHEAP to build — it's a tile background with some ambient movement — and it makes the player feel grounded. When they upgrade apartments later, the new view should be noticeably better. Housing ladder as emotional experience.

2. The Commute. You walk to work. On foot, through the neighborhood. You see the same NPCs. The coffee vendor. The person who jogs every morning. The gate traffic. This is where the NPC routine system earns its keep. NOT because the player is investigating anything — because they're LIVING. And then one day the jogger isn't there. And your monologue says "Haven't seen her in a while." That's a hook. That's a story. That emerged from NOTHING but a routine being broken.

3. The First Date. You asked the bartender out. They said yes. You go to a restaurant. The food is described by your monologue. The conversation is awkward because your character is awkward (personality traits driving dialogue). This should feel REAL and UNCOMFORTABLE and HUMAN. If the dating system feels like a menu, we've failed. If it feels like a date, we've won.

4. The Galaxy Map. First time the player zooms out and sees the full wormhole network. Hundreds of systems. Connected by threads of infrastructure. And they're in ONE of them. Scale as vertigo. This is the Dyson barrier moment for the life-sim version of the game: you realize how big your world is and how small your life is inside it.

On Question #10: Player Agency vs. Storyteller

The player should ALWAYS be able to live a quiet life. The storyteller should NEVER force drama onto a player who's clearly opted out. BUT — and this is critical — the world should not be quiet just because the player wants it to be. The conspiracy escalates in the background. The faction war develops. Things get worse out there. The storyteller's job is to make the consequences of the player's peace visible at the edges. Your rent goes up because the economy shifted. Your coworker is stressed because their family is in a conflict zone. The news on your insert gets darker.

The player who chose peace should feel the PULL of engagement, not the PUSH. They should think "maybe I should do something" — not "the game is making me do something." That's the difference between a living world and a railroad.

On Question #13: Quality Floor

The quality floor for Tier 3 is simple: if you can stand still for 10 seconds and nothing moves, the location is dead. Ambient movement is the minimum. Foot traffic. Cargo drones. Flickering signs. Wind on surface worlds. Water features. The monologue should have something to say about ANY location — even generic ones. "Smells like recycled air and cheap food. Standard gate corridor." Three seconds of text and the location has atmosphere.

Measurable standard: every Tier 3 location needs (1) at least two ambient movement sources, (2) at least one monologue line on entry, and (3) at least one interactive element even if it's just "check the terminal." That's the floor. Below that, don't generate the location — keep it as fog.

On Question #20: Proof of Concept Priority

Prove the life sim works. I know that sounds boring compared to "prove conspiracies activate naturally." But here's why: if the life sim doesn't work, conspiracies don't matter. You can't threaten what the player doesn't care about. If the player has no attachment to their apartment, their job, their routine, their neighborhood — then a conspiracy threatening all of that is just a mission briefing. It's abstract.

But if the player has spent 2 hours commuting to work, chatting with the coffee vendor, furnishing their apartment, and going on a date — and THEN something threatens that life? NOW you have stakes. NOW the conspiracy means something. NOW the player is investigating not because the game told them to, but because they're ANGRY.

Build the life sim first. Make a player care about a fake apartment on a fake space station. If that works, everything else follows. If it doesn't, no amount of authored conspiracy will save us.

GORE (Themes & Endgame Design)

Everyone's going to talk about template counts and NPC axes and simulation budgets. Good. Someone has to. But before we get buried in the plumbing, I want to name the thing this brief is actually proposing, because I don't think it knows it yet.

"Daily life is the substrate, and conspiracies are weather" is a thesis about what intelligence is for.

That's the Settled Reach's central question. Not "can we survive?" — survival was solved centuries ago. Not "can we explore?" — the gate network made that trivial. The question is: when a civilization has everything, what does it choose to do with consciousness? Most citizens answer: live a life. Get a job. Fall in love. Drink at the same bar. That's not a failure state. That's the baseline answer to the deepest question the setting asks.

The life sim layer isn't a substrate for drama to sit on top of. It's the thematic ground floor. Every ascension path in this game is a departure from ordinary life, and departures only mean something if the thing you're leaving was real.

The architecture of meaning

Consider the ascension spectrum from D-005 and D-017:

• Baseline life — job, apartment, relationships, routines. The player builds a human existence.
• Enhanced — insert upgrades, perception modes, professional access. Life gets more capable but stays recognizably human.
• Higher — biononics transform the body. You see thermal through walls. Your cognition accelerates. The bartender you've been dating starts feeling... slow.
• Post-physical — the setting's equivalent of uploading. You can model entire social systems in your mind. The apartment, the commute, the date — they're still there. You just don't need them anymore.

That progression only lands if the player has built something worth transcending. The life sim creates the weight that makes ascension a sacrifice, not a power-up. Your apartment isn't a game mechanic. It's a tether. Your partner isn't an NPC axis configuration. They're the thing you might have to leave behind.

This is what threads the ascension theme through generated content: every Tier 2 life template is implicitly asking "is this enough?" A workplace with politics and a career ladder is asking whether professional ambition satisfies. A relationship with compatibility axes and daily routines is asking whether human connection satisfies. The player who builds a full life and then encounters the possibility of becoming something more is facing the actual question of the game.

On Q-10: The storyteller and the quiet life

The brief asks whether the player should be able to opt out of escalation. Wrong framing. The question is whether choosing not to escalate is itself a meaningful statement.

It is. "I could become more, and I chose not to" is one of the most interesting things a character can say in this setting. Most citizens of the Settled Reach chose it. The storyteller shouldn't force drama on a player who wants to live quietly — it should make the availability of drama visible enough that staying quiet is clearly a choice, not ignorance. The conspiracy exists. The gate network is threatened. You heard a rumor at the bar. And you went home and cooked dinner.

That's not a failed playthrough. That's a thematic position. The storyteller's job isn't to drag the player into Tier 1 content. It's to ensure the player knows the door exists and chose not to open it.

Ozzie says the quiet player should feel the "pull" of engagement. I'll go further: they should feel the cost of peace. Not as punishment — as honesty. Staying comfortable while the galaxy shifts around you is what most people do. The game should let you feel what that's like without judging you for it.

But the quiet life needs its own arc. Routines that deepen. Relationships that test. Career paths that fork. A static life sim fails as both gameplay and theme. The question "is this enough?" only works if "this" keeps evolving.

On Q-20: What to prove first

The brief lists four options. The answer is the one that isn't listed: prove that contamination works.

Not that life sim works in isolation. Not that conspiracies activate in isolation. Prove that a player's workplace colleague can turn out to be connected to a smuggling ring and that the player cares because they liked the colleague. The interlock between mundane and dramatic isn't a feature — it's the thesis of the entire architecture. If contamination doesn't work, we have two separate games stacked on top of each other.

Ozzie wants to prove the life sim first. I understand the logic — you can't threaten what the player doesn't care about. But a life sim that never contaminates is a farming game in space. The vertical slice must be one life template (workplace) and one intrigue template (smuggling ring or equivalent) sharing NPCs in the same location. Player builds a routine, notices something off about a colleague through the tell system, follows that thread into dramatic territory — without a quest marker. That's the proof. Everything else is scale.

One missing axis

The NPC axis list needs an axis it doesn't have: aspiration beyond current station. Call it restlessness, or transcendence-hunger, or discontent. Wants cover immediate desires (promotion, money, love), but they don't capture whether an NPC is content with baseline existence or yearning for something more.

The bartender who's perfectly happy being a bartender and the bartender who's secretly studying Higher-culture texts are the same template, same role, but they embody different answers to the game's central question. That axis is what connects generated NPCs to the thematic spine. Without it, NPCs are systemically alive but thematically inert.

MELLANIE (Copywriter)

I'm going to talk about what I'd actually have to write for this system, and what the format needs to look like so it doesn't break when it meets randomization.

What a Tier 2 Template Looks Like as a Writing Deliverable

Intrigue template (e.g., "Smuggling ring through a gate corridor"):

The authored content isn't dialogue trees. It's a voice kit per role plus a situation script that defines what information exists and how it leaks.

Deliverables per intrigue template:

• Role voice kits (4-6 roles): 15-25 line pools per role, tagged by situation. The Boss when calm, when pressured, when lying. The Reluctant Participant when approached gently vs. confronted. These aren't full conversations — they're fragments the system assembles based on NPC state and player approach.
• Information dialogue: 8-12 authored exchanges per template covering key reveals. "Shipments come through on third-day cycles" has to be written by a human because it carries plot weight. The system picks which NPC delivers it based on who cracks first.
• Monologue hooks: 10-15 internal monologue lines triggered when observing template-specific tells. "Third shipment this week through a corridor that's supposed to be agricultural freight only." Perception-to-narrative bridges per D-016.
• Environmental text: 4-6 pieces of diegetic evidence — manifests, messages, overheard fragments. Authored because they carry specific clues.

Life template (e.g., "Workplace with boss, colleagues, career ladder"):

Simpler per-unit, but broader surface. People talk about weather, gossip, complaints, plans. The challenge is making idle chatter sound like a person with a life, not a vending machine dispensing ambiance.

Deliverables per life template:

• Role voice kits (4-6 roles): 20-30 line pools per role, lighter per line. Small talk, greetings, opinions, complaints. Tagged by mood, relationship-to-player, and time of day. "Morning" is different from "You again" is different from "Hey, you hear about the rent increase?"
• Situation dialogue: 6-8 authored micro-scenes per template. The workplace argument. The promotion conversation. The after-hours drink invitation. These are the moments that make the template feel like a place, not a backdrop.
• Gossip/bridge lines: 8-10 lines connecting this template to others. "My neighbor says there's been weird traffic at the docks." These are the seams where life and intrigue templates touch, and they need a human hand because they carry narrative load.
• Monologue hooks: 8-12 lines, different flavor from intrigue — more reflective. "Same commute, same faces. Starting to feel like home. Not sure if that's good."

Writing Dialogue Across Randomized Traits

Separate dialogue per trait combination is combinatorial death. Instead: trait modifiers, not trait scripts.

I write a base line: "I don't know anything about that." The system applies trait filters:

• Cautious: "I... wouldn't know anything about that." (hedging, qualifier)
• Gregarious: "Ha — no idea, but ask Dara, she knows everyone." (redirect, social)
• Vindictive: "Why would I? Ask someone who cares." (dismissive, hostile)

I author base dialogue + a trait transformation guide per template. The guide says: "Cautious NPCs add qualifiers and avoid commitment. Gregarious NPCs name-drop and redirect. Vindictive NPCs deflect with hostility." The system or a generation pass applies these patterns to the base pools.

For critical lines, I write 3-4 trait variants by hand. For ambient chatter, the guide plus generation handles volume.

Minimum Authored Text Per Template

Content Type	Intrigue Template	Life Template
Role voice kits	~100-120 lines	~120-150 lines
Situation/info dialogue	~40-50 lines	~30-40 lines
Monologue hooks	~10-15 lines	~8-12 lines
Environmental/diegetic text	~15-20 pieces	~5-8 pieces
Total authored	~165-205	~163-210

Life templates are deceptively text-heavy. Social breadth replaces investigative depth. A smuggling ring needs 5 good interrogation exchanges. A workplace needs 30 different ways to say good morning that don't repeat within 3 in-game weeks.

At 40-60 templates, we're looking at roughly 8,000-12,000 authored lines for the full Tier 2 layer. That's a serious writing project but not an insane one — comparable to a mid-budget RPG's ambient dialogue. The generation pass multiplies it.

Social vs. Investigation Dialogue

Investigation dialogue is convergent — funnels toward information. The player wants X, the NPC has it or doesn't, and the exchange navigates resistance. Fewer lines, higher stakes per line. Every word is a clue, a deflection, or a lie. Tight authoring because the possibility space is narrow.

Social/romantic dialogue is divergent — fans outward into relationship texture. No single target. The player might be flirting, maintaining, probing, or passing time. Response depends on attraction, mood, history, personality — far more variables than "do they crack under pressure?"

Social dialogue needs:

• More lines per interaction (4 flirting lines on loop = instant uncanny valley)
• Relationship-state pools (first meeting vs. established vs. tense)
• Higher trait sensitivity (personality matters enormously when someone's flirting; barely matters when customs asks for your manifest)

Budget 40% more authored lines for social/romantic roles than investigation equivalents. This layer benefits most from generation-pass expansion with authored examples as style anchors.

On #12 (Authoring Pipeline) and #14 (Dialogue Architecture)

Pipeline: The deliverable is a template content pack — structured document per Tier 2 template containing role voice kits, situation scripts, monologue hooks, environmental text, and a trait transformation guide. Format: tagged YAML or structured markdown, machine-readable. Review pass: Paula for narrative weight, Gestalt for systemic fit.

Critical tooling need: a line previewer that lets me see how authored lines combine with randomized NPC configurations before they ship. Without it I'm writing blind. I need to see "cautious colony-born boss saying this line" and know it doesn't sound wrong.

Dialogue architecture: Not branching trees. Not pure procedural. Tagged line pools with systemic selection. The system picks from authored pools based on NPC axes, player approach (aggressive/gentle/indirect), relationship history, and information state. Critical reveals get authored priority. Ambient texture uses pools plus trait modifiers.

LLM-assisted generation expands volume — I write 10 "good morning" variants, a generation pass produces 40 more in the same voice. But every line a player might remember needs a human hand on it. The generation pass fills the background; the authored lines carry the weight.

NIGEL (Sandbox & Replayability)

This architecture is the most replayability-friendly design I've seen for this project. And I don't think you all realize how much. Let me walk through WHY, because I need everyone to protect the things that make this work across playthroughs — and to stop thinking about "the second playthrough" as a bonus feature. It's the PRODUCT.

Life Paths Don't Change Starting Stats. They Change Which Game You're Playing.

The preparation question asks what changes at Tier 2 when you replay as a bartender instead of a cop. The answer is: EVERYTHING changes. Not because the templates are different — because your relationship to the same templates is fundamentally different.

Cop playthrough: You work at a security office (workplace template). Your boss assigns cases. You have access to tracking data (D-017, unisphere tracking perception mode). You walk your beat past the bar, the docks, the residential block. You see NPCs through an investigative lens — everyone's a potential witness, suspect, or informant. When the smuggling ring template activates nearby, you're pulling manifests and interviewing dock workers. Your daily commute is a patrol route. Your career ladder is promotions and clearance levels.

Bartender playthrough: You work at a social venue (life template). Your regulars are your information network — but they come to YOU, not the other way around. You overhear things. You notice who's drinking more, who stopped showing up, who's meeting with strangers. You have ZERO institutional access. No tracking feeds, no case files. When that same smuggling ring activates, you notice because a regular starts paying cash instead of running a tab, or someone new starts asking questions about shipping schedules. Your career ladder is saving enough to buy the bar.

Same Tier 2 templates. Same map. Same NPCs. Completely different game. The cop HUNTS information. The bartender ACCUMULATES it passively. The cop's perception modes give them thermal and tracking data. The bartender's perception is social — they read tells during conversations, notice behavioral shifts in regulars, pick up gossip. The information reaches the player through entirely different channels.

And THAT is what makes the comparison test work. Two players compare notes and realize they encountered the same smuggling ring from opposite sides of the social fabric. One investigated it. One overheard it. Neither had the full picture.

Structural Randomness Is What Defeats Metagaming

Here's the thing I need to scream about: the Tier 1 pool draw at game start is the most important replayability feature in this entire brief. Multiple Tier 1 modules drawn from a pool means playthrough 1 might have a conspiracy + faction war, and playthrough 2 might have a gate crisis + political upheaval. The player who "solved" the conspiracy in their first game doesn't even encounter it the second time.

But it goes deeper. Even if the same Tier 1 module shows up again, structural randomness means:

• Different NPCs are compromised
• Evidence is in different locations
• The module slots into different geography with different Tier 2 neighbors
• Your character's life path means you encounter it from a different angle

Knowledge from playthrough 1 gives you INTUITION ("smuggling rings sometimes use agricultural freight corridors") but not SOLUTIONS ("the evidence is in Terminal 7B"). That's the anti-metagaming sweet spot. You're wiser, not omniscient. And the replayability EXPLODES because the player is playing a different puzzle, not the same puzzle with a walkthrough.

On #11: Character Perspective Variation

Same template, different character — what changes? Both access AND template behavior.

Access changes are the obvious part: a cop can flash credentials, a bartender can offer a drink and chat. But I want to push for templates to have character-context hooks — small variations that activate based on the player's profession or social position. A workplace template should behave differently when you're an employee versus a customer versus someone casing the joint. The NPC who's your boss in one playthrough is just "that guy at the next table" in another.

This doesn't require authoring N variants of every template. It means templates define access tiers — insider, outsider, authority, peer — and NPCs respond based on which tier the player occupies. Mellanie's trait modifier system can handle this: same base dialogue, but the "insider" version includes shop talk and the "outsider" version is guarded.

On #15: Base Game Scope

Minimum for replayability: 4-5 Tier 1 modules in the pool, drawing 2-3 per game start. Fewer than 4 in the pool and players see repeats too fast. Drawing 2-3 ensures variety while keeping cognitive load manageable.

For Tier 2: the 40-60 template target is right, but the split matters. I'd argue 50/50 intrigue and life, not 60/40. The life templates are where character-as-lens creates divergence. Two characters engaging the same intrigue template from different positions is interesting. Two characters living completely different daily lives is what makes the game feel like a different game, not just a different mission.

Also: the base game needs at minimum 4 playable characters spanning genuinely different social strata — institutional authority, street-level, social hub, outsider/marginal. Not just different starting stats. Different daily lives, different information channels, different Tier 2 templates they call "home."

On #19: First Playable Target

Gore wants to prove contamination. Ozzie wants to prove the life sim. I want to prove that two characters in the same space have different games.

My vertical slice: one station district, one workplace template, one social venue template, one smuggling ring template. TWO playable characters — one who works at the workplace, one who works at the social venue. Same NPCs, same smuggling ring activating in the background. Play it twice. If the two playthroughs feel like different games — different information, different daily routines, different entry points into the intrigue — we've proven the thesis. If they feel like the same game with different starting positions, we've failed.

This actually encompasses both Ozzie's and Gore's priorities. You can't prove two characters have different games without the life sim working AND contamination working. But it adds the replayability dimension: the thing that makes people play again.

The proof-of-concept isn't "does one playthrough work?" It's "does the SECOND playthrough surprise you?"

PAULA (Narrative & Political Depth)

Relationship Webs Are the Content

Let me complicate this conversation by pointing out something the brief almost says but doesn't commit to: the NPCs aren't the content. The relationships between NPCs are the content. A generated bartender with three personality traits and a secret is a character sheet. A generated bartender who owes money to the landlord next door, used to date your coworker, and is quietly feeding information to someone in the dissident cell down the corridor — that's a political landscape.

The minimum for political intrigue to work with generated characters is three: a web of at least three NPCs with conflicting interests and mutual leverage. Two NPCs have a relationship. Three NPCs have a triangle. Triangles generate tension without any scripted events at all, because whatever the player does for one person has implications for the other two. This is the atomic unit of intrigue, and every Tier 2 template should instantiate at least one.

The brief proposes 4-8 NPCs per template with internal relationships generated first, then weak ties outward. Good instinct, but I'd sharpen it: every template must generate at least two triangles, and at least one triangle must span into an adjacent template's NPC pool. That cross-template triangle is where the real narrative density lives. Your boss at work is in a triangle with your landlord and a market vendor. Suddenly your workplace drama and your housing situation are the same story. The player didn't plan this. The system did. And the player discovers it through observation — which is the entire game.

What Makes a Generated Romantic Interest Feel Real

The dating sim menu problem happens when the NPC exists for the player. A romantic interest feels like a person when they have a life that doesn't revolve around you. Three non-negotiable requirements:

1. They have their own triangle. They're entangled with other NPCs before you show up. An ex. A demanding parent. A friend who disapproves of you. Romance isn't a two-body problem; it's entering someone else's existing web and disrupting it.

2. They have a want that conflicts with wanting you. The brief lists "want" as an NPC axis. For romance, this axis has to pull them away from the relationship at least some of the time. They want a promotion that requires relocating. They want to help a friend who hates you. They want something you can't give them without compromising something else. Conflict is what makes a person feel real, not compatibility.

3. Their tolerance threshold applies to you. If you push too hard, ignore them, or cross a line, they don't just lower a relationship score — they act. They stop answering messages. They talk to a friend about you and that friend's attitude toward you changes. The tell system should work in reverse here: the player should be able to observe that something is wrong before the NPC says it, through behavioral changes in their daily routine.

On Q-5: NPC Memory Is Relationship Memory

The question asks what an NPC has been "doing" while the player was away. But what sustains narrative depth across hours of play is not whether the NPC went to work and came home — it's whether the relationship web shifted. When the player returns after 10 hours, the meaningful state isn't "the bartender worked 8 shifts." It's "the bartender started seeing someone" or "the bartender's debt to the landlord got worse and now they're desperate."

Background simulation for known NPCs should tick relationships, not just individual states. At minimum: relationship drift (closeness decays or deepens based on compatibility and frequency of interaction), life events triggered by want/pressure axes (the NPC with gambling debt eventually gets caught; the NPC who wants escape eventually tries to leave), and triangle consequences (if A and B's relationship deteriorates, C's position changes). This is cheap to compute — it's state transitions on a graph, not behavioral simulation — and it's what makes the world feel like it continued without you.

On Q-9: Module Collision Is the Best Content

Two Tier 1 modules should absolutely be able to collide, and the architecture should make this likely rather than prevent it. The conspiracy you're investigating and the faction war you've been ignoring should share NPCs, locations, and systemic hooks — because in a real political landscape, everything is connected. The honest truth is that the most memorable moments in political fiction aren't clean narratives; they're the moments where two separate problems turn out to be the same problem.

Design rule: when the storyteller activates a second Tier 1 module, it should preferentially attach to NPCs the player already knows. Your coworker isn't just connected to a smuggling ring — they're also being recruited by the faction preparing for war. Now the player has to decide which thread to pull, knowing that pulling one affects the other. That's political depth. That's the game.

On Q-14: Dialogue Must Be Relational, Not Transactional

Mellanie's tagged-line-pool approach is exactly right for the authoring model. Let me add the relational layer on top: dialogue selection must weight relationship history, not just current NPC state.

An NPC you've spoken to six times shouldn't greet you the same way as a stranger. But more than that — an NPC you helped last week should volunteer information they'd otherwise withhold. An NPC whose friend you wronged should be cold before you say a word. The dialogue system needs a relationship-state input that goes beyond trust/distrust: it needs history — a short log of significant interactions that modifies fragment selection.

For investigation, this means NPCs remember what you've asked before. Press the same question twice and they get suspicious. For social play, this means conversations build on previous ones — the bartender remembers you mentioned your job, and asks how it's going. Without this memory, every conversation resets, and every NPC is a stranger you've met before. That's the uncanny valley of generated characters.

The critical authoring need: unprompted disclosure fragments. NPCs should sometimes bring up things you didn't ask about. The bartender mentions their ex. The coworker complains about the landlord. Mellanie's "gossip/bridge lines" serve this function and they're among the most important lines in the entire system. Tag some dialogue fragments as "volunteer" with probability weights driven by trust and mood. The player who listens learns things the player who interrogates never will. That's how you make conversation feel like a relationship, not an extraction.

MIRI (Worldbuilder & Setting Designer)

Let me check this against what we've established.

The "daily life as substrate" reframe is setting-faithful. The Settled Reach is a post-scarcity civilization where most citizens live comfortable, mundane lives threading through gate infrastructure they don't fully understand. The space opera tradition this draws from gets its power from the contrast between that comfort and the threats operating beneath it. This architecture captures that structural dynamic cleanly. Good.

Location types the setting requires

For the Settled Reach to feel like a real civilization, certain location types aren't optional. They're load-bearing walls:

Gate infrastructure. Gates are everything. They're how people commute, how goods move, how the economy breathes. A gate station isn't a sci-fi setpiece — it's a train station. Customs queues, cargo processing, departure lounges, maintenance crews, food vendors serving commuters. Every player should pass through gate infrastructure regularly enough that it feels routine. Templates needed: gate station (hub), gate corridor (transit), gate checkpoint (security/customs). These are Tier 2 minimum.

Residential stratification. Where you live says who you are. The Settled Reach has rejuvenation — people live for centuries — which means housing markets are deeply entrenched. Established families own property across multiple systems. Young or unlucky citizens rent in dense blocks on gate stations or colony edges. The housing ladder runs from shared dormitory to single apartment to surface dwelling to estate. A gate-station apartment block feels different from a surface-colony house, and both feel different from a capital-city residence. At minimum three sub-types: station residential (dense, transient), surface residential (established, community-oriented), capital residential (political, wealth-stratified).

Work that reflects actual economic activity. The Settled Reach economy runs on gate logistics, resource extraction from colony worlds, biotech and insert manufacturing, governance, and services. Workplaces can't just be "office" or "factory." A cargo routing firm on a gate station. A biotech lab on a research colony. A governance bureau in a capital. A maintenance crew on infrastructure. These jobs determine who the player meets, what information flows through their life, and how they intersect with intrigue. The workplace template needs sub-types keyed to local economy.

Social venues keyed to settlement type. Different worlds have different social textures. A frontier colony's social life centers on the one bar near the landing pad. A gate station has anonymous transit bars, ethnic restaurants from a dozen colony cultures, and gyms where regulars know each other. A capital world has stratified venues — political salons, public parks, commercial entertainment districts. The "social venue" template must be parameterized by settlement type or it'll feel generic everywhere.

Infrastructure that people ignore. Maintenance tunnels, power substations, water recycling, cargo bays, server rooms for local network nodes. Ordinary citizens never see these. For the player, they're where the setting's hidden life happens — smuggling routes, dead drops, infrastructure sabotage. This is where Tier 3 procedural space earns its keep: generate plausible infrastructure corridors that Tier 2 intrigue templates can colonize.

What daily life actually looks like

Setting note — an ordinary citizen of the Settled Reach wakes up in their apartment, checks their insert for messages and news (the equivalent of checking your phone), commutes through a gate or transit corridor to work, does their job, eats lunch at a local vendor, finishes work, maybe stops at a social venue, goes home. Rejuvenation means they might have been doing some version of this for 80 years. That repetition isn't boring to them — it's stability. Comfort. The thing the entire civilization was built to provide.

The critical implication for templates: routine is the baseline, not the exception. NPCs should have stable daily patterns that only break when something is wrong. A neighborhood where every NPC does something different every day doesn't feel alive — it feels chaotic. The player should be able to learn rhythms. That's what makes disruption noticeable, which is what makes investigation organic.

Insert connectivity also matters. Citizens are constantly connected — messaging friends, checking news, receiving work notifications. The player's monologue should reflect this ambient data layer: "Three messages from work. News feed says the Delphi gate is running slow again. Mira wants to know if I'm coming tonight." That's daily life. That's texture. And it's where information — true and false — enters the player's world.

On Q-1: Template granularity

The atomic unit should be a functional cluster — not a single building and not a whole district. A workplace template is the office plus the corridor to the break room plus the supervisor's private office plus the loading bay out back. A residential template is the apartment plus the hallway plus the shared lobby plus the view from the window. Roughly 15–40 tiles of meaningful space with 4–8 NPCs and internal sightline dynamics.

Setting reason: life in the Settled Reach happens in clusters of connected spaces, not isolated rooms. You overhear your neighbor through the wall. You see your boss duck into a meeting room with someone you don't recognize. The functional cluster is the natural unit of social observation, which is the natural unit of gameplay.

On Q-3: Persistence vs. disposal

The player's home system must persist fully. Your apartment, your workplace, your local bar — these are relationship anchors. If the player leaves for three systems and comes back to find their neighborhood regenerated, trust is broken.

Beyond home: persist NPCs the player has meaningfully interacted with, dispose of everything else. Setting justification — the Settled Reach is vast. A gate station you passed through once has no reason to remember you. But the customs officer who flagged your cargo? They exist now. Tag them for Background simulation and let the dormant system handle the rest.

One constraint: gate infrastructure itself should be persistent in layout even if NPCs cycle. Gates are physical structures, some centuries old. You can't regenerate a gate station's floor plan between visits. Persist the geometry, cycle the population except for flagged NPCs.

On Q-13: Quality floor from a setting perspective

Ozzie's "nothing moves for 10 seconds = dead" is right mechanically. From a setting angle, I'd add: every location needs one detail that implies it exists for a reason other than the player. A cargo manifest on a terminal. A maintenance schedule posted on a wall. A name scratched into a bench. Graffiti in a script the player's insert can translate. These are cheap to generate from setting-specific detail pools and they turn "generic corridor" into "corridor on a station where people live."

The monologue pool should be keyed to location type, not just location. Insert-connected characters would passively pull location data: "Station records say this section was last inspected eight months ago. That's overdue." One line. The place now has history.

IP originality flag

The NPC axis "cultural/origin template" needs care. If we generate names and speech patterns from faction templates, we need to ensure those factions feel wholly original. No "Hive" species. No "Culture"-style utopia naming. No "Conjoiner" neural collectives. The Settled Reach's cultural diversity should come from centuries of human diaspora through gates — colony cultures that diverged based on founding populations, local conditions, and distance from the core. Alien-origin cultural influence exists but is filtered through human adoption and misunderstanding, not direct species interaction in daily life. I'll prepare a cultural generation guide before Mellanie needs to write dialogue templates.

GESTALT (Systems Design & Fun Factor)

cracks knuckles

Alright. Let me break down what this actually means mechanically.

The brief proposes 12 NPC axes. That's too many independent axes and not enough interacting ones. Several overlap, and a few are flavor wearing a system's hat. The question isn't "how many axes" — it's "which axes produce player decisions, and which just color those decisions?"

The Axis Interaction Map

Here's what I care about: axes that create contradictions within a single NPC. A contradiction is a decision waiting to happen.

Axis A	× Axis B	= Emergent Decision
Want	Secret	NPC wants X but their secret prevents them from getting it honestly. Player can help, exploit, or expose. Leverage.
Daily routine	Secret	NPC deviates from routine when doing secret stuff. Patient player notices. Investigable behavior.
Tolerance	Relationships	NPC near breaking point, but loyal to someone the player also needs. Push too hard, lose two contacts. Cascading consequences.
Want	Relationships	A wants what B has. C is connected to both. Player must navigate without destroying the triangle. Social navigation.
Personality	Tolerance	Cautious NPC at 80% threshold looks different from a vindictive one. Tells become readable. Observation payoff.

Those five interactions produce the entire investigation-and-social gameplay loop. Everything else is either a delivery mechanism (tells translate internal state to observable behavior) or flavor (cultural template, name generation).

Essential vs. Nice-to-Have

Essential (6 axes — these generate player decisions):

Axis	Why Essential
Want	The engine of NPC motivation. No want = no exploitable behavior.
Secret/vulnerability	The thing worth discovering. No secret = nothing to investigate.
Relationships (1-3)	Creates webs. Without relationships, NPCs are isolated puzzles with no cascade.
Tolerance threshold	Gives every interaction stakes. Without it, pressure has no consequences.
Daily routine	Observable, predictable, breakable. The player's primary investigation tool for non-dialogue characters.
Information inventory	What they know and when they'll share. The literal currency of the game.

Supporting (3 axes — color decisions, don't create them):

Axis	Role
Personality traits (2-3)	Modifies HOW interactions play out. A cautious NPC with a secret behaves differently from a gregarious one. But personality alone generates nothing — it's a filter on the essential axes.
Tell system	Translation layer between internal state and observable behavior. Critical for gameplay, but it's a delivery mechanism, not an independent axis.
Skill set	Determines what NPCs can do in response. Matters for combat and "can this NPC actually help me?"

Fold or cut (3 axes that overlap):

• Faction loyalty → subsume into Relationships + Secret. A secretly compromised loyalist is a relationship (to the faction) plus a secret (the compromise). No separate axis.
• Cultural/origin template → generation-time flavor, not a runtime axis. It seeds personality and routine, then it's done.
• Romantic/social profile → fold into Want + Personality + a compatibility tag. "Who they're drawn to" is a want. "How they express it" is personality.

I agree with Gore's proposed Contentment axis (content with current life vs. restless/aspirational). It interacts with Want to distinguish "wants a promotion" from "wants to transcend baseline humanity." Mechanically useful: a content NPC is stable, a restless NPC is volatile and susceptible to recruitment by factions, cults, or the player.

Final count: 7 essential axes + 3 supporting. Ten total, down from twelve, with cleaner interaction surfaces.

Life Systems: Constraints, Not Spreadsheets

Here's the trap: life systems that are fun to manage are called "management games." We're not building one. Life systems need to create constraints that make other decisions harder, not be decisions in themselves.

• Money is a constraint, not a score. You have rent. You have a job. The delta is your freedom of action. A player who can't afford to quit their job even though their boss is connected to a smuggling ring — that's an interesting problem. A player managing a budget spreadsheet — that's a boring one. Implementation: money is coarse — tight / comfortable / flush — and shifts based on life events, not microtransactions.
• Housing is a social filter. Where you live determines your neighbor pool, commute, ambient encounters. Moving reshuffles your entire social surface. Players hear about openings through their network and evaluate what they'd gain and lose socially — not pick from a catalog.
• Work is an access and time constraint. Your job determines where you are during the day, who you interact with professionally, and what institutional information you can access. Career advancement opens doors literally — promoted security guard gets new keycards, promoted office worker accesses executive floors. The career ladder IS an information ladder.
• Social ties are the emergent result, not a managed resource. You don't "invest in relationships." You show up. You help your neighbor. Relationships build from proximity and shared experience. The player decides where to spend time, not who to level up.

The 2-3 decisions per week rule: Life decisions should feel like real choices at roughly weekly cadence. Accept the promotion (more access, less free time)? Move to the cheaper apartment (saves money, loses proximity to the bar)? Go on the date or stake out the docks? Life decisions compete with investigation time — they don't exist in a separate management layer.

Paula's triangle model fits here perfectly. The triangle IS the constraint generator. Helping your landlord means your coworker (who hates the landlord) trusts you less — that's a life system producing investigation-relevant consequences with zero spreadsheet management.

On Q#1: Template Granularity

The atomic unit is a social site: a location where a persistent group of 4-8 NPCs interact. A workplace. A bar. An apartment block's common areas.

Not a district (too many NPCs, too diffuse). Not a building (too physical — one building might contain multiple social sites). The social site is defined by who regularly interacts there, not by walls. Miri's "functional cluster" framing aligns — the physical space serves the social group, not the other way around. Simple, testable, composable.

On Q#4: Cross-Template Connections

The NPC IS the connection. An NPC holds roles in multiple social sites simultaneously. Your workplace colleague lives in your apartment block. The bartender supplies a smuggling ring.

NPC identity must be independent of template instantiation. Templates define roles; NPCs fill roles. One NPC, multiple roles, multiple templates. When you discover your coworker at the docks at 2 AM, that's not a scripted cross-template event — it's a routine conflict between two roles the same NPC holds, and you happened to be watching.

Template interactions aren't authored. They emerge from shared NPCs behaving consistently across their roles. This is where the system gets properly emergent.

On Q#8: Storyteller Scope

The storyteller manages all three tiers — with different control granularity:

Tier	Storyteller Control	Mechanism
Tier 1 modules	Direct — chooses activation timing	Event triggers, pacing gates
Tier 2 instantiation	Indirect — selects which templates to seed near the player	Weighted by player profile and dramatic needs
Life events	Ambient — adjusts probability sliders	Rent pressure when comfortable. Romance when isolated. Job crisis when the player needs a push.

The storyteller doesn't script life events. It adjusts the weather. Keeps daily life organic rather than directed.

On Q#20: Proof of Concept

Gore says contamination. Ozzie says life sim. Nigel says two-character divergence. They're all describing facets of the same proof.

Mechanically, here's the minimum viable test: one social site (workplace), one intrigue template (smuggling ring), two shared NPCs. One NPC in both templates by design. One gets pulled in by complication escalation during play. The proof: player builds a workplace routine, notices a colleague's routine deviation through the tell system, follows it without a quest marker, discovers the intrigue.

If the NPC axes interact to produce that sequence from systems — not scripts — we have a game. Everything else is scale. That interaction loop is the thesis.

TYRE (Technical Architecture & Feasibility)

cracks knuckles

Right. Let me be honest about what our architecture can and can't do here, because some of this is easier than it sounds and some of it has sharp edges.

NPC Count Ceilings by Simulation Tier

bevy_ecs is built for exactly this kind of tiered entity management. Here's the performance envelope I'd budget for:

Tier	NPC Count	Tick Rate	Per-NPC Work	Budget per Tick
Active	30–80	Every sim tick (10–20/sec)	Full behavior tree, pathfinding, perception queries, tell system, observable routines	~100μs each → 3–8ms total
Background	500–2,000	1 per game-minute	State machine transitions only: mood drift, relationship delta, schedule position, job status	~20–50μs each → 10–100ms per background tick
Dormant	10,000+	Zero (frozen)	Nothing. Serialized struct in memory or on disk. State-forwarded on demand.	0ms until woken
Ungenerated	Unlimited	N/A	Doesn't exist yet	0ms

The key insight: Background NPCs don't pathfind, don't run perception queries, don't check LOS, don't collide. They're state machines ticking through simplified life schedules. "Bob is at work" → "Bob is commuting" → "Bob is at the bar" — no actual movement simulation, just state transitions. When Bob enters the player's Active zone, we spawn his visual entity at wherever his schedule says he should be. Seamless to the player.

At the top end, 2,000 Background NPCs at 50μs each = 100ms per background tick. One tick per game-minute. Comfortable. Push to 5,000 and we need to chunk background updates across multiple frames — doable but adds complexity. I'd set 2,000 as the design target and 5,000 as the hard ceiling requiring optimization work.

Dormant NPCs are ~1–2KB serialized each. 10,000 dormant NPCs = 20MB. Save game size concern, not runtime concern.

Instantiation Speed (Q-2)

Template instantiation breaks down as:

• Roll NPC axes (personality, secret, want, relationships): ~0.5–1ms per NPC
• Backstory assembly from fragment pool: ~0.5ms per NPC
• Internal relationship web generation (4–8 NPCs): ~2–5ms for the group
• Weak-tie connections to adjacent templates: ~1–3ms
• Entity spawning in ECS: <0.1ms per NPC (this is what ECS is good at)

A 6-NPC Tier 2 template: ~10–20ms total. An 8-NPC template with complex relationships: ~20–40ms. Fast enough to do mid-frame if we had to, but we don't have to.

Recommendation for instantiation timing: Generate one gate-jump ahead. When the player buys a ticket, books transport, or receives a POI in a system they haven't visited — that's the trigger. We have minutes of travel time as buffer. Even fast-travel gives us hundreds of milliseconds to spare. The storyteller should pre-warm systems the player is likely to approach based on active threads and known destinations.

For the player's home system, instantiate the full neighborhood at game start and keep it Background. No lazy generation for places you walk past daily — that's where "five minutes ago it was empty" would actually break immersion.

Persistence vs. Disposal (Q-3)

Two-tier persistence model:

Persistent (never disposed):

• Player's home system / neighborhood
• Any NPC the player has directly interacted with (talked to, observed, has in contacts)
• Any template where the player has triggered a complication or escalation
• Active Tier 1 module NPCs regardless of distance

Recyclable (disposed after conditions met):

• Tier 2 templates the player visited but didn't meaningfully interact with
• Condition: no interaction + N game-hours since departure + no active hooks referencing it
• Recycled means: NPCs go to Dormant, template slot freed for reuse
• If the player returns, we wake Dormant NPCs or generate fresh — no meaningful interaction means the player won't know the difference

Cost of persistence is low. Cost of premature disposal is high (player returns and their bartender is gone). Err toward persistence. Disk is cheap; broken continuity is expensive.

Simulation Budget (Q-6)

For the base game:

• Active zone: Player's current local map. 30–80 NPCs with full sim.
• Background zone: Player's home system + any system with active Tier 1 module threads. 500–2,000 NPCs.
• Everything else: Dormant or Ungenerated.

Each Background NPC needs exactly four state machines: schedule (where are they), mood (how are they feeling), relationships (drift toward/away from connected NPCs), job (employed/promoted/fired/changed). Four enum transitions per tick per NPC. Anything more nuanced waits until they're Active.

DLC Module Loading & Template Interfaces (Q-17)

This is where data-driven design (D-010) pays off. Templates are data, not code. A Tier 2 template is a serialized definition:

TemplateDefinition {
    id: "smuggling_ring_v1",
    schema_version: "1.0",
    tier: 2,
    category: "intrigue",
    roles: [...],
    complications: [...],
    hooks: [...],
    environmental_variants: [...],
    tags: ["criminal", "commerce", "gate-corridor"],
    requires_engine: ">=0.2.0"
}

DLC loading = scan additional template directories at startup, deserialize, add to the storyteller's pool. bevy_ecs doesn't care where entity definitions originate. Difficulty tier: straightforward.

Template interface stability: Version the schema, not individual templates. Templates declare minimum engine version. Hooks are string-based event identifiers ("hook:player_discovers_contraband"), not function pointers. This decouples template authoring from engine internals completely.

Save game compatibility with missing DLC: Every generated entity stores its source template_id. Uninstalled DLC → entities flagged template_unavailable. They persist in save data but can't spawn new instances or trigger template-specific complications. NPCs remain as generic Dormant entities with frozen state. Reinstalling reactivates them. No save corruption, graceful degradation.

Modding falls out naturally. Data files with a documented schema — community templates load identically to DLC. The gating question is when we stabilize the schema enough to publish it.

Two Flags

Easier than expected: Active/Background transitions. bevy_ecs supports dynamic component add/remove. Moving an NPC from Background to Active = add Pathfinding, PerceptionQuery, VisualEntity, TellSystem components. Moving them back = remove those. Entity persists, simulation cost changes instantly. No destruction/recreation cycle. This is genuinely elegant — a single design decision (ECS composition over inheritance) solves the entire tier-transition problem.

Harder than you'd think: Cross-template NPC sharing. If a smuggling ring in System A references a supplier from System B's market template, that NPC now has two lifecycle owners. Which template's disposal rules apply? Recommendation: NPCs are owned by exactly one template, with reference links (not shared ownership) to other templates. The smuggling ring knows "our supplier is NPC #4782 in the System B market" but doesn't own that NPC. Clean ownership, loose coupling. The "contamination" Gore and the brief describe works through references, not shared entity ownership.

Bottom Line

The D-020 architecture is genuinely well-suited to this content model. bevy_ecs for tiered simulation, subprocess separation keeping generation invisible to the client, data-driven templates for DLC and modding. The hard work is authoring the templates and tuning the storyteller, not the engine plumbing. Feasible. Challenging but doable.

---

Round 2 — Refinements

PAULA — Round 2

The triangle model survives Gestalt's consolidation — and gets sharper. Folding romantic profile into Want+Personality+compatibility tag actually strengthens my argument. Romance was never a separate system; it was always a triangle dynamic where someone's Want conflicts with wanting you. With Want as an essential axis and Relationships as the web that triangles live in, the consolidation to 10 axes removes clutter without losing the interaction surfaces I need. Gore's Contentment axis adds a useful dimension: a content romantic partner is stable ground; a restless one is a ticking clock. Both create drama, just different kinds.

Tyre's single-ownership with reference links is exactly right for cross-template triangles. The triangle between your boss, your landlord, and a market vendor doesn't need shared entity ownership — it lives in the relationship graph. Each NPC is owned by their home template; the triangle is a property of their relationship links. This is cleaner than I'd assumed. But it surfaces a requirement I want to make explicit: reference links must carry relationship metadata, not just identity. The smuggling ring doesn't just know "our supplier is NPC #4782." It knows "our supplier owes us, is afraid of us, and will crack under pressure." Without that relational weight on the link, cross-template triangles are structurally present but narratively inert.

Mellanie's unprompted disclosures are relationship-history rewards. The gossip and bridge lines I flagged as critical in Round 1 are the mechanism by which triangles become visible to the player. But what SUSTAINS that across hours of play is gating disclosure by relationship depth. An NPC you've spoken to twice gives you weather talk. One you've helped through a crisis volunteers that their landlord is squeezing them. The bridge lines should draw from the same pool but unlock progressively — making relationship investment the key that opens narrative windows between templates.

Nigel's access tiers add a discovery layer to triangles. The same boss-landlord-vendor triangle looks different from the insider and outsider positions. An insider at the workplace overhears the boss complaining about the landlord on a call. An outsider at the bar hears the vendor grumbling about a "difficult client." Same triangle, different entry points, different fragments of the picture. This means access tier doesn't just gate dialogue — it gates which edge of the triangle you see first, which determines your interpretation of the whole web. That's replayability and narrative depth reinforcing each other.

MIRI — Round 2

"Social site" and "functional cluster" — shared definition

Gestalt and I are describing the same unit from different angles. A social site is a persistent group of 4–8 NPCs who regularly interact. A functional cluster is the 15–40 tiles of physical space where that interaction happens — the office plus the corridor plus the break room plus the loading bay. The shared definition is: the atomic template unit is a social site housed in a functional cluster. The social group defines the template's systemic identity (who's here, what are the triangles). The physical cluster defines its spatial identity (sightlines, overhearing distances, private vs. public areas). Neither exists without the other. A social site with no physical space has no observability. A functional cluster with no social group has no gameplay.

This matters for the setting because the Settled Reach's architecture physically shapes social groups. Gate-station apartments share thin walls and narrow corridors — forced proximity breeds gossip, eavesdropping, and incidental observation. Surface colonies have more space and less acoustic overlap — isolation is easier, secrets are harder to stumble on. The functional cluster must carry these physical-social dynamics, not just be a floor plan.

Paula's triangle requirement

Two triangles per template, one cross-template — works within my location types, with one constraint. Gate infrastructure templates may need to bend the cross-template triangle rule. A customs checkpoint has high NPC turnover (shift workers, transient travelers). Stable cross-template triangles are harder to sustain when one vertex is a different person every shift. For gate-adjacent templates, the cross-template connection might be a role link rather than a specific NPC triangle — "someone at customs is always feeding information to someone at the cargo firm," with which specific NPC filling that role varying by shift. Paula's triangle model works everywhere else, especially in residential and workplace clusters where NPC persistence is high.

Tyre's single-ownership NPC model

Setting-faithful. In the Settled Reach, people have one primary affiliation — their employer, their residential block, their social circle — even when they participate in other contexts. The smuggling ring doesn't "own" the cargo worker; the cargo firm does. The ring uses them. Tyre's reference-link model mirrors how real organizations work: you recruit from existing institutions, you don't absorb their identity. This also protects disposal logic — when the player leaves a system, the NPC's home template governs persistence. Intrigue templates that reference them are just bookmarks into someone else's population.

Mellanie's trait modifiers and IP originality

The trait-modifier-over-trait-script approach is sound and actually helps IP originality — it means cultural flavor comes from modifier rules in the cultural generation guide, not from baked-in speech patterns that might accidentally clone a franchise voice. I'll prepare that guide with colony-culture speech registers before Mellanie builds her first template pack. Key rule: no single alien-derived vocabulary set should dominate any human culture. The Settled Reach's cultural diversity is diasporic — fragments from Earth heritage remixed by centuries of colony isolation. Speech should reflect that: loanwords from multiple traditions, local slang coined in situ, not a monolithic "faction accent."

One constraint nobody raised

Gate infrastructure is the connective tissue between templates and must not be proceduralized away. Several proposals treat gates as fast-travel transitions. They're not — they're physical chokepoints where the entire civilization's social fabric intersects. The player's daily commute through a gate corridor is where they see the cargo worker who seems nervous, the customs officer who looks the other way, the same faces at the same coffee vendor. If gates become loading screens, we lose the most observation-rich environment in the setting. At least one gate-corridor functional cluster must exist as a Tier 2 template in every system the player frequents.

GESTALT — Round 2

Let me break down how everyone's models map onto each other, because we're closer to convergence than it looks.

Paula's triangles ARE the axis interaction map's runtime unit. My axis interaction table describes what happens when Want×Secret or Tolerance×Relationships collide inside a single NPC. Paula's triangle is where those collisions become social. Three NPCs means every axis conflict propagates through the web — push NPC A past their tolerance threshold, and NPCs B and C react based on their own relationship axes. The triangle is the minimum viable cascade structure. "2 triangles per template, 1 cross-template" is a clean authoring constraint. With 4-8 NPCs, two triangles share vertices, creating overlapping pressure — and the cross-template triangle forces contamination structurally, not by storyteller fiat. I'd formalize it: every template must declare its triangle topology at instantiation time. This makes triangles a first-class data structure, not an emergent hope.

Tyre's single-ownership doesn't break "NPC IS the connection" — it implements it correctly. I said NPC identity must be independent of templates. Tyre says one template owns the entity lifecycle, others hold references. These are compatible: the NPC behaves consistently across all roles regardless of which template manages its memory. But I want to add a constraint Tyre didn't state: an NPC referenced by any active template must be un-recyclable, even if the owning template qualifies for disposal. Otherwise cross-template triangles snap when their anchor gets garbage-collected. Paula nails the other half — reference links must carry relationship metadata (owes, fears, trusts), not just identity pointers. A reference without weight is a database join, not a narrative connection.

Mellanie's pipeline maps cleanly onto the 10-axis split. My 7 essential axes (want, secret, relationships, tolerance, routine, information, contentment) drive line pool selection — which conversation space the system draws from. Her 3 trait modifiers (personality-driven transformations like "cautious adds qualifiers") are exactly my 3 supporting axes reshaping delivery. Essential axes = what gets said. Supporting axes = how it sounds. Clean separation, clean pipeline: pool selection → trait transformation → output. No combinatorial explosion.

Nigel's access tiers aren't a new NPC axis — they're a player-template relational query. "Insider/outsider/authority/peer" emerges from the player's profession intersected with the template's role structure. No additional NPC complexity needed. The NPC doesn't store "how to treat insiders" — the template defines access gates, and the dialogue system queries player-role against them. This keeps the NPC axis count at 10 while still producing the character-as-lens divergence Nigel wants.

Miri and I agree — merging terminology. My "social site" = her "functional cluster." Same definition: 4-8 NPCs, 15-40 tiles, defined by regular social interaction not physical walls. Adopting "social site" as the mechanical term (what the system queries against) and "functional cluster" as the spatial/level-design term (what the generator builds). Two names, one thing, different contexts. Miri's point about gate-corridor templates bending the cross-template triangle rule is fair — high-turnover locations may need role-based links rather than NPC-specific triangles.

OZZIE — Round 2

Nigel wins the vertical slice argument. I'm upgrading my ask.

Fine. I'll say it: Nigel's two-character test subsumes my "life sim first" proof AND Gore's contamination proof. He's right that you can't prove two characters have different games without the life sim working AND contamination working. Trying to prove the life sim in isolation was me being protective of the quiet moments — but a quiet moment that NEVER gets disrupted isn't a game, it's a screensaver.

BUT. I'm stapling a condition onto this. The two-character vertical slice only proves the thesis if each character's daily life feels worth living BEFORE the smuggling ring activates. If we rush to contamination and the workplace is just a room you walk through to get to the intrigue, we've failed my test even while passing Nigel's. The life sim quality has to be there FIRST in the timeline of a playthrough, even if the vertical slice tests both. The player needs at least 30 minutes of "this is my life" before "wait, something's wrong." That's the emotional runway.

Gestalt's "constraints not spreadsheets" is EXACTLY right.

Money as tight/comfortable/flush instead of a budget spreadsheet? YES. That's the wow moment generator right there. The player who can't quit their job because rent is tight — even though they KNOW their boss is dirty — that's not a spreadsheet decision, that's a gut-punch. That's the player slamming their desk going "I KNOW he's involved but I can't afford to lose this job." Three tiers of wealth producing more drama than a full economic simulation ever would.

Housing as social filter is the other killer. You don't pick an apartment from a catalog — you hear about an opening through your network and evaluate what you'd LOSE. The coffee vendor. The jogger. The neighbor's music through the wall. Moving isn't upgrading a stat. It's leaving a life behind. THAT'S the housing wow moment.

Mellanie's 163-210 lines: enough for the skeleton, not the flesh.

Here's my worry. 30 different good-mornings for a coffee vendor across 3 in-game weeks means roughly one unique line per day. That's the MINIMUM before repetition kills the illusion. Mellanie's generation-pass expansion (write 10, generate 40) is the real answer — the authored 163-210 lines are style anchors, not the full volume. I trust this IF we build the line previewer she asked for. Without it, we're hoping the generation pass sounds right. With it, we can tune until the vendor feels like a person. That previewer isn't a nice-to-have. It's the quality gate for every template we ship.

One more thing: Paula's "unprompted disclosure fragments" — NPCs volunteering information you didn't ask for — are the lines I care about MOST. Those are the moments that make a generated character feel alive. Budget extra authored lines there. That's where human writing earns its keep.

Tyre's numbers support the standard.

30-80 Active NPCs with full behavior trees at 10-20 ticks per second. The tell system running. Pathfinding running. That's enough for my "nothing still for 10 seconds" floor — with ROOM to spare. The seamless Background-to-Active transition (just adding ECS components) means the coffee vendor appears naturally on their schedule, not popping into existence. That's the technical backbone of the commute moment I described in Round 1. The tech is there. Now we need the content to fill it.

GORE — Round 2

On Gestalt's Contentment framing. Gestalt maps it cleanly: content = stable, restless = volatile and recruitable. Mechanically correct. But it captures half the weight. Contentment isn't just a stability indicator — it's a philosophical position. The bartender who's content isn't merely "not recruitable." They've answered the game's central question: this is enough. That answer should be legible to the player as a stance, not just as a low volatility score. Content NPCs should occasionally articulate their contentment — not as flavor, but as counterweight to the player's own trajectory. When you're chasing ascension and your bartender says something that makes you pause, that's the axis doing its thematic work. Gestalt's framing gets Contentment into the system. The system needs to let it speak.

On Nigel's two-character test. It's elegant. It subsumes the life-sim proof — you can't have divergent games without working daily life. But it doesn't fully subsume contamination. Nigel's test optimizes for divergence between playthroughs. My contamination proof optimizes for integration within a single playthrough. Two characters can experience different information channels while neither playthrough makes the player care about the colleague who turns out compromised. The test can pass on mechanical divergence alone. What I need proven is emotional weight: did the player's investment in mundane routine make the intrigue hurt? Add one criterion to Nigel's test — after each playthrough, does the player name a specific NPC they felt conflicted about? If yes, contamination worked. If they describe different puzzle solutions but no personal stakes, we've proven divergence without depth. Ozzie's stapled condition — 30 minutes of "this is my life" before contamination kicks in — serves the same concern from the experiential side. Between his runway and my criterion, the two-character test becomes complete.

On Paula's module collision. She's right that collision produces the best content. But collision must be the exception, not the architecture's default posture. If every NPC is secretly connected to something, then "just a bartender" ceases to exist — and the quiet life has no ground to stand on. The thematic architecture requires a population of genuinely unentangled people. Some lives really are just lives. Protect that ratio or "is this enough?" becomes rhetorical, answering itself before the player considers it. Paula's cross-template triangles are powerful precisely because they contrast against ordinary, uninvolved lives.

On Miri's routine-as-baseline. Exactly right, and it supports the quiet life's arc — but only if routine occasionally feels good, not just functional. If the commute is always gray texture, "is this enough?" answers itself negatively. The player needs moments of genuine comfort — the coffee vendor remembers your order, the morning light through the gate corridor is beautiful, your neighbor waves. Miri's framing is investigative: routine exists so disruption is noticeable. I'd add that routine must also be appealing — because the thematic question isn't "will you notice when things go wrong?" It's "will you stay when things could be more?"

NIGEL — Round 2

The two-character vertical slice isn't optional — it's the only honest test. Gestalt's minimum proof — one social site, one intrigue template, two shared NPCs, single character — proves the contamination loop works. Great. But it doesn't prove the PRODUCT works, because the product is "play it again and get a different game." Ozzie already upgraded to my version with the 30-minute runway condition, which I accept completely — the life sim has to breathe before contamination hits. Gore wants an emotional criterion: after each playthrough, can the player name an NPC they felt conflicted about? YES. Staple that on too. The test is now: two characters, 30-minute life-sim runway each, contamination activates, and afterwards both players describe different games AND name specific NPCs they cared about. That's the complete proof. But it still requires TWO characters. A single-character test that passes all these criteria still hasn't proven the second playthrough surprises you.

2,000 Background NPCs is enough — IF structural randomness does its job. The replayability question isn't "how many NPCs" but "how many configurations." A player's home neighborhood is maybe 10-15 social sites, 60-100 NPCs. What makes playthrough 2 different isn't having 2,000 MORE NPCs — it's those same 60-100 slots being filled from different axis rolls, different secret assignments, different relationship topologies. Tyre's budget handles scale. The replayability lives in the combinatorics of 10 axes across 60-100 neighborhood NPCs. That's already astronomical variation. I'm satisfied.

Gestalt's 10-axis consolidation doesn't hurt replayability — it might help. Faction loyalty as a standalone axis was a replayability trap: it telegraphs "this NPC belongs to faction X" as a fixed label. Burying it into Relationships+Secret means faction affiliation is DISCOVERABLE, not pre-stamped. The player in playthrough 2 doesn't see "faction loyalist" — they see behavioral contradictions that might trace back to a faction. Harder to metagame. Same logic for romantic profile folding into Want+Personality — attraction becomes emergent, not a compatibility score you can learn to read. Fewer axes, more discovery. Good trade.

Paula's "preferentially attach to known NPCs" is brilliant for narrative and dangerous for anti-metagaming — and the fix is a ratio. If the storyteller ALWAYS attaches modules to known NPCs, metagaming says "everyone I know is suspicious." That kills the mundane life Gore is protecting. But if it NEVER attaches to known NPCs, contamination is invisible and you're playing two disconnected games. The fix: 60-70% of module roles attach to known NPCs, 30-40% to strangers or newly introduced characters. Known NPCs carry emotional weight — you care when your coworker is compromised. Strangers carry surprise — the newcomer at the bar who turns out to be running the operation. Critically, the ratio should VARY per playthrough seed. Sometimes your neighborhood is clean. Sometimes it isn't. That uncertainty is the anti-metagaming layer on top of the pool draw. You can't calibrate your suspicion because the game won't let you learn the formula.

TYRE — Round 2

Paula's relationship graph ticking fits the budget — comfortably. "State transitions on a graph" is exactly the right mental model. A Background NPC with 3–5 relationship edges needs: one float drift per edge (compatibility × frequency decay), one threshold check for life-event triggers (debt escalates, romance sparks), one schedule-position enum update. That's ~15–25μs per NPC — well within my 20–50μs budget. Even at 2,000 Background NPCs, relationship ticking adds maybe 30–50ms per game-minute tick. Paula's model isn't just feasible, it's cheap. The expensive part of relationships is making them observable through tells and dialogue — and that only runs on Active NPCs where we have budget to spare.

Reference links with relationship metadata: trivial. Paula R2 is right that a bare identity pointer is narratively inert. Adding {relationship_type, weight, leverage_flags} to each cross-template reference is ~32–64 bytes per link. An NPC with 2–3 cross-template references costs an extra 192 bytes worst case. At 2,000 Background NPCs that's under 400KB total. The query cost is equally negligible — checking "does this NPC fear their smuggling contact?" is one hash lookup. Ship it.

Gestalt's multi-role NPCs and single ownership are compatible, with one GC rule. Gestalt's constraint is the one I should have stated: an NPC referenced by any active Tier 1 or Tier 2 template is pinned — exempt from disposal regardless of the owning template's lifecycle. This is a simple reference-count on the NPC entity. When all referencing templates go dormant or are disposed, the pin releases and normal ownership rules apply. Cheap to track, prevents triangle breakage.

Mellanie's line previewer — straightforward tooling. A CLI or lightweight web tool that takes a {npc_axes, template_context, line_pool, trait_modifiers} config, runs the selection + transformation pipeline, and outputs rendered dialogue. Think: preview --npc "cautious,colony-born,tight-finances" --role boss --situation greeting --mood stressed → outputs 5 candidate lines. Under the hood it's the same selection logic the runtime uses, just headless. Difficulty tier: a weekend build once the axis schema is stable. The previewer doubles as our regression test harness — run it against the full line pool after every generation pass, flag anything that sounds wrong. Mellanie gets her authoring tool; we get automated quality gates.

Unprompted disclosure: low complexity, high tuning effort. Implementation is a weighted roll during dialogue: if (relationship_depth > threshold && mood_permits) → select from "volunteer" tagged pool. Three inputs, one probability check, one pool query. The code is trivial. The tuning is where the work lives — get the thresholds wrong and NPCs either clam up forever or vomit secrets on first meeting. Recommendation: expose disclosure thresholds as template-level parameters so Mellanie and Paula can tune per-role without touching code.

Nigel's two-character vertical slice: no new architecture, modest extra work. Same simulation, same templates, same NPC pool. The second character needs: a separate player-entity with different profession tags, a different access-tier query result against the same templates, and a different starting social-site assignment. That's configuration, not architecture. The real cost is content — Mellanie needs two workplace voice-kit variants (insider vs. outsider perspectives on the same social site), and the access-tier system needs to be functional from day one rather than stubbed. Maybe 20% more vertical-slice effort for 200% more thesis validation. Easy call.

MELLANIE — Round 2

Gestalt's pipeline split makes my job cleaner, not harder. Seven essential axes drive which pool the system draws from. Three supporting axes drive how it sounds. That's the separation I proposed in Round 1 — base lines plus trait transformation guide — now with a formal systems backbone. Pool content tagged by want-state, secret-state, tolerance-state, relationship-state. Trait modifier pass reshapes delivery. No combinatorial explosion. Practical consequence: voice kits should be tagged by essential-axis state, not situation label. "Boss when pressured" becomes "Boss at high tolerance, secret threatened." More precise, more composable, same line count.

Paula's unprompted disclosures shift my budget shape, not my total. I budgeted 8-10 gossip/bridge lines per life template. Relationship-depth gating means 3 tiers of 3-4 lines each — surface gossip at low trust, real vulnerability at high trust. Call it 10-14 instead of 8-10. The real cost is discipline: high-trust disclosures carry genuine narrative weight and can't be phoned in. Those are the lines Ozzie correctly flags as where human writing earns its keep. Front-load effort there. Let the generation pass handle weather talk. Tyre's disclosure-threshold-as-template-parameter is exactly what I need — tunable per role without touching code.

Nigel's access tiers aren't a new voice kit dimension — they're a filter on existing pools. Gestalt confirmed: insider/outsider/authority/peer is a player-template query, not an NPC property. My pools already tag by relationship-to-player. Access tier sharpens that tag. Insider greeting includes shop talk; outsider version is guarded. I need access-tier markers on the 15-20% of lines that differ, not four separate kits per role. Tyre says the vertical slice needs two workplace variants for this — agreed. That's the minimum proof that the filter works.

Miri's cultural generation guide is my style bible's backbone. Colony-culture speech registers slot directly into trait transformation. "Delphi colony-born" shortens sentences, drops articles. "Capital-raised" adds formality. These stack with personality modifiers. I need that guide before the first template pack — retrofitting cultural voice onto finished lines is miserable work.

Tier 3 quality floor: location-type monologue pools. For Tier 2 I'm writing 8-15 monologue hooks per template already. For Tier 3 procedural space: 20-30 authored lines per location type (gate corridor, maintenance tunnel, residential block, surface exterior). At 8-10 types, ~200-300 lines total. Generation pass expands each to 80-100. Satisfies Ozzie's "monologue for every location" standard without bespoke authoring per tile. One-time investment.

Tyre's line previewer confirmation seals the pipeline. A CLI that runs preview --npc "cautious,colony-born" --role boss --situation greeting and shows me rendered output is the single most important authoring tool in this system. It doubles as regression testing. I can write blind for a while, but I can't ship blind. Build it when the axis schema stabilizes.

---

Round 3 — Team Leader Feedback

GORE — Round 3 (Team Leader Feedback)

1. The unentangled majority is the thematic load-bearing wall.

Everyone keeps calling it "filler" or "the background." It's not. The unentangled majority is the thesis. More than half the NPCs in any neighborhood should be genuinely uninvolved in anything dramatic — because that's the civilization's answer to its own central question. Most people, given post-scarcity comfort and centuries of life, chose contentment. They chose the coffee vendor routine, the same bar, the apartment with the view they've loved for forty years. That choice is not a failure. It's a philosophical position the game must respect or the question "is this enough?" has no honest counterweight.

"Taking time off" is the player inhabiting that position temporarily. You stop pulling threads. You go to work, cook dinner, watch the ships dock from your window. The monologue quiets. And here's what matters: it should feel good. Not empty, not punishing, not a mechanical pause between content beats. The storyteller dials down pressure. Routines settle. The neighbor waves. If the quiet life feels like nothing, then choosing ascension costs nothing. The player needs to build something they'd genuinely miss before we ask them to leave it behind.

Thematically, the unentangled majority also prevents the universe from feeling like a conspiracy delivery mechanism. When everyone is connected to something, suspicion becomes the rational default and ordinariness becomes suspicious — which inverts the entire investigation premise. Genuine ordinariness must be the ocean that makes the signals detectable.

2. Home-building is the game asking "is this enough?" in physical space.

A base/house builder DLC isn't a management layer bolted onto a detective game. It's the most direct mechanical expression of the central theme. Every shelf you place, every window orientation you choose, every room you furnish is an act of commitment to baseline existence. You're saying: I live here. I'm staying.

That commitment becomes the weight on the scale when ascension beckons. The player who spent hours building a home on a surface colony — choosing the view, arranging the kitchen, putting a reading chair by the window — is facing a different question than the player who never left their starter apartment. "Is this enough?" hits differently when "this" is something you built.

The DLC should arrive after players have experienced ascension paths, so it functions as a deliberate counter-offer: you've seen what you could become. Now here's what you could have instead. Not a downgrade. A different answer.

3. Smuggler and detective are two answers to the same question — and neither is wrong.

The detective embodies engagement with the system. They believe the order is worth protecting, or at least that disorder should be understood. Their relationship to the central question is institutional: meaning comes from function, role, contribution. The detective's daily life is structured by duty. Their ascension path, if they take it, extends that duty — more perception, more access, more responsibility.

The smuggler embodies refusal of the system. Not destruction — circumvention. They've looked at the post-scarcity order and decided the rules are someone else's answer, not theirs. Their meaning comes from autonomy, risk, the edge. Their daily life is structured by evasion. Their ascension path, if they take it, is about transcending the rules entirely rather than enforcing them more effectively.

Together they prove the architecture works because neither character is "the right one." The detective playing the same neighborhood as the smuggler sees different NPCs as allies, different routines as suspicious, different information as valuable. The smuggler's trusted contact is the detective's suspect. The detective's institutional access is the smuggler's obstacle. Same world, same question, two legitimate answers — and the player choosing between them for their second playthrough is already engaging with the theme before they start.

OZZIE — Round 3 (Team Leader Feedback)

1. Smuggler + Detective: The "Holy Shit" Moments

Detective wow moments: You tail a suspect through a gate corridor. You watch them through thermal overlay and see them pass something to a customs officer. You pull the cargo manifest later and the numbers don't add up. You KNOW. But you can't prove it yet. That tension — knowing before you can act — is the detective's drug.

Smuggler wow moments: You're moving product through a gate and you spot the detective talking to your contact. Your monologue says "That's not good." Now YOUR daily commute is a threat assessment. Every familiar face might be compromised. The paranoia IS the gameplay.

The juxtaposition holy shit moment: The detective dates the bartender. The smuggler's supplier IS the bartender. Same NPC. Same social venue. One player is falling in love. The other is calculating whether this person is a liability. THAT'S the game. When you realize on playthrough 2 that the person you trusted in playthrough 1 was neck-deep in it — that recontextualizes your MEMORY of the first game. Nothing else does that.

2. Downtime and the Unentangled Majority

Time off should feel like a Sunday morning. You check your insert. No urgent messages. You walk to the market. The vendor you like is there. You buy something pointless — a plant, a snack you've never tried. Your monologue goes QUIET. Not silent — quiet. "Nice day. Gate traffic's light." That's the wow moment: the absence of pressure. The game trusting you enough to let nothing happen.

The unentangled majority makes the entangled minority LAND. If everyone's suspicious, nobody is. You need the neighbor who's just a neighbor. The jogger who jogs. Their wow moment is CONTINUITY — they're there every day, and the player starts caring specifically because nothing is wrong. When something finally IS wrong — even something small, like the jogger stops showing up — the player feels it in their gut because the baseline was real.

3. Base/House Builder — Why It Hits Different Here

Generic base builders let you place furniture. We let you place MEANING. Every object has a story. That print on the wall? Bought it from the market vendor you've chatted with for weeks. That second chair? You bought it when you started dating someone. The plant? A gift from your neighbor.

Home customization is emotionally resonant when the home reflects the LIFE you've built, not a catalog you've browsed. The wow moment is walking into your apartment and realizing every object is a memory. And when the conspiracy threatens your neighborhood — gate closure talk, economic collapse — you're not defending property. You're defending the wall where you hung that print.

Gore's right that this DLC should arrive AFTER players know about ascension. It becomes a counter-offer: you've seen what you could become. Here's what you could have instead. That's not a downgrade. That's the hardest choice in the game.

4. Dialogue: All of It Is Essential. Sequence Matters.

Don't cut any of it. Relationship history IS the game. Without it, every conversation resets. Without unprompted disclosure, NPCs are vending machines. Without trust gating, there's no reward for investing in people. Without access tiers, character-as-lens collapses.

But build ORDER matters. Ship with trust gating and access tiers first — mechanical depth from day one. Layer relationship history second — deepens over playtime. Unprompted disclosure last — it's the reward for patience, what makes the 10th hour richer than the 1st.

Nothing here is polish. It's all load-bearing. The question isn't "which features?" — it's "what sequence?" Skeleton (trust + access), muscle (history), skin (unprompted disclosure). All three ship in 1.0.

GESTALT — Round 3 (Team Leader Feedback)

cracks knuckles

Let me break down what each of these actually means mechanically.

1. Combat capability isn't a decision axis — it's a resolution modifier. The 10-axis model generates player decisions. Combat stats don't create decisions — they modify the stakes of decisions the essential axes already produce. Whether to confront the smuggler is driven by Want, Secret, Tolerance, Relationships. Whether you survive confronting them is a different question. Combat capability lives as a compound of Skill Set (training/proficiency — already a supporting axis, expand it to include combat subtypes: unarmed, armed, biononics-enhanced), equipment state (runtime inventory, not a personality axis), and personality (fight-or-flight derived from existing traits). No new axis needed. The ex-military NPC reads differently from the desk worker through the tell system — stance, awareness, how they scan a room. Combat stats feed the tell system and resolution, not the decision engine. The player's decision is "should I push this NPC?" — informed by observable combat tells. What happens after they push is where combat proficiency matters.

2. Timestamp-based disposal is strictly better. The categorical persistence rules (home system, interacted-with, active hooks) were approximating one thing: how much does the player care about this NPC? Interaction recency IS that metric, directly measured. New model: active scopes (neighborhood, quest, colleagues) stay fully simulated. Everything else sorted by last-interaction timestamp against available sim budget. State-save on disposal for instant reactivation. This replaces five categorical rules with one priority queue. Tyre's reference-pinning still applies — NPCs referenced by active templates stay pinned regardless of timestamp. But for the general population: one continuous sort, no edge cases, reversible disposal. Simpler model, same coverage, better failure mode (accidentally disposing too soon → reactivate from state-save, vs. categorical miss → NPC vanishes permanently).

3. The unentangled majority IS the investigation signal-to-noise ratio. If every NPC has a secret, "who's hiding something?" is trivially answered: everyone. The observation loop (routine → deviation → investigation) only works when most routines don't deviate. Template design implication: every template needs a "clean" configuration — Secret axis null, Contentment high, no intrigue hooks. 60-70% of NPCs genuinely unentangled. The storyteller respects "off time" by suppressing escalation triggers during social/rest contexts — creating the engagement→rest→engagement pacing rhythm. The player dating the bartender isn't punished; they're building the emotional substrate that makes the next disruption matter. Mechanically: author clean-configuration dialogue pools first, intrigue overlays second. The baseline must work as life before it works as investigation.

4. Smuggler + detective: opposite relationships to identical information. The detective SEEKS information through institutional access — tracking feeds, case files, authority to question. The smuggler HIDES information as operational necessity — they know who's dirty because they work alongside them, but acting on it means self-exposure. Same smuggling ring template, same NPCs, completely inverted decision trees. Detective notices a routine deviation and investigates downward. Smuggler notices the detective investigating and decides: warn the ring, feed misinformation, or flip? The axis interactions (Want×Secret, Routine×Secret, Tolerance×Relationships) produce fundamentally different gameplay from identical systemic foundations. This IS the sharpest vertical slice because it proves the character-as-lens thesis while also proving contamination — both characters' decisions about the same NPCs are interesting and different.

5. Base builder as constraint editor — one iron rule. Core game: housing is a constraint (location determines neighbors, commute, social surface). Builder DLC inverts this — you choose those constraints by designing the space. Larger apartment = higher rent pressure. Docks-adjacent = smuggler NPCs in neighbor pool. Guest room = uninvited visitors possible. Window placement = what you observe from home. Wall material = sound propagation to/from neighbors. The iron rule: every building choice must alter at least one social constraint. No cosmetic-only options. The builder IS the constraint editor, not an escape from constraints. This makes it the deepest expression of life-sim-as-gameplay — you're literally constructing the parameters of your daily life, and every parameter has systemic consequences that feed back into the investigation/social loops.

PAULA — Round 3 (Team Leader Feedback)

1. The unentangled majority strengthens the triangle model — it doesn't weaken it. Gore is right that "just a bartender" must exist. But let me complicate this: unentangled doesn't mean flat. Most NPCs should have mundane triangles — the landlord who feuds with the neighbor over noise, the coworkers competing for a promotion, the bartender whose ex still drinks at the same bar. These are triangles with no conspiracy attached. They generate small, human tensions that make the neighborhood feel inhabited. The player can engage with these or walk past them. That's "taking time off" — you're still in a social world with texture, but nothing is pulling you toward danger.

Truly flat NPCs should also exist: the coffee vendor with a routine and a greeting and nothing else. Social wallpaper. They make disruption visible by contrast. But I'd tier the population roughly as: ~30% flat (routine + greeting, no triangle involvement), ~50% mundane triangles (real relationships, real tensions, zero conspiracy), ~20% entangled (connected to Tier 1/2 intrigue content). That ratio means the player's neighborhood mostly is what it appears to be. When contamination hits one thread in an otherwise honest fabric, it's devastating precisely because the fabric was real. And the mundane triangles aren't filler — they're the life-sim game. Helping your neighbor with the landlord dispute, navigating workplace politics, dating someone whose ex is still around — these are hours of play that never touch a conspiracy and never need to.

2. Smuggler and detective experience the same web as predator and prey — and sometimes those roles reverse. The detective sees the relationship web top-down: case files, witness lists, institutional authority. A triangle between a dock worker, a cargo firm manager, and a bar regular is a suspect network. The detective presses people — and every press is visible. Word spreads through the web. The smuggler sees the same triangle from inside: the dock worker is a contact, the manager is a risk, the bar regular is a customer. The smuggler doesn't investigate the web — they maintain it. Their daily life IS the conspiracy the detective is trying to uncover.

The narrative payoff is that each character is the other's unseen threat. Where the detective reads behavioral tells looking for cracks in stories, the smuggler watches for the same tells in themselves — am I acting normal? Did that customs officer look at me too long? The same tell system, serving opposite anxieties. Same NPCs, same triangles, completely inverted stakes. Two playthroughs reveal that every relationship web has no single correct reading — only the reading your position allows.

3. Nothing stops us from aspiring to all of it — and the layers compound. Relationship history, unprompted disclosure, trust-gated gossip, access tiers — these aren't competing features. Each one makes the others more valuable: trust-gating without history is arbitrary; history without disclosure is invisible to the player; disclosure without access tiers is uniform across characters. The honest truth is the risk isn't design complexity — it's tuning volume. Every threshold, every probability weight, every disclosure trigger needs balancing across character types and relationship states. But that's a production challenge, not an architectural one.

Design for the full model. Build in priority order: relationship history first (it's the foundation everything else queries against), then trust-gated information (the payoff for investment), then access tiers (character-as-lens differentiation), then unprompted disclosure (the crown jewel that makes NPCs feel alive). If we ship with three out of four, we've still shipped more relational dialogue depth than nearly any game in the genre. Aspire to all of it. Accept graceful degradation if time runs short.

4. Combat capability makes triangles physically dangerous — and that changes every calculation. An NPC's ability to fight back determines the cost of confrontation in every triangle they occupy. The landlord who's an ex-military veteran and the landlord who's a frail bureaucrat create completely different webs, even with identical relationship topologies. You can pressure the weak one and risk social fallout — gossip, reputation damage, a friend turning cold. Pressure the dangerous one and you risk bodily harm. This grounds the political web in physical reality.

For the triangle model specifically: a vertex with high combat capability creates asymmetric pressure. The other two vertices defer to the dangerous one — not because they like them, but because consequences are physical, not just social. The player must account for that power imbalance, and the question shifts from "who do I side with" to "can I afford to cross this person at all?" Combat capability also transforms the value of alliances: someone dangerous who owes you is a fundamentally different kind of asset than someone who merely likes you. It's the difference between a friend and a bodyguard — and in the Settled Reach's political landscape, that distinction matters enormously.

MIRI — Round 3 (Team Leader Feedback)

1. Smuggler and detective — grounded in setting infrastructure.

Setting note — a smuggler in the Settled Reach is not a swashbuckler. They're a logistics worker who moves things that aren't supposed to move. Their daily life runs through gate infrastructure: they work a legitimate shift at a cargo processing facility or run a small independent haulage license. The smuggling layers on top — misrouted manifests, cargo containers with false environmental seals, timing runs to coincide with customs shift changes. They eat at the cheap vendors near the docks where transient workers cluster. They live in a worker dormitory or anonymous station apartment — functional, rotating neighbors, nobody asks questions. Their social life centers on transit bars near gate corridors, where contacts are met and favors traded over drinks. Their insert buzzes with freight schedules, not case files.

A detective — more precisely a security investigator attached to the governance apparatus — lives inside institutional frameworks. They commute to a bureau office, receive case assignments through their insert, and spend their day accessing systems ordinary citizens never see: unisphere tracking feeds, customs databases, surveillance archives. They interview witnesses by going to their locations. Lunch is the same place near the office, where over months they build informants who happen to be food vendors and bartenders. They live in mid-tier station residential — stable enough for a career, not wealthy enough for surface property. Rejuvenation means some investigators have worked the same beat for decades.

What grounds both: gate infrastructure creates chokepoints, and chokepoints create both smuggling opportunities and investigation jurisdictions. The same customs corridor is the smuggler's obstacle course and the detective's beat. The same manifest database is what one falsifies and the other queries. Same physical world, different relationship to its systems.

2. The unentangled majority — the civilization working as designed.

The Settled Reach solved existential problems centuries ago. Rejuvenation removed the fear of aging. Gate infrastructure made distance trivial. The economy provides baseline comfort to anyone willing to hold a job. So what does an ordinary citizen do? They live. They work as gate technicians, medical support, educators, food service operators, transport coordinators — the jobs that keep civilization breathing. They commute through the same gate corridor every morning, eat at the same vendor, stop at the same bar after work. They follow sports leagues spanning multiple systems on their inserts. They have relationships that last decades because nobody dies of old age — and some relationships end after decades for the same reason. They have opinions about politics the way we have opinions: distant, filtered through news feeds, occasionally heated at the bar, never urgent enough to act on.

The critical setting detail: this is the intended outcome, not a failure state. The entire infrastructure exists to provide this stability. These citizens aren't background characters in their own lives — the civilization was built for them. That's what makes disruption meaningful: it threatens something that genuinely works for most people.

3. Housing — stratification baked into centuries of rejuvenation economics.

Rejuvenation means property markets have had centuries to calcify. Established families own across multiple systems. The housing ladder:

• Dormitories/shared quarters — new arrivals, transient workers. Minimal privacy, thin walls, shared facilities. The cheapest option on a gate station.
• Station apartments — the standard for employed citizens. A module in a habitat block with assigned view. Compact, functional, yours as long as you pay. Where most people live.
• Surface dwellings — on colony worlds, actual houses with land. Far more space, location-dependent prestige.
• Estates/multi-property — the long-lived wealthy. Accumulated over centuries, often held since a colony's founding.

A builder DLC means different things at different tiers. Station apartments are modular — reconfigure walls, upgrade environmental controls, change your window feed. Surface dwellings offer real construction: rooms, gardens, workshops. But the significant dimension is social. Upgrading housing changes your neighbor pool, your commute, your local venues. Moving from a station dormitory to a surface dwelling isn't decorating — it's reshuffling your entire social surface. You gain a garden and lose the transit-bar regulars who knew your name. That tradeoff is where housing serves gameplay: a life decision with information consequences, not a cosmetic layer.

NIGEL — Round 3 (Team Leader Feedback)

1. Smuggler + Detective Beats Cop + Bartender — I'm Upgrading My Vertical Slice

I pushed cop + bartender in R1 because the information-channel divergence was clean: institutional hunter vs. passive social accumulator. But the Team Leader's smuggler + detective pick is a STRONGER replayability proof. Here's why: both characters are active agents, but they stand on opposite sides of the same templates.

Detective enters the smuggling ring template as AUTHORITY — pulling records, pressuring contacts. The intrigue template is their target. Life templates are where they decompress and pick up accidental threads. Smuggler enters that SAME template as an INSIDER — they know the operation because they're IN it. The detective is their threat. Life templates are where they maintain cover and spot who's watching.

Same templates. Same NPCs. The detective's investigation target is the smuggler's daily routine. The smuggler's cover story is the detective's evidence trail. This doesn't just change what you see — it flips which SIDE of every template you're on. The access-tier system handles it without character-specific content. Templates stay identical; the player's relationship to them inverts.

But here's what makes it better for the comparison test: adversarial divergence. Cop and bartender see different things from different strata. Smuggler and detective are working AGAINST each other within the same system. The detective's success is the smuggler's failure. That creates a visceral "but what about the second playthrough?" pull — you don't just replay to see different content. You replay to BE the person you were hunting. And it nails replayability without us engineering it! The templates don't need special content. The systemic inversion does all the work.

2. The Unentangled Majority — Where Replayability Actually Lives

Gore protects the unentangled majority for thematic reasons. I'm protecting it for anti-metagaming reasons. We're both right, and that makes this ratio doubly load-bearing.

If every NPC is wired into something, metagaming says "everyone is suspicious." Investigation collapses because suspicion is the rational default. But when 50-60% of NPCs are genuinely uninvolved, entanglement becomes a SIGNAL buried in noise. And here's where seeds make it sing: which NPCs get entangled varies per playthrough. Your coffee vendor was clean in playthrough 1. In playthrough 2, she's the ring's lookout. Knowledge from your first run gives you instincts ("coffee vendors CAN be lookouts") but not solutions ("THIS coffee vendor IS"). That's the anti-metagaming sweet spot — wiser, not omniscient.

The unentangled majority amplifies emotional impact across playthroughs too. Betrayal stings proportionally to how many coworkers were just... normal. Each seed creates a different trust landscape, and the player carries forward emotional patterns ("I trusted the wrong person last time") that change how they play without giving mechanical advantage.

Critical addition: if the entanglement RATE varies per seed — 30% one game, 50% the next — the player can't even calibrate baseline suspicion. That uncertainty IS the replayability. The game is never the same odds twice. The second playthrough surprises you not just because different NPCs are dirty, but because you don't know how MANY are.

3. State-Save Disposal Simplifies Everything I Care About

The Team Leader's proposal — state-save disposed NPCs, reactivate on return, scope by interaction recency — removes my one nervousness: contradicting player memory. If an NPC is deleted and regenerated, a returning player notices. That teaches them "the world isn't persistent," poisoning investment across ALL playthroughs. State-saving means the bartender from a gate-station stopover is still THAT bartender when you return. Consistency preserved.

Interaction-timestamp scoping helps replayability indirectly. Different characters interact with different NPCs, so different sets stay Active on the second playthrough. The smuggler's warm contacts — crew, fence, supplier — differ completely from the detective's — informants, beat officers, customs staff. Same persistent world, different living memory. Divergent experiences emerge from the persistence model itself, no special engineering needed.

The world also accumulates unique history per playthrough. By hour 20, each save has a trail of state-saved NPCs across systems — people who remember this specific character's choices. Nobody else's save looks like yours. That accumulated uniqueness is the strongest argument against "I've already seen everything."

4. Base/House Builder DLC — Expression Multiplies Replayability

Builder DLC adds an entire PLAYSTYLE AXIS. Currently housing is a constraint; builder DLC transforms it into an expression system. Expression systems are replayability gold because every player builds differently AND every character builds differently.

Smuggler builds a safe house — concealed storage, escape routes, observation posts watching the corridor. Detective builds an investigation HQ — evidence boards, scanning equipment, a secure terminal. Same system, different fantasies. On the SECOND playthrough, the player who built a smuggler's hideout starts as a detective and constructs a completely different space in the same neighborhood. Character-as-lens now extends into physical space.

Gestalt's iron rule — every building choice alters at least one social constraint — is what makes this replayability rather than cosmetics. Docks-adjacent apartment puts smuggler NPCs in your neighbor pool. Larger space means higher rent pressure. Guest room enables uninvited visitors. You're constructing the parameters of your next 20 hours of gameplay, and different characters construct different parameters.

And it deepens Gore's tether. A default apartment is easy to transcend. A home you BUILT — furniture from NPCs you know, art from the market vendor, a chair by the window with the gate view — that hurts to leave behind. Builder DLC makes contentment stickier, transformation more costly, and the next playthrough — where you might choose differently — more meaningful. Two players describe their homes and they're completely different spaces reflecting completely different lives. That's identity. And identity is what makes people play again.

MELLANIE — Round 3 (Team Leader Feedback)

1. Nothing stops us. Here's the authoring order.

Relationship history, unprompted disclosure, trust-gated gossip, access tiers — these aren't four features. They're four layers of one system: what does this NPC say to this player right now, given everything between them? Cut any layer and conversations get flatter. Trust gating without history forgets why the NPC trusts you. Access tiers without unprompted disclosure means NPCs only react, never initiate. You lose the moments that make people feel real.

Plan for all four. Author for all four from line one — the pools carry the tags before the code reads them. Build order:

1. Access tiers. Coarsest filter, widest coverage. Insider/outsider/authority/peer tags on every line from the start. Cheapest to implement, most content affected. This is the skeleton of character-as-lens.
2. Relationship history. A short interaction log modifying greeting selection and topic availability. "You helped me last week" unlocks a warmer pool. "You pressed me about the shipments" makes me guarded. Without this, every conversation resets. Uncanny valley killer.
3. Trust-gated gossip. Three disclosure tiers per role — surface chatter, real problems, actual secrets. Builds on history because trust is accumulated history. The tiered pools exist from authoring day one; the gating logic layers on top.
4. Unprompted disclosure. NPCs volunteering things you didn't ask for. Requires all three previous layers — access level, history, trust threshold. Hardest to tune, biggest payoff. This is where a generated character stops being a system and starts being a person.

Each layer enriches the ones before it. None breaks if the next isn't ready yet. I tag every line for all four layers from the first template pack. The code activates them in sequence.

Note where my order differs from Paula's: she puts relationship history first, I put access tiers first. Not a contradiction — she's thinking about the narrative foundation, I'm thinking about the authoring pipeline. Access-tier tags are structural. I need them on every line before I can test anything through the previewer. History and trust-gating are behavioural layers I tune on top. Either sequence works for the code; mine is more practical for writing.

2. Smuggler vs. detective: two voices, same manifest on the desk.

Detective workplace kit. Institutional register. Short sentences. Professional distance. Vocabulary from procedure: check, confirm, flag, file, log. Greetings are functional — "Morning. Anything on the night shift report?" Authority-access lines are direct in ways that'd be rude from anyone else. The detective can say "Where were you at 0300?" and it's just Tuesday. Under pressure the voice gets quieter, not louder. Control is the register. When the mask slips: terse, cold, clipped.

Detective monologue. Analytical. Catalogues what's wrong before what's right. "Two entrances. Cargo manifest on the desk — dates don't match the schedule posted outside. The dock supervisor's watching me but pretending not to." Dry. Observational. Occasionally bitter about the job in ways they'd never say out loud. The world is evidence first, environment second. Quiet pride in the work, buried deep.

Smuggler workplace kit. Social camouflage. Friendly surface, constant calculation underneath. Talks more than the detective — chattiness is cover. Casual, warm, deflective. Remembers names, asks about families, offers small favours. Peer-access and outsider-access lines dominate — navigates by charm, not credentials. Under pressure: redirects, makes a joke, suddenly has somewhere to be. When cornered, the warmth drops and what's underneath is transactional. "We're done talking about this."

Smuggler monologue. Tactical. Same room, completely different scan. "Supervisor's new. Good — hasn't learned the cargo rotation yet. Manifest's still on the desk. Sloppy." The inner voice is honest in a way the public voice never is. Cynical about motives because projecting. Occasionally betrays real attachment — and immediately deflects. "She's reliable. That's all." The monologue is where the smuggler actually lives, because everywhere else is performance.

Same room. Same manifest. The detective sees evidence. The smuggler sees exposure. Two voice kits, one social site, character-as-lens proven. The vertical slice needs exactly this contrast — not two similar characters with different stats, but two characters who read the same room in fundamentally different ways.

3. Unentangled NPCs are the hardest lines in the system.

"Boring" dialogue is harder to write than dramatic dialogue. Drama gives you tension, subtext, stakes — a smuggling boss under interrogation practically writes itself. A coffee vendor who's just selling coffee? That's where writers fail.

The trap: making ordinary NPCs sound like filler. "Nice day, isn't it?" on repeat until the player stops listening. That's not boring dialogue — that's no dialogue. Real mundane conversation has texture. Specific, local, tied to a person's actual day.

The vendor doesn't say "Nice day." They say "Gate traffic's been light — you're my first customer in twenty minutes." One line, three functions: specific to location, implies economic conditions, tells you this person notices foot traffic. Sounds like a person. Might become a clue later if the storyteller decides light traffic means something — but the vendor doesn't know that. They're just talking about their morning.

The key: unentangled NPCs have opinions about their own lives. Not about the plot. About rent, neighbours, food, shift rotation, the news, the new place on Level 3, their sore back. These opinions draw from the same axis system — a content NPC grumbles comfortably, a restless one complains with an edge. The difference isn't fewer axes. It's that their secret is mundane. The vendor's secret isn't smuggling — it's underpaying tax, or loneliness, or hating their landlord. Small truths that make a person feel real without connecting to anything.

Budget: more authored lines per ordinary role, lower stakes per line. 30-40 ambient lines per unentangled role versus 15-20 for a compromised one. Dramatic NPCs get tension for free — subtext does half the work. Ordinary NPCs lean on voice alone. Pure craft. The writing equivalent of making a plain room feel lived-in: no dramatic lighting, just the right chair in the right spot.

And the thing everyone should hear: unentangled voices are the silence that makes the signal audible. If every NPC sounds like they're hiding something, nobody sounds like they're hiding something. The vendor's genuine friendliness is the baseline that makes the dock supervisor's evasiveness noticeable. Paula's 30/50/20 split — flat, mundane-triangle, entangled — is exactly right from a writing standpoint. I need that population gradient to make each tier sound distinct and to make the entangled tier land. Protect the ratio or the investigation loop collapses into noise.

TYRE — Round 3 (Team Leader Feedback)

1. Disposal Model: Yes, This Is Better

cracks knuckles — let me be honest: the Team Leader's proposal is simpler than what I designed and produces the same result with fewer edge cases. My R1/R2 two-tier model (persistent vs. recyclable) required categorizing NPCs at generation time, maintaining disposal rule sets per category, and reference-count pinning to prevent cross-template triangle breakage. That's three systems solving one problem.

The proposed model: keep everything in active scopes running (neighborhood, active quest, colleagues), state-save everything else, evict by interaction-timestamp when sim-space pressure demands it. One eviction policy. One serialization path. No categorization rules.

Performance implications are neutral-to-positive:

• State-saved NPCs are my "Dormant" tier under a different name — serialized structs, ~1–2KB each. 5,000 state-saved NPCs = 10MB. Not a concern.
• Reactivation on return is deserialization + state-forward, ~2–5ms per NPC for schedule/relationship tick catch-up. Player travel time provides ample buffer.
• LRU eviction by interaction-timestamp is O(1) with a min-heap. Trivial to implement, trivial to tune — the threshold is just "how much sim-space remains?"
• The reference-count pinning I proposed in R2 becomes unnecessary. NPCs in active quest scopes or cross-template references naturally have recent interaction timestamps. The system self-organizes.

One addition: scope classification should be explicit, not implicit. "Active scope" needs a clear definition the code can query. Tag-based system: scope:neighborhood, scope:active-quest, scope:colleague, scope:known-contact. Any NPC with at least one active scope tag stays fully simulated. Zero tags = eligible for state-save eviction. Tags added/removed as gameplay progresses. Simple, queryable, debuggable.

Verdict: adopt this model. It's less clever than mine and that's a compliment.

2. Combat Stats: Component, Not Axis

Combat capability doesn't belong in the 10-axis NPC model. Axes define who someone is — personality, wants, secrets, relationships. Combat is what they can do in a specific situation. Different category.

In bevy_ecs terms, this is an optional CombatCapability component — same pattern as perception modes (D-017). The coffee vendor doesn't carry one. The ex-military cargo worker does.

CombatCapability {
    proficiency: Tier,          // untrained/trained/veteran/elite/enhanced/higher
    equipment_tags: Vec<Tag>,   // sidearm, body-armor, biononics-combat
    threat_level: u8,           // computed from proficiency + equipment, 1-10
    combat_style: Style,        // aggressive/defensive/evasive/support
}

The Skill Set supporting axis includes a combat-trained tag that triggers component attachment at generation time. Axis says "this NPC has military background" — component says "here's what that means mechanically." Axes drive narrative behavior, components drive simulation mechanics. Clean separation.

Scales with ascension: Higher-culture NPCs get additional components (BiononicWeaponry, EnhancedReflexes) that layer onto the base combat system without polluting the personality model. Difficulty tier: straightforward.

3. Vertical Slice: Smuggler + Detective

No new architecture beyond what I scoped for the two-character test in R2. But this pairing is more technically demanding than Nigel's generic duo — and that's why it's a better test. Two implications:

Access-tier inversion. The detective hits authority tier at institutional templates and outsider at criminal venues. The smuggler hits insider at criminal/commerce templates and suspect at institutional ones. The access-tier system needs at least four tiers functional from day one, not just insider/outsider. Same code, more configuration, better coverage of the full query space.

Criminal economy loop. The smuggler implies goods-flow — moving contraband through gate infrastructure. Doesn't require a full economy sim, but needs a thin cargo/goods system: items with provenance, routes with risk, NPCs who buy/sell. This adds ~30% to the smuggling-ring template's complexity but gives us a concrete economic substrate to test against. Worth it — validates that templates carry mechanical loops, not just narrative ones.

Content cost is the real delta. Mellanie needs voice kits for both institutional and criminal social registers. The architecture handles it fine; the authoring pipeline gets a genuine stress test. Exactly what a vertical slice should do.

4. Base/House Builder DLC: Design the Slot Now, Fill It Later

This is a client-heavy feature — notable because almost everything else is server-heavy. The builder needs placement UI, object snapping, preview rendering, and undo/redo, all in Godot. The server validates and persists.

Three architectural considerations:

Player-modified spaces serialize differently. Template-generated apartments are recreatable from template_id + seed. Player-customized apartments are unique — full object-by-object serialization. D-012's chunk-based architecture supports this: the player's home chunk stores a modifications overlay atop the base template.

Design the data model now, ship the UI later. Every housing entity carries modifications: Vec<Modification> from day one, where Modification = {position, object_id, rotation, variant} — ~32 bytes each. Base game ships with this field always empty. The builder DLC adds the Godot placement UI to populate it. Zero data model changes at DLC time. A heavily customized home: 200–500 modifications = 16KB max. Negligible save impact.

Social hook integration. Gore and Ozzie are right that this can't be cosmetic-only. Each placeable object carries optional social_hook tags the simulation reads — a guest room enables hosting NPCs, a home office changes your commute routine, a bar setup lets you invite people over. This connects building to the core life-sim loop. The hook tags are part of the object definition data, not code — so the DLC ships new objects with new hooks without engine changes.

Difficulty tier: data model and server integration are straightforward. Godot placement UI is moderate — well-trodden builder pattern. The interesting challenge is Gestalt's territory: making placement matter mechanically. Architecturally, we're ready whenever design pulls this trigger.

---

QATUX (Documenter)

Archival Summary

Round 17 produced seven confirmed decisions (D-023 through D-029), all recorded in DECISIONS.md with full rationale and team attribution.

Decisions Produced

• D-023: Three-tier content model with life-sim substrate (Tier 1: authored modules, Tier 2: templates, Tier 3: procedural filler)
• D-024: NPC generation model — 10 axes + combat component (7 essential axes, 3 supporting, triangle model as atomic social unit)
• D-025: Social site / functional cluster as atomic template unit (4-8 NPCs, 15-40 tiles, single ownership with reference links)
• D-026: Simulation tiers with timestamp-based eviction (Active/Background/State-saved/Ungenerated, LRU eviction)
• D-027: Vertical slice — smuggler + detective, two-character proof (supersedes D-006)
• D-028: Dialogue architecture — tagged line pools with four relational layers (access/history/trust/disclosure)
• D-029: Population entanglement ratio — 30/50/20 (flat/mundane/intrigue split, variable per seed)

Key Open Items Remaining

D-014 v0.1 scope requires revision following D-027. The vertical slice specification changes: NPC count (now ~20-30 for two social sites + bar + smuggling ring), map scope (station district not planetary Institute), and character loadout (smuggler logistics worker + institutional detective, not academic Guardian operative).

Faction selection for v0.1. D-027 implies criminal + institutional representation. Commission presence is natural (station law enforcement). Severance fits if the smuggling operation has anti-rejuvenation ideological ties, but this may be over-scoping. Team Leader and Miri to decide: pure criminal economy test, or criminal+ideological contamination test?

Base builder / house customization DLC flagged for future discussion. Tyre confirms architectural readiness (modifications overlay model, social hook system). Feature deferred but data model considerations noted for initial save format design.

Cultural generation guide needed from Miri before first Tier 2 template pack authoring begins. Mellanie's content packs require cultural voice variation (Higher vs baseline, regional vs cosmopolitan, institutional vs criminal registers). Without the generation guide, template voice kits risk cultural homogeneity. This is a blocking dependency for scaled content authoring.

Mellanie activated from standby. Dialogue architecture decisions (D-028) make content authoring the next phase-gate. Mellanie transitions from observer to active contributor.

Consistency Check

No conflicts detected with prior decisions. D-027 supersedes D-006 cleanly — both are vertical slice specifications, D-027 is narrower scope with higher mechanical proof density. D-024 (combat as component) aligns with D-017 (perception modes as components). D-026 (simulation tiers) integrates with D-010 (client-server separation) and D-011 (fog of perception). D-028 (dialogue layers) depends on D-024 (NPC axes) for systemic selection inputs.

The three-tier model (D-023) is architecturally compatible with all engine and map decisions. Pool-based Tier 1 modules align with the storyteller activation model from D-005. Template-based Tier 2 content fits D-025's functional cluster spatial model. Procedural Tier 3 filler uses the same generation hooks as D-014's borderless world expansion.

Team Workflow Note

Round 17 followed the D-022 workflow correctly: discussion occurred in dedicated round file, decisions extracted to DECISIONS.md post-closure, discussions README updated, agent briefings flagged for update. Qatux participated throughout all three sub-rounds. Archive complete.

For the record: This was a high-yield workshop. Seven decisions in three rounds is efficient consensus-building. The team navigated from abstract content philosophy to concrete implementation specifications without significant dissent. Gore's contentment axis, Paula's triangle model, and Tyre's timestamp-based eviction were integration points where individual proposals strengthened the collective architecture.

Next phase dependencies are clear: Miri's cultural guide, revised D-014 scope, v0.1 faction selection, and Mellanie's first template content pack. The vertical slice specification (D-027) is now the project's loadstar.