jpmschweitzerandClaude Opus 4.6
9a5c9c4408
Standardized YAML frontmatter on all 38 files with title, description, type, workshop, agent, and round fields. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
33 KiB
title, description, type, status, workshop, agent, round, created
| title | description | type | status | workshop | agent | round | created |
|---|---|---|---|---|---|---|---|
| Nigel Round 4 — MobileChunk and Soft Re-generation | MobileChunk replayability, one NPC five lenses, soft re-generation coherence, and D-record sign-off | workshop | archived | generator-architecture | nigel | 4 | 2026-02-27 |
Generator Architecture Workshop — Round 4: Nigel (Replayability & Procedural Generation)
Date: 2026-02-27 Role: Replayability Advocate Task: OQ-R4-A (MobileChunk replayability), OQ-R4-E (One NPC / Five Lenses), OQ-R4-F (soft re-generation coherence), final sign-off on 12 D-ready items.
OQ-R4-A: Vessel Architecture — Accepting MobileChunk, Specifying Replayability Requirements
Accepting the Architecture; Moving Forward
The lead has settled vessel architecture as entity-carried MobileChunk. My instanced district model is retired. I want to briefly record why I don't fight this: the persistent world-entity model offers one replayability property my model couldn't — vessels have a history. A ship that has docked at fifteen ports, been used to smuggle contraband twice, and hosted an assassination attempt on voyage seven is a RICHER object than a vessel that only exists during voyages. The entity-carried model enables vessels to accumulate simulation state across time. That is, on reflection, better for the game I want to make.
What follows is my full specification of what the MobileChunk model needs from a replayability perspective.
The Stage and the Cast
The core conceptual frame for vessel replayability: the vessel is a stage; the manifest is the cast.
The stage (ChunkData interior) stays the same across voyages. The same ship has the same corridors, the same cabins, the same social spaces. This is correct and good — familiarity with the ship is earned knowledge that players can exploit on subsequent voyages. A player who has ridden the Tide-loop cargo hauler before knows where the service access is, knows which cabin is near the galley, knows the captain's usual table. That knowledge is capital they've built. The replayability isn't in rediscovering the layout; it's in who's aboard and what they know.
The cast (passenger manifest) changes every voyage. The same ship, the same route, but entirely different dramatic potential.
Manifest Seeding Strategy
Every voyage produces a new passenger manifest. The seed formula:
voyage_manifest_seed = derive_seed(master_seed, "vessel_manifest", vessel_entity_id, voyage_index)
voyage_index increments on every departure, not on every calendar day. A ship that makes three voyages on the same in-game day has three distinct manifests.
Manifest composition:
The manifest has two zones:
- Fixed slots — crew. Always present, same personnel per voyage. The cook is the cook. The first mate is the first mate. These NPCs are generated from the vessel's generation seed (not the voyage seed) and persist across all voyages. They develop relationships with repeat passengers over time.
- Variable slots — passengers. Drawn fresh per voyage from the eligible NPC pool at the departure location at the time of departure.
Passenger eligibility criteria (evaluated at journey-start):
- NPC is present at the departure location at departure time
- NPC has a plausible purpose for this route (home port on the other end, active relationship at the destination, commercial reason, assigned by faction, fleeing a situation)
- NPC satisfies berth class (if the ship has class-stratified cabins, as Miri's
BoundedLinearmodel specifies) - NPC is not in a simulation state that prevents travel (hospitalized, under house arrest, actively mid-scene)
From the eligible pool, passengers are selected by voyage_manifest_seed. This means:
Voyage A and Voyage B of the same ship, same route, drawn from overlapping NPC pools — different manifests. The eligible pool at departure time changes between voyages because the simulation has run. NPCs complete trips, return home, get tied up elsewhere. The manifest isn't just a random draw from a static pool; it's a draw from whatever pool actually exists at departure, seeded deterministically.
What Makes Voyage 5 Different from Voyage 1?
Five independent sources of variation across voyages of the same ship:
1. Manifest composition The passengers change every voyage. The entanglement configuration among those passengers — who's connected to whom, who's being watched, who's carrying knowledge the player wants — is seeded from the voyage manifest seed. Voyage 1: a smuggling ring operative sharing the ship with their handler, both unaware the player is investigating the same ring. Voyage 5: entirely different cast, different dramatic potential.
2. NPC knowledge state drift Even repeat passengers (NPCs who've traveled this route before) have different knowledge states on voyage 5 than voyage 1. The simulation has run. A NPC who on voyage 1 was unaware of a conspiracy is on voyage 5 the primary witness to it. Same NPC, same 10-axis generation, different information inventory because the world has changed.
3. Storyteller state DramaDensity can vary per voyage. A route that was Low-density when the player first traveled it becomes a Flashpoint voyage when the storyteller has activated drama modules that involve NPCs aboard this ship.
4. In-transit events The storyteller can seed in-transit events (delays, emergencies, unexpected dockings) from a per-voyage event seed. Voyage 3: smooth crossing. Voyage 5: unexpected stop at an unscheduled port while the Commission investigates a distress signal. The player who knows this route intimately has never encountered THIS version of it.
5. Crew relationship state The crew accumulates relationship state across voyages. The cook who was neutral toward the player on voyage 1 has warmed (or soured) by voyage 5 depending on simulation events. The crew provides persistent social continuity that makes repeat voyages feel like returning to a place with memory, not resetting to a blank state.
The Temporal Constraint Mechanism
The arrival deadline is the primary dramatic engine aboard a vessel. The player can see it: their neural insert displays arrival time. This creates guaranteed temporal pressure without the generator engineering it.
For replayability: the temporal constraint means the player cannot do everything every voyage. On a six-hour crossing, there are perhaps four meaningful conversations possible, three explorations of the ship's spaces, and one incident. The player must choose. Different voyages, different choices, different outcomes — even with the same cast.
The storyteller can modify arrival time (delay, emergency, diversion). A voyage that was supposed to be six hours becomes nine. Three additional hours of enforced proximity. The NPC who had almost opened up, who was one more meal away from revealing what they know, now gets that meal. This is the highest-leverage storyteller tool aboard a vessel: not changing who's present, but changing how long they're all stuck together.
Replayability Requirements for MobileChunk (For the D-Record)
Formalizing as verifiable requirements:
R-V-1: Voyage manifest seeded per departure, not per vessel. The same ship on different voyages must have meaningfully different passenger lists. The guarantee: at least N passengers must differ between adjacent voyages (N = floor(variable_slots × 0.5) — at least half the variable slots turn over).
R-V-2: Crew is persistent, passengers are variable. Crew NPCs persist across all voyages of the same vessel. This creates historical continuity. Passenger slots are refilled from the eligible pool at each departure.
R-V-3: In-transit events are voyage-seeded, not vessel-seeded. The same ship should not always produce the same incidents. Each voyage gets its own event seed, derived from the voyage manifest seed.
R-V-4: Arrival time is storyteller-modifiable. The temporal constraint is a storyteller instrument. The deadline can be extended (delay, diversion) or shortened (emergency early docking) based on narrative needs.
R-V-5: Vessel interior does NOT re-generate per voyage. The interior ChunkData is fixed at vessel generation time. Players who learn the ship's layout have earned that knowledge. The replayability is in the cast and the events, not in rediscovering the stage.
R-V-6: Vessel carries ChunkMutations for accumulated damage. A vessel that has been boarded, damaged, or modified during a voyage carries those modifications forward. The hole the player blasted through the bulkhead on voyage 3 is still there on voyage 5 unless repaired. Vessels accumulate history.
OQ-R4-E: "One NPC, Five Lenses" — Replayability Check
The Question
Miri demonstrates that one sufficiently complex NPC can provide all five playstyle entry hooks simultaneously (investigation anchor, tycoon economic chokepoint, dating sim social presence, political drama nexus, assassination latent hook). The question: does this create a district that plays the same every time?
Answering the Direct Question
Within a single seed: YES, the same complex NPC provides the same hooks every visit. The investigation hook is always the same anomaly. The economic chokepoint they control is always the same resource. The same seed produces the same NPC with the same 10-axis profile. The district will feel "solved" once the player understands the NPC.
But this is the correct behavior. Same seed = same world. The point of a single complex NPC is not variation within a seed — it's that this is a place where ONE person MATTERS. In a small community, that's realism: there is a person here whose story is the story of this place. Exhausting that story in one thorough visit is accurate, not a design failure.
Across seeds: FULLY VARIABLE. Different seed = different NPC. Different anomaly. Different economic stranglehold. Different romantic archetype. Different political position. Different reason to be in a quiet backwater. The one-NPC minimal district provides maximum seed-to-seed variation because the entire dramatic content of the world is packed into one NPC, and that NPC is generated fresh per seed.
But One NPC Is Not Enough for Intra-Seed Variation
The instinct to require 2-3 NPCs for minimal districts is correct, but the REASON is not replayability across seeds. The reason is emergent drama requires relationships.
Gestalt's minimum functional triangle is three nodes. One NPC cannot be in a triangle alone. Without relationships between NPCs, there are no triangles. Without triangles, the drama is a monologue — one person's story, fully contained within themselves. A monologue is finite. Once you've heard it, it's heard.
Two NPCs in tension produce a story. Three NPCs produce emergent behavior that none of them individually contains. The triangle is where unpredictable outcomes emerge from predictable NPC behaviors.
The Minimum for Meaningful Replayability Within a Seed
For a minimal-complexity insignificant district:
- 1 NPC: A character study. High seed-to-seed variation. Zero intra-seed variation after the first thorough visit. Not replayable.
- 2 NPCs: A relationship. Intra-seed variation from relationship dynamics (trust building/breaking, information transfer between them over time). Limited emergent behavior.
- 3 NPCs: A triangle. Intra-seed variation from triangle dynamics, shifting alliances, cascade effects when one node changes state. This is the minimum for genuine emergent narrative.
My recommendation: The minimum NPC count for a district to be replayable within a seed is 3 — one functional triangle. For minimal-complexity districts, those 3 NPCs can each be simpler than a Full-complexity NPC (less information inventory, fewer entanglements, narrower routine), but there must be 3.
Does the 10-Axis Model Support One NPC Providing All Five Hooks?
Yes, completely. The 10-axis model (Want, Secret/vulnerability, Relationships, Tolerance threshold, Daily routine, Information inventory, Contentment + 3 supporting) can absolutely encode an NPC who:
- Has a Want that's an investigation hook (they want something that implicates them in something)
- Has a Secret that's an assassination latent hook (they're someone significant in hiding)
- Has Relationships that create economic chokepoints (they're the only one with the off-world comm codes)
- Has a Tolerance threshold that creates political drama (they're near the edge; push them and the community fractures)
- Has a Routine that creates dating sim opportunity (they gather with others every evening)
But having all five hooks in one NPC compresses the drama to a point. It works for the initial playstyle discovery (the world IS this person), but it creates a district that has no further depth once the NPC is understood.
Conclusion on OQ-R4-E: One NPC CAN provide all five hooks. The 10-axis model supports this. Across seeds, this produces maximum variation. Within a seed, this produces minimum emergent behavior. The answer to whether it's enough: no — the minimum for intra-seed replayability is 3 NPCs (one functional triangle), even for minimal-complexity districts. Miri's five minimum content types should be distributed across a minimum triangle, not collapsed into a single NPC.
OQ-R4-F: Soft Re-Generation Coherence
The XOR Problem
Gestalt proposes original_seed XOR event_seed for large-scale post-event re-generation. The question: does XOR-seeded regeneration produce caused variation or random variation?
The answer is: it produces deterministic but incoherent variation. Here's why this fails.
XOR combines two bit patterns arithmetically. The result has no semantic relationship to either input. original_seed XOR event_seed will always produce the same output (deterministic — good), but that output has zero structural relationship to the original seed (incoherent — bad). The content generated from the XOR seed would be as likely to produce an upscale market quarter as a burned-out ruin, because the XOR seed has lost all information about what the original district was.
This violates Ozzie's principle ("destruction must be CAUSED") and Miri's cultural response model (aftermath must be an intensification of existing character, not a transformation of it). A Frost-heritage community that has experienced a violence event should look MORE Frost (tighter, colder, doors closed, nobody talks to strangers) — not like a random re-roll that might be more Tide than the original.
The Replayability Dimension of Re-Generation
Before proposing an alternative, I need to address the replayability question directly: does any form of re-generation produce different-enough results across playthroughs of the same seed?
The answer depends on what varies. If the post-event state is deterministic from seed + event type + event location + pre-event state, then two playthroughs of the same seed that both experienced the same event in the same location would produce identical post-event states. That's correct — same seed = same world. The variation across playthroughs comes from whether the event happens and when, not from random variation in the aftermath.
What the storyteller controls: whether and when to activate a fragility, trigger a trauma event, or cause structural damage. Different playthroughs of the same seed can have different event histories, producing different delta layers on the same base world. That's the replayability engine — not random reseeding of chunks.
The Alternative: Typed Modification Instead of Re-Generation
No event actually requires chunk re-generation. Every event type maps to a ChunkMutations overlay:
Fire/explosion:
- Tile overrides: burned floor tiles, collapsed wall tiles, ash-layer objects
- Structural changes: specific walls marked as destroyed
- Object removal: combustible objects replaced with debris objects
- NPC modifications: displacement of NPC home points (their space no longer exists)
Building collapse:
- MultiBlockReservation modified: vertical extent reduced
- Floor zone zone_type changed from active to Ruins for affected z-levels
- Structural changes at block level: entry points sealed
Economic catastrophe:
- NPC contentment/want axes modified for residents
- Object states changed (shops closed, market stalls empty)
- No physical world changes — the buildings are still there, the market stalls are still there, but they're empty and the NPCS know it
Political upheaval:
- Triangle activation changes (storyteller fires suppressed triangles)
- NPC relationship modifications
- Access tier changes (previously Semi-Private zones lock down to Restricted)
- No physical world changes
In NONE of these cases is chunk re-generation needed. The ChunkMutations overlay handles all of them. Tyre's existing architecture is sufficient.
The one case that might seem to require re-generation: district-scale catastrophe (full district destroyed). My answer: this isn't re-generation — it's a new setting type. A district that has been catastrophically destroyed becomes a Ruins district. The ruins are generated from the SAME master seed, using the original district generation, with a damage overlay applied at Phase 2 that marks everything as destroyed. The ruins look like THAT district's ruins, not randomly generated rubble — because the tiles underneath are the same tiles, just tagged as destroyed.
Formal Position on OQ-R4-F
XOR-seeded re-generation is the wrong tool. It produces deterministic but semantically incoherent results. It cannot satisfy the "caused, not random" requirement.
The correct approach:
- Small-scale events →
ChunkMutationswith event-typed tile overrides and structural changes - Medium-scale events →
ChunkMutationswith broader structural changes + NPC state modifications - Large-scale catastrophe → Original ChunkData + comprehensive damage overlay at Phase 2; setting type changes to Ruins; no re-seeding
Replayability outcome: The modification layer varies per playthrough (different events occur at different times, or don't occur at all). The base world never changes (same seed = same underlying district). Two playthroughs of the same seed can have completely different physical worlds in a district if one playthrough triggered the gas explosion and the other didn't. That is the replayability engine. The variance is in EVENT HISTORY, not in random chunk re-generation.
What I'd put in the D-record: "Post-event world modification is handled exclusively through the mutation overlay system (ChunkMutations / WorldStateDelta). The generator never re-runs for a generated district. Re-seeding via XOR is explicitly rejected as producing semantically incoherent results."
Final Replayability Sign-Off: 12 D-Ready Items
For each item, I apply the Comparison Test: would two instances of this mechanism (same template, different seeds) produce distinguishable player experiences? And: would the same instance across two playthroughs of the same seed produce meaningfully different play?
D-READY-1: DistrictLayoutMode — Grid and Organic Support
PASS — SIGNED OFF.
Comparison Test: Grid district vs. Organic district of the same type → structurally different player experience (spatial reconnaissance is genuinely different; the assassin who uses a mental grid template is wrong in Organic mode). Same seed always produces the same layout mode — correct. The variation is across world regions, not within a seed.
One replayability note for the D-record: the distribution of Grid vs Organic districts must itself be seeded (different seeds produce different proportions of Grid/Organic across their worlds). If every generated world has Grid at the center and Organic at the margins, that's a predictable pattern a player can exploit. The distribution proportion should vary per seed.
D-READY-2: Guarantee Tier System — Universal / Full-Only / Conditional
PASS — SIGNED OFF.
Comparison Test: Two Full-complexity districts (different seeds) → both satisfy Tier 2 guarantees, but the SPATIAL REALIZATION differs per seed (which specific chunk is the Traffic Chokepoint, where the Encounter Corridor runs, what the Elevated Vantage position overlooks). The guarantee doesn't determine configuration, only presence. This is the right design — minimum guaranteed content without constraining configuration to a template.
One caveat: the guarantee system must not anchor archetypes to fixed positions within the district footprint. If the guarantee states "must have an Elevated Vantage" but the generator always places it northeast of the Social Hub, experienced players will use that pattern. The implementation must verify that archetype spatial positions vary in angular distribution across seeds. This is my earlier hard requirement, still standing.
D-READY-3: TrianglePurpose Enum
PASS — SIGNED OFF (with a note).
Comparison Test: Triangle purpose tags don't produce variation — they produce RELEVANCE. Two instances of the same triangle model with different purpose tags activate under different player playstyle conditions. This is not a variation mechanism; it is a targeting mechanism that prevents the storyteller from surfacing wrong-playstyle drama at the wrong moment.
For the D-record: TrianglePurpose is not a replayability feature — it is a multi-playstyle accessibility feature. Its replayability contribution is indirect: by activating the right triangles for the player's current lens, it ensures that drama which exists in the world is surfaced to the player who can engage with it, rather than being invisible noise.
D-READY-4: WallBackside / TileBehindState
PASS — SIGNED OFF (with a condition).
Comparison Test: Same district type, different seeds → different wall backside configurations. The proportion of HiddenRoom vs. ServiceVoid vs. StructuralFill must vary per seed, not be fixed by template. If every commissary wall always has a ServiceVoid on the other side, exploration is template-matching. The specific backside assignment must be seeded.
Condition for the D-record: Backside assignments within a template must have seed-driven variation in their specific distribution. The template can constrain TYPES (this district type can have HiddenRooms; this template slot is always StructuralFill) but the specific assignment per wall tile should vary. 90% automated tagging (Tyre's number) should mean 90% from seeded probabilistic rules, not 90% from fixed template values.
D-READY-5: Dynamic Modification via Overlay (Not Re-Generation)
PASS — SIGNED OFF.
This is the most important replayability mechanism in the architecture. The overlay model makes the same-seed-different-playthroughs scenario possible. The base world is identical across all playthroughs of the same seed. The modification history diverges based on what events the simulation has produced. Two players who started the same seed, made different decisions, and triggered different events have genuinely different physical worlds after those events — while sharing the same generator baseline.
This is EXACTLY how replayability should work: same world, different history.
D-READY-6: ZonePalette Modifier System
PASS — SIGNED OFF.
Comparison Test: Industrial farmland vs. rustic farmland → visually distinguishable. Frost-heritage industrial farmland vs. Tide-heritage industrial farmland → also distinguishable. The modifier system produces substantial combinatorial variety from a small base set.
Replayability note: palette combinations are static per district per seed. The same seed always produces the same palette. The variation is across seeds and across district types — not within a playthrough or across playthroughs of the same seed. This is correct. Visual identity of a place should be stable.
One note I want in the D-record: palette modifiers should influence NPC appearance as well as environment appearance. A Frost-heritage district should have NPCs whose clothing/gear palette is consistent with the Frost material grammar. This extends the "caused not random" principle to NPC appearance — people dress like they're from here.
D-READY-7: Horizon View Corridor as Coastal Guarantee
PASS — SIGNED OFF.
Comparison Test: Two coastal districts (different seeds) → both have horizon view corridors. The corridors are in different locations, overlook different portions of water, have different surrounding context. The MOMENT is guaranteed; the specific experience is seeded.
From a replayability standpoint: the horizon view is one of Ozzie's primary Wow Moments. Its value is partly in its unexpectedness — the player turns a corner and sees the ocean. If the horizon view corridor is always in the same relative position to the district entry point, experienced players expect it and the Wow diminishes. The reservation should constrain the corridor's existence and minimum width, but not its position. Let the generator place it wherever the spatial configuration produces it, as long as it exists.
D-READY-8: Assassin Lens Spatial Guarantees (A-1 through A-4)
PASS — SIGNED OFF (with the angular variation requirement).
Comparison Test: Two Full-complexity districts (different seeds) → both satisfy A-1 through A-4. The Elevated Vantage is in a different position. The Egress Multiplicity routes run different directions. The Temporal Opacity Window is at a different day-phase.
My standing hard requirement: archetype placement must vary in angular position across seeds, not just in distance from center. This requirement applies directly to the Elevated Vantage (A-1) and the relationship between it and the Traffic Chokepoint. If the Elevated Vantage is always north of the Traffic Chokepoint, every assassination approach is the same elevation/angle relationship regardless of seed. The guarantee system must verify that the angular distribution of archetype positions across multiple seeds is not clustered.
This is a verification requirement, not just a generation requirement. The guarantee audit should fail if archetypes are generated in positions that form a predictable template.
D-READY-9: Heritage Grammar Overlay for Non-Urban Palettes
PASS — SIGNED OFF.
Comparison Test: Frost-heritage farmland vs. Tide-heritage farmland → substantially different organizational grammar (Frost: individual plots, fenced separations, minimal communal space; Tide: open gradients, communal gathering areas, fluid spatial boundaries). This is not just cosmetic — the spatial grammar affects which playstyle affordances are naturally present (Tide produces more obvious Social Hub expressions; Frost produces more physical_distance informal zone expressions).
For the D-record: Heritage grammar is a generative input, not a decorative overlay. It shapes which archetypes are easy to satisfy and which are difficult. A Frost-heritage district has natural physical_distance informal zones but needs deliberate effort to produce an Encounter Corridor. This constraint shapes the district's playstyle affinity — which is the right level of influence.
D-READY-10: Non-Urban Informal Zone Typology
PASS — SIGNED OFF.
Comparison Test: Frost-heritage wilderness district → physical_distance informal zone type. Tide-heritage maritime district → social_permission informal zone type. The type is determined by heritage root — consistent and predictable. The LOCATION within the terrain is seeded.
One replayability note: the three informal zone types (social_permission / physical_distance / utilitarian_cover) create meaningfully different gameplay even when the player knows which type they're in. Social_permission means cover is about convention, not geography — you can be seen, you just can't be judged. Physical_distance means cover requires travel — you have to physically remove yourself. Utilitarian_cover means you need a functional excuse for your presence. Each type demands different strategies. The variation is not in "what is the informal zone" but in "how does one USE it." That's deep replayability from a simple typology.
D-READY-11: Vertical Scale Architecture
PASS — with a replayability condition for the D-record.
Comparison Test: Two Full-complexity skyscrapers (same heritage, different seeds) → different floor zone assignments? If z-band assignments are purely deterministic by building function (corporate building always has labor on 1-5, operations on 6-20, executive on 21-30), then experienced players can predict what's on floor 30 without visiting. That's the Second Station Syndrome applied vertically.
Condition for the D-record: z-band boundaries must have seed-variation within cultural constraints. The cultural model constrains the ORDERING (labor below operations below executive), but not the exact floor numbers. A corporate building in one seed has executive starting on floor 22; in another seed, floor 35. The player who knows "executive floors are in the upper third" is working with useful knowledge, but can't skip exploration — they still need to find the actual executive zone.
Secondary replayability property of vertical scale: vertical access routes are playthrough-history dependent. A player who befriended the building's head of facilities on an earlier encounter can now access the service elevator directly. A player who damaged the main elevator bank in a previous event now has to find an alternative route. The building's access topology is fixed by the seed; which routes are available to the player at any moment is determined by their relationship and event history. Same building, different access experience per playthrough.
D-READY-12: Trauma Events as EraModification Subtypes
PASS — SIGNED OFF.
Comparison Test: Same district before and after a PhysicalDestruction trauma event → different NPC pattern weight distribution (ANCHOR/WITNESS/REMNANT increase; normal distribution suppressed). The change is predictable from the heritage root — Frost communities respond differently to trauma than Tide communities. Two Frost-heritage communities that experience the same trauma type respond similarly (same cultural grammar). Two communities with different heritage roots diverge.
This is the correct behavior. The trauma response is not random — it is culturally CAUSED. Ozzie's principle is satisfied.
One replayability note: the decay rate (how quickly the cultural aftermath fades toward baseline) should be seeded at generation time with variation. A Frost-heritage community might always be slow to recover (heritage root determines the baseline rate), but the SPECIFIC rate for this community is seeded (some Frost communities are 20% faster to recover than the mean; others are 20% slower). This prevents trauma response from being perfectly predictable from heritage root alone — it adds the "this specific community" dimension that makes individual settlements feel distinct.
Summary: Round 4 Positions
OQ-R4-A resolved: MobileChunk accepted. The vessel is a stage (persistent interior) + a cast (per-voyage manifest). Manifest seeded from derive_seed(master, "vessel_manifest", vessel_id, voyage_index). At least 50% of variable passenger slots must turn over between adjacent voyages. Crew persistent across voyages. In-transit events voyage-seeded (not vessel-seeded). Arrival time is storyteller-modifiable. Vessel accumulates ChunkMutations across its lifespan. Requirements R-V-1 through R-V-6 stated for the D-record.
OQ-R4-E resolved: One NPC can provide all five playstyle hooks simultaneously. 10-axis model supports this fully. BUT: one NPC produces zero intra-seed emergent behavior (no relationships = no triangles = no emergence). The minimum for meaningful intra-seed replayability is 3 NPCs (one functional triangle), even for minimal-complexity insignificant districts. Miri's five minimum content types should be distributed across a minimum triangle.
OQ-R4-F resolved: XOR-seeded re-generation is explicitly rejected — it produces deterministic but semantically incoherent results. The correct approach: typed ChunkMutations overlays for all event scales. The generator never re-runs for a generated district. District-scale catastrophe produces a Ruins overlay on the original ChunkData, not a re-seeded replacement. Replayability comes from event history divergence across playthroughs, not from random chunk re-generation.
12 D-ready items: All signed off. Three replayability conditions to include in D-records:
- (D-READY-1) Grid/Organic distribution proportion must vary per seed — not fixed to a predictable geographic pattern.
- (D-READY-8 / D-READY-2) Archetype placement must vary in angular position per seed — not just distance from center. This is verifiable and testable. The guarantee audit should fail if archetypes cluster in predictable angular positions across a test run of N seeds.
- (D-READY-11) Z-band floor boundaries must have seed-variation within cultural ordering constraints — experienced players should know "executive is in the upper zone" without knowing which exact floor that begins on.