Files

T

jpmschweitzerandClaude Opus 4.6 9a5c9c4408 docs(docs): add frontmatter to generator-architecture workshop

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>

2026-03-13 23:40:37 +01:00

22 KiB

Raw Permalink Blame History

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

title	description	type	status	workshop	agent	round	created
Round 1 Notes	Qatux compilation of all round 1 participant outputs and emerging themes	workshop	archived	generator-architecture		1	2026-02-27

Generator Architecture Workshop — Round 1 Notes

Compiled by: Qatux (Documenter) Date: 2026-02-27 Source files:

docs/workshops/generator-architecture/gestalt-round1.md
docs/workshops/generator-architecture/tyre-round1.md
docs/workshops/generator-architecture/miri-round1.md
docs/workshops/generator-architecture/araminta-round1.md
docs/workshops/generator-architecture/nigel-round1.md
docs/workshops/generator-architecture/ozzie-round1.md

Overview

Round 1 produced six independent domain assessments of the generator architecture. Despite writing without cross-referencing each other, participants arrived at strong consensus on the pipeline model and the role of authored templates. The primary open tensions are in pipeline stage ordering, the depth of variation the quarter system will actually deliver, and the question of whether cultural variation translates mechanically (or remains decorative).

Section 1: Consensus Points

C-1: The Pipeline Model

All participants implicitly or explicitly accepted the Cities Skylines top-down pipeline as the working model:

Geography
  → Infrastructure (transport, utilities, Meridian coverage)
    → Amenities & Services (social site type selection)
      → Population (NPC density, entanglement seeding)
        → Zoning (access tier assignment per zone)
          → Block Generation (chunk merge strategy, multi-block reservations)
            → Chunk Fill (D-025 template instantiation, sub-chunk quarters, NPC placement)

Sources: Gestalt §3, Miri §5, Nigel §2, Tyre §5.3 pipeline table.

Each participant mapped their domain to this pipeline and found it accommodated their requirements without modification. No participant proposed a different top-level structure.

C-2: Spatial Hierarchy (Confirming D-094)

All technical discussion respected the four-level hierarchy without question:

Level	Sim tiles	Visual tiles	Physical	Role
Chunk	64×64	32×32	32m	Streaming + serialization unit
Block	128×128	64×64	64m	Generator planning unit (4 chunks)
District	512×512	256×256	256m	Template composition unit (16 blocks)
Quarter	32×32 sim	16×16 visual	16m	Fill-time constraint only — not a hierarchy level

The quarter clarification is important: Tyre stated explicitly that quarters are a generation-side layout concept, invisible to the runtime after fill. The server sees tiles; quarters exist only during the generation pass. Araminta and Nigel both accepted this framing.

Sources: Tyre §1.4, §4.4; Araminta §2.1; Nigel §3.

C-3: "Templates Are Authored; Placement Is Generated"

Stated independently by three participants in nearly identical terms:

Tyre: "D-025 templates are authored. Skeleton placement is generated. The generator arranges templates, not tiles."
Miri: "D-025 templates are bricks. The district skeleton is the architectural plan. The generator writes architectural plans from ingredients; human authors craft the bricks. The generator never touches the bricks themselves."
Gestalt: "The generator places the slots; templates fill them."

This principle governs the generator's relationship to the entire D-025 social site library. It is the mechanism that preserves hand-authored content quality while enabling procedural arrangement.

Sources: Tyre §2.3, Miri §5 (Q-036 reconciliation), Gestalt §3.

C-4: District Skeleton as Atomic Generator Output (Q-036 direction)

Strong convergence: the district skeleton is the generator's primary compositional output. Tyre proposed a concrete Rust data structure (DistrictSkeleton). Miri endorsed the skeleton concept with a complementary description of its contents (society profile → social site slots + spatial positions + access topology + NPC capacity + triangle assignments). Gestalt mapped each of its seven spatial guarantees to specific fields of the skeleton.

No participant proposed an alternative atomic unit.

Sources: Tyre §2, Miri §5 (Q-036 reconciliation), Gestalt §2 and summary table.

C-5: Grey Economy as Negative Space in the Generator

Three participants independently required the generator to model what official zoning does NOT account for:

Gestalt (Guarantee 2): Every district must have at least one zone with meridian_coverage: degraded or minimal — explicitly generated, not incidental.
Miri: "The grey economy occupies the spaces that official zoning doesn't account for. The generator must model what's NOT in the official map — which corridors are maintenance-only, which zones have dead spots, which blocks have unofficial access routes. This is not flavor; it's where the investigation happens."
Ozzie: "A place I'm not supposed to be" is a non-negotiable player experience requirement. Spaces that "feel like I wasn't meant to find it."

Araminta adds a visual corollary: maintenance/service zones must always have at least one empty quarter per block of service access type, and the dark/sparse visual treatment that distinguishes them.

Sources: Gestalt §2 Guarantee 2, Miri §5 point 4, Ozzie "What I Need" §2, Araminta §3.2.

C-6: Cultural Variation Must Produce Mechanical Variation

All six participants required that the society profile parameters (Q-032's six categories) translate into gameplay outcomes, not just aesthetics:

Miri: "If cultural variation doesn't translate to mechanical variation... then 300 distinct cultural profiles produce only the illusion of variety."
Gestalt: Cultural variation produces "different investigation difficulty, specific NPC behavior patterns, and specific contraband moral texture."
Nigel: The cultural layer "has the highest variety payoff per authored ingredient" and trust-building timelines, access tier thresholds, and NPC pattern distributions must all vary by culture.
Ozzie: "If the generator is just varying layout geometry and art themes, the 50th station is just the 50th palette swap. If it's varying the SOCIAL AND POLITICAL TEXTURE — who's in charge, what they want hidden, what grudges are still warm — then I will happily explore the 100th."

The mechanism: privacy_level and trust.building_rate are the key translation parameters (Miri). Cultural parameters feed NPC behavior systems; they cannot be a pure naming/art variable.

Sources: Miri §1 and §4, Gestalt §3 amenities stage, Nigel §2 Stage 3, Ozzie "What Makes the 50th Station Exciting."

C-7: Transit District as Generator Validation Fixture

Two participants independently proposed expressing the v0.1 Transit District (D-093) as generator output to validate the schema:

Tyre §5.4: "Express the v0.1 Transit District as a hand-authored DistrictSkeleton + hand-authored ChunkData for each of its 64 chunks." This validates the schema, creates a test fixture, and validates the content pipeline.
Gestalt §5 (Q-C): Proposed the "minimum viable district" — 1 workplace, 1 bar, 1 maintenance spine, 1 transit node, 1 restricted zone, 2 active triangles — as the minimum that satisfies all seven spatial guarantees.

These are complementary: Tyre's is a schema validation exercise; Gestalt's is a gameplay completeness check. Together they form a complete validation plan for the generator's district output format.

Sources: Tyre §5.4, Gestalt §5 Q-C.

Section 2: Open Questions (for Round 2)

OQ-1: Pipeline Stage Ordering — Population Before or After Zoning?

The tension: Gestalt says NPC secrets must have plausible staging grounds before NPCs can be validly generated, creating a feedback loop between Population and Zoning. Miri places template instantiation at the district skeleton stage (after Zoning, before Block Generation). Tyre's pipeline table places NPC Population as the last stage (v0.3+), after Chunk Fill.

These three descriptions are not fully compatible. The ordering of Population relative to Zoning and Block Generation is unresolved.

Gestalt's position: Population follows zoning for assignment, but population requirements constrain zoning (feedback loop required). Miri's position: Triangle assignments happen at the district skeleton stage, after zoning but before block generation. Tyre's table: NPC Population is last — a v0.3+ implementation concern.

For Round 2: Tyre needs to address whether the feedback loop Gestalt describes creates implementation complexity. Can the skeleton stage partially resolve NPC type requirements (without full NPC generation) to guarantee spatial staging grounds exist?

Raised by: Gestalt Q-A; Miri §5 "when D-025 templates get instantiated"; Tyre §5.3 pipeline table.

OQ-2: Seed Architecture — Single Master Seed or Per-Stage Seeds?

Nigel asks explicitly: "Is it a single master seed that derives all sub-seeds deterministically, or does each stage have its own seed parameter? I need to know whether 'same seed, different character selection' produces the same world with different lenses, or genuinely different worlds."

Tyre mentions determinism (D-010) but does not address how the seed propagates through stages.

Stakes: If same seed + different character = same world with different lenses, then character choice is a filter on the same information space. If same seed + different character = different worlds, the seed architecture is more complex and the reproducibility requirement (Q-030) more demanding.

For Round 2: Tyre to specify seed propagation architecture.

Raised by: Nigel §1 Guarantee 3, §7; Tyre §5.3 (mentions determinism, not seed propagation).

Ozzie raises this directly: "Does the choice of what fills a quarter have downstream consequences — does a garden quarter mean something different about who lives nearby versus a shack quarter? If the quarters are just aesthetic choices, I'll see through them on the second station."

Nigel endorses the quarter system as a replayability engine (§3) but focuses on physical geometry variation (chokepoints, routes). Ozzie wants to know if the fill type has social meaning.

The question for Round 2: Does the flavor structure assignment in unclaimed quarters (Nigel §4) feed back into NPC generation? A "market stall" quarter should presumably attract different NPC types than a "Commission kiosk" quarter. Is that dependency in the generator's data flow?

Raised by: Ozzie "Quarter System" section; Nigel §4.

OQ-4: Historical Palimpsest — Can the Generator Produce Layered History?

Ozzie's most demanding requirement: "Can it make stations that feel like they were built in layers by different people with different plans?" She distinguishes CAUSED irregularities (this building is L-shaped because something made it this way) from RANDOM irregularities (the RNG said so).

Miri's era-stratification system (Era 1/2/3) and historical event modifier pass are the proposed technical mechanisms. Nigel proposes historical events as a seed modifier leaving "physical traces."

The unanswered question: What specific causal relationships does the generator encode? When the generator produces an L-shaped building, does it record WHY that shape exists, and can that reason surface as environmental storytelling? Or is the L-shape purely geometric, with the player inventing the reason?

Raised by: Ozzie "Hand-Crafted vs Obviously Procedural"; Miri §2 "Historical events"; Nigel §5 Layer 4.

OQ-5: Society Profile YAML as Serde-Compatible Schema

Miri asks Tyre directly: "Can the content pipeline consume the society profile YAML format (from wiki-review R4) as a serde-compatible schema? The parameter depth is significant."

This is a technical feasibility question. The society profile has six categories, with blend weights, NULL states, and nested parameters. It needs to be consumable by the Rust pipeline.

Raised by: Miri §5 "For Tyre."

OQ-6: Era Stratification in Chunk Data Structure

Miri asks Tyre: "How does era-stratification map onto the chunk data structure? The Z-level model (D-093) is confirmed, but does the chunk system have era fields?"

Tyre's DistrictSkeleton has a zone_palette: Vec<ZoneDefinition> field, which presumably could carry era tags. But this is not made explicit in Tyre's Round 1. Araminta's visual rules depend heavily on era tags being present at block level before chunk fill runs.

Raised by: Miri §5 "For Tyre"; Araminta §1.3.

OQ-7: Size of the Cultural Ingredients Space

Nigel asks Miri: "How large is the cultural ingredients space? (Q-032 specifics.) The variety payoff of the cultural composition layer depends entirely on how many distinct ingredient combinations produce distinguishable district personalities."

Miri provides a conservative estimate ("comfortably exceeds 300 meaningfully distinct societies") but does not enumerate the combination count. Nigel's own calculation (300 worlds × 2 characters × 20 distinct cultural compositions = 12,000 meaningfully distinct games) depends on the "20 distinct compositions" assumption, which may be conservative or generous.

Raised by: Nigel §7; Miri §1 "How 300 worlds get variety."

OQ-8: Visual Vocabulary for Flavor Structure Categories

Nigel proposes six flavor structure categories for unclaimed quarters (informal economy indicators, settlement indicators, economic stress indicators, faction presence indicators, plus two others). He asks Araminta what visual vocabulary distinguishes these categories.

Araminta's empty quarter types (plaza, service alley, courtyard, vehicle staging, structural gap) overlap partially with Nigel's flavor categories but use different taxonomy. These two systems need reconciliation before chunk fill can be specified.

Raised by: Nigel §7; Araminta §3.2.

Section 3: Dissent and Alternative Proposals

D-1: Ozzie's Skepticism About Second Station Syndrome

Ozzie is not convinced the quarter system alone defeats structural recognizability. Her concern: "Even with quarter variation, if the blocks are always the same size and the streets are always aligned, I'll feel the skeleton underneath." She wants the generator to vary the SOCIAL AND POLITICAL skeleton, not just geometry within a fixed structural skeleton.

This is not a rejection of the quarter system — it's a demand that the quarter system be a consequence of social/political variation, not a separate aesthetic variation pass. Her test: "two players should be able to compare notes and find genuinely different investigation experiences."

This aligns with Nigel's Guarantee 1 (Structural Non-Repeatability Per Seed) but frames it from the player-experience side rather than the systems side.

Source: Ozzie "The Generation Sins," "What Makes the 50th Station Exciting," "Quarter System" section.

D-2: Gestalt vs. Miri on Triangle Template Instantiation Stage

Gestalt's position: "Triangle configuration is determined at the population stage... social graph → chunk fill. NOT: chunk fill → social graph. The generator must not produce spatial arrangements and then try to fill them with compatible social graphs. The social graph drives spatial requirements."

Miri's position: Templates are instantiated "at the district skeleton stage, after zoning but before block generation," with "triangle assignments (who's in conflict with whom across templates)" as a skeleton field.

These are not fully contradictory — Miri may be describing template type selection while Gestalt describes NPC assignment to triangle positions — but the vocabulary gap obscures whether they agree. Round 2 needs a shared definition of "triangle instantiation" before this can be resolved.

Sources: Gestalt §5 Q-B; Miri §5 "when D-025 templates get instantiated."

D-3: Nigel's Reframing of the Generator's Purpose

Nigel argues the generator's promise is not "300 worlds" but "300 × (characters) × (cultural combinations) × (seed entropy) distinct game experiences." He explicitly names structural randomness (who's entangled, where evidence is, which triangles are active) as more important than geometric variation.

This reframes the generator's success criteria from "produces 300 visually distinct districts" to "produces distinct investigative experiences." The implication for architecture: entanglement assignment, triangle configuration, and evidence placement are first-class generator outputs — not emergent from spatial placement.

No other participant dissents from this, but Araminta's contribution focuses entirely on spatial/visual coherence without addressing investigative experience variation. Whether these two framings are in tension will depend on how the pipeline's final stage ordering shakes out.

Source: Nigel §1, §6, §7 summary table.

Section 4: Cross-Cutting Themes

Theme 1: Every Pipeline Stage Serves Gameplay

Gestalt frames all generator requirements as "ultimately guarantees about asymmetric information production." Araminta frames all visual rules as serving "a player who can read where they are, what tier of access they're in, and where cover is." Nigel frames every variation axis by its "impact on player." Ozzie demands that every space "pay rent."

The convergence: no pipeline stage is permitted to be purely aesthetic or purely technical. Infrastructure must produce surveillance topology AND stealth topology. Zoning must produce access tier palette. Block generation must produce chokepoints. Visual grammar must communicate spatial function.

This is an implicit shared principle that Round 2 should make explicit, as it has implications for generator validation: the test of a generated district is not "does it look right" but "does it play right."

Theme 2: Dual-Reading Spaces

Gestalt's G-08 principle ("every ring location reads as mundane; criminal function visible only to those who know") appears independently in two other voices:

Ozzie requires spaces that feel "like I wasn't meant to find it" — discovered, not guided-to.
Miri requires grey economy spaces that "occupy the spaces official zoning doesn't account for."

The generator must produce spaces that serve a manifest function (visible to all) and a latent function (visible to those with specific knowledge or access). This is not a post-generation content layer — it is a generation-time property. The template tag at chunk fill time must encode both functions.

Theme 3: Physical History as a Generator Input

Multiple participants want the generator to produce spaces that feel like they have a history, not just a current state:

Miri's era-stratification (Era 1/2/3 as construction layers) is the primary mechanism.
Miri's historical event modifier pass adds anomaly traces on top of steady-state generation.
Nigel's historical event seed (§5 Layer 4) makes history leave physical traces — repurposed buildings, blocked corridors, NPCs with long-memory grievances.
Ozzie demands "historical palimpsest" — layers built by different people with different plans.

The common thread: history is not flavor text; it is a structural input that produces physical consequences. A district that experienced a corporate merger 10 years ago has two architectural styles and NPCs with residual loyalty conflicts.

Whether the generator can produce this — and whether era tags alone are sufficient, or whether a dedicated historical event layer is required — is unresolved.

Theme 4: Minimum Viable District as v0.1 Deliverable

Tyre and Gestalt both propose concrete v0.1 deliverables that validate the generator schema without running the generator:

Tyre: hand-author the Transit District as a DistrictSkeleton to validate schema expressiveness. Estimated ~3-4 developer-days.
Gestalt: define the "minimum viable district" (1 workplace, 1 bar, 1 maintenance spine, 1 transit node, 1 restricted zone, 2 active triangles) as the completeness check.

These should be treated as a single deliverable: a hand-authored DistrictSkeleton representing the Transit District, validated against Gestalt's seven spatial guarantees. If the schema can express all seven guarantees for the Transit District, it can express any generator output.

Section 5: Qatux Observations

Implicit Decision Emerging

The following implicit decision appears to be forming across Round 1 and should be formally proposed in Round 2 for Jeroen's confirmation:

Proposed decision: The district skeleton is the primary atomic output of the generator. It contains: social site slots (type + position), access topology, NPC capacity and pattern distribution per site, triangle assignments (NPC conflict topology), cultural modifier tags, multi-block reservations, corridor spine, access points, and zone palette assignments. The generator selects and arranges D-025 templates into skeletons; it does not modify templates. This is consistent with D-025 and Q-036.

If confirmed, this should become a formal D-record.

Flag: Vocabulary Divergence on "Triangle Instantiation"

Gestalt and Miri use overlapping vocabulary with potentially different meanings:

Term	Gestalt's meaning	Miri's meaning
"Triangle instantiation"	NPC assignment to triangle positions (Population stage)	Selection of D-025 template types for social sites (Skeleton stage)
"Social graph drives spatial"	NPC conflict topology must precede spatial placement	Templates carry triangle connections as a skeleton field

This may be a complementary split (Miri = which template types go where; Gestalt = which NPCs fill which roles in those templates) rather than a contradiction. Round 2 should establish shared vocabulary.

Flag: Araminta and Nigel's Empty Quarter Taxonomies Need Reconciliation

Araminta's empty quarter types: plaza, service alley, courtyard/garden, vehicle/cargo staging, structural gap (undeveloped).

Nigel's flavor structure categories: informal economy indicators (market stalls, vendor carts, repair shops), settlement indicators (container gardens, seating clusters, shrines), economic stress indicators (shacks, abandoned equipment, unauthorized storage), faction presence indicators (Commission kiosks, union halls, corporate branded infrastructure).

These two systems overlap but do not map cleanly to each other. Before chunk fill can be specified, a unified taxonomy of "what fills an unclaimed quarter" must be agreed.

Round 1 complete. Summary ready for Round 2 use. — Qatux

22 KiB Raw Permalink Blame History Unescape Escape