settled-reach/docs/workshops/wiki-review/round4-si.md

Round 4 — SI (Project Manager)

Responding to the lead's direction. The 300-world reframe changes everything I said in Round 3. Let me redo the math and the plan.

---

The Fundamental Shift

Round 3 assumed 20 hand-authored districts. The lead says 300 worlds, generator-driven. This isn't a 15x scaling of the same plan. It's a different kind of project entirely.

Old model: Writers produce districts. Tooling accelerates writers. New model: Engineers produce generators. Writers produce the INPUTS generators consume. Tooling IS the product.

My Round 3 bottleneck analysis (Tier 1 content = 900-1760 hours for 20 districts) is obsolete. At 300 worlds, hand-authoring Tier 1 content for each is ~15,000-26,000 hours — 8-14 person-years. Obviously not happening. The question becomes: what MUST be hand-authored, and what do generators produce from authored inputs?

---

1. What Changes at 300-World Scale

Content tiers reframe

Tier	Old Model (20 districts)	New Model (300 worlds)
Tier 1 (FRIEND, MIRROR)	Hand-authored per district. Bottleneck.	Pool of authored archetypes + generator composition. Write 30-50 FRIEND templates. Generator selects, skins with cultural parameters, assigns to district. Still the most expensive content — but amortized across 300 worlds, not written 300 times.
Tier 2 (Working Profile)	Hand-authored per district.	Generator-produced from templates + cultural brief + functional motivation. The 10-axis model IS a generation spec. Feed it a cultural brief, a functional motivation, and a social site role → out comes a Tier 2 NPC. Human review for quality, not for creation.
Tier 3 (Sketch)	Already procedural.	Fully procedural. Cultural brief + naming algorithm + role slot → complete Tier 3 NPC. No human touch.

The new bottleneck

Not content authoring. Generator quality. If the generators produce flat, predictable, samey NPCs — the game is dead. The bottleneck shifts from "how many writers do we have" to "how good are the generators at producing characters that feel authored."

This means: the writers' job isn't to write 300 districts. It's to write the SEED CONTENT that makes generators produce interesting output. Every style brief, every cultural parameter, every naming algorithm, every thematic pattern definition — these are generator inputs, and their quality determines the quality of 300 worlds.

---

2. Revised Production Model

Phase 1: Proof of Concept (v0.1)

• 1 district, fully hand-authored
• Proves the interaction model works
• SI focus: Standard sprint management. Small team.

Phase 2: Generator Development (v0.2-v0.5)

• Build and validate ALL generators:
  • NPC generator (Tier 2 + Tier 3)
  • Social site generator
  • Triangle generator
  • Naming algorithm per cultural group
  • Dialogue line expansion system
  • District layout generator
  • Pool composition system
  • FRIEND/MIRROR skinning system (authored templates + cultural parameters)
• Test each generator against hand-authored Sova output as quality benchmark
• 3-5 test districts generated, human-reviewed, iterated
• SI focus: Sprint planning around generator milestones, not content milestones. Each generator is an epic with clear acceptance criteria: "generated output is indistinguishable from hand-authored at Tier X."

Phase 3: Content Pipeline at Scale (v0.6-v0.10)

• Cultural parameter matrices for all cultural groups
• Thematic pattern definitions for all 9+ patterns
• FRIEND/MIRROR template pool (30-50 authored templates)
• PC archetype definitions for all 8
• Generator runs producing 300 worlds
• Automated validation + human spot-check review
• SI focus: Batch production management. Track generator coverage (how many worlds generated, validation pass rate, spot-check quality scores). Flag generators that produce below-threshold output.

Phase 4: Polish (pre-v1.0)

• Hand-authored elevation of key worlds (starting districts, narrative hubs)
• Cross-world consistency checks
• Gate topology finalization
• Playtesting at scale
• SI focus: Triage. Which worlds need human touch? Prioritize by player traffic patterns (starting worlds, hub worlds, quest-critical worlds).

---

3. The Generator Development Roadmap

This replaces my Round 3 tooling roadmap. Generators aren't tools that help writers — they ARE the content pipeline.

Core generators (must exist before Phase 3)

Generator	Input	Output	Acceptance Criteria
Cultural Parameter Generator	Ingredients menu selections	Complete cultural parameter set (naming rules, access tier profile, trust defaults, economic parameters, social norms)	Output passes Miri's review as "a society, not a matrix entry"
NPC Generator (Tier 2)	Cultural params + thematic pattern + functional motivation + social site role	Complete 10-axis NPC profile with dialogue tags, routine, relationships	Output is indistinguishable from hand-authored Tier 2 in blind review
NPC Generator (Tier 3)	Cultural params + role slot	Minimal NPC with name, routine, greeting, 3-5 generic lines	Feels like a person, not a placeholder
FRIEND Skinner	Authored FRIEND template + cultural params + district context	Culturally-skinned FRIEND with district-appropriate name, relationships, cultural tells, contradiction arc adapted to local context	Emotional beats land. Warmth phase feels genuine. Contradiction is culturally grounded.
Social Site Generator	District type + cultural params + economic function	Social site with role slots, triangle structures, location descriptions, atmosphere	Site feels purposeful and inhabited
Triangle Generator	Social site + NPC assignments + thematic texture	Triangle with conflicting interests, mechanical expression, cascade effects	Triangle produces a genuine FORK — not just tension
Naming Algorithm	Cultural params (heritage roots, drift stage, naming conventions)	Names at all levels: first names, surnames, place names, business names, slang	Names feel culturally coherent, not random
Dialogue Line Expander	10-15 authored seed lines + voice card + cultural params	40-60 expanded lines matching voice, cultural register, tag taxonomy	Expanded lines pass Mellanie's "read aloud" test
District Composer	Cultural params + economic function + thematic texture + NPC pool + social site templates	Complete district with all content YAML	District feels like a place with a story, not a content dump

Support generators (needed for quality at scale)

Generator	Purpose
Pool Compositor	Assembles seed-time pools per district from global NPC/contraband/role pools
Gate Topology Generator	Produces system connectivity graphs following Sacred/Profane/Middle Kingdom rules
News Ticker Generator	Produces district-appropriate ticker headlines from economic conditions + political state
Environmental Text Generator	Produces signs, menus, graffiti from cultural params + district type
PC Starting State Generator	Produces PC-candidate NPC specs per archetype per district from archetype definition + district context

---

4. What the Writers Actually Write (at 300-world scale)

The writers' deliverables fundamentally change. They no longer write districts. They write generator inputs and quality benchmarks.

Authored content (irreducible)

Deliverable	Quantity	Purpose
Cultural ingredient definitions	30-50 ingredients across all categories	Generator menu items. Each ingredient has mechanical parameters + flavor text + naming rules.
Thematic pattern definitions	9 patterns (FRIEND, MIRROR, ANCHOR, GHOST, CATALYST, THRESHOLD, REMNANT, SYSTEM, NOBODY)	Generator templates. Each pattern has: emotional function, required content depth, phase structure, composition rules with functional motivations.
Functional motivation definitions	6 motivations (HANDLER, WITNESS, TURNCOAT, CIVILIAN, OPERATOR, SKEPTIC)	Generator templates. Each motivation has: gameplay function, dialogue disposition, investigation/navigation role.
FRIEND template pool	30-50 authored FRIEND archetypes	Generator draws from these. Each is a full contradiction arc, tell system, warmth phase — but culturally unspecified. The skinner adds cultural dress.
MIRROR template pool	15-25 authored MIRROR archetypes	Same model — authored emotional arc, culturally skinned per district.
PC archetype definitions	8 archetypes	Full voice cards, moral arc structures, starting knowledge templates, agency boundary templates.
Voice card library	8 PC voices + 20-30 NPC voice archetypes	Generator selects voice, line expander matches it.
Benchmark districts	5-8 hand-authored districts	Quality benchmarks for generator comparison. Include Sova (v0.1) + 4-7 diverse districts covering different cultural groups and thematic textures.
Seed dialogue per voice archetype	10-15 lines × 30 voice archetypes = 300-450 lines	Line expander input. The most important authored content at scale — every generated line descends from these seeds.

Total irreducible authored content

Category	Hours
Cultural ingredients (30-50)	100-200h
Thematic patterns (9)	40-60h
Functional motivations (6)	20-30h
FRIEND templates (30-50)	600-1100h
MIRROR templates (15-25)	200-400h
PC archetypes (8)	120-180h
Voice card library (30)	60-90h
Benchmark districts (5-8)	400-650h
Seed dialogue (300-450 lines)	50-80h
Total	1590-2790h

That's 1-1.5 person-years of writing. Less than my Round 3 estimate for 20 hand-authored districts. But the output supports 300 worlds instead of 20. The leverage is in the generators.

---

5. 8 PC Archetypes — Production Implications

The lead wants 8 archetypes at v1.0 with fluid transitions between them. Here's what that means for production:

Per-archetype deliverables

Deliverable	One-Time	Per-District (generator handles)
Voice card	4-6h	0 (global)
Moral arc structure	4-6h	0 (global)
Knowledge attribute vocabulary	2-3h	0 (global)
Monologue trigger set	2-3h	0 (global)
Starting knowledge template	3-4h	~0.5h (generator adapts to local context)
Agency boundary template	2-3h	~0.5h (generator adapts)
Archetype transition rules	4-6h (per transition pair)	0 (global)

Archetype transition matrix

8 archetypes = 56 possible transitions (8 × 7). Not all are valid. The lead says transitions are a game mechanic — the player changes jobs, moves districts, shifts social position.

Production question: How many transition paths need authored content?

If each valid transition has a brief narrative beat (5-10 lines of monologue marking the shift), and ~30% of transitions are valid (17 paths), that's 85-170 authored lines. Manageable.

The bigger production implication: every district must support all 8 archetypes. That means every social site needs role slots for each archetype. Every NPC needs access tier mappings for each archetype. Every pool needs archetype-tagged candidates.

At 300 worlds, this is a generator constraint, not an authoring task. The District Composer must produce districts where all 8 archetypes have viable starting positions and meaningful social integration. The acceptance criterion: "a player using any archetype in any generated district can find a community, a conflict, and a reason to care within the first 5 minutes."

Rollout (per lead's direction)

Release	Archetypes	New This Release
v0.1	2 (Insider, Investigator)	Both
v0.2	3 (+Newcomer)	Newcomer = tutorial vehicle
v0.3-v0.4	5 (+2 TBD)	Expand thematic coverage
v0.5-v0.8	7 (+2 TBD)	Near-complete
v1.0	8 (+1 TBD)	Full set

Each new archetype requires: voice card, moral arc, knowledge vocabulary, monologue triggers, starting knowledge template, agency boundaries, transition rules to/from existing archetypes, and generator updates so all existing worlds support the new archetype.

The retroactive content problem: When archetype 3 (Newcomer) launches in v0.2, Sova Transit District (hand-authored for 2 archetypes) needs Newcomer content. When archetype 4 launches, ALL existing districts need archetype 4 content. At 300 worlds, this is a generator re-run, not a hand-authoring task — but the benchmark districts need manual updates.

---

6. Sacred/Profane/Middle Kingdom — Project Management Mapping

The lead approved Nigel's framework. Let me map it to what I manage — sprint planning, dependency tracking, and release scoping.

What this means for sprint planning

Layer	Sprint Implication
Sacred (rules of the system)	Engine development. Changes are rare and expensive. Once a Sacred system ships, it's locked unless we're willing to regenerate all content that depends on it. Sacred systems are v0.1-v0.2 deliverables. After that, they're frozen.
Profane (parameters)	Content production. Pool definitions, cultural parameters, naming algorithms, thematic pattern instances. This is the bulk of ongoing sprint work. Profane content is cheap to add, cheap to modify, and the generator pipeline handles it.
Middle Kingdom (sacred structure, profane cast)	The generators themselves. Templates that define structure (Sacred) but whose specific population is drawn from pools (Profane). Generator development is v0.2-v0.5. After that, generators are maintained, not rebuilt.

Sprint category allocation (long-term steady state)

Category	% of Sprint Capacity	Focus
Sacred (engine)	10-15%	Maintenance, performance, new system prototyping
Profane (content)	40-50%	New cultural groups, FRIEND/MIRROR templates, archetype additions, pool expansion
Middle Kingdom (generators)	20-25%	Generator improvements, new generator types, quality threshold tuning
Validation & QA	15-20%	Automated testing at scale, spot-check reviews, playtest feedback integration

---

7. Quality Assurance at 300 Worlds

This is the problem that scared me most about the reframe. At 20 districts, humans review everything. At 300 worlds, humans can't.

Automated quality gates

Gate	What It Checks	When It Runs
Schema validation	All YAML valid, required fields present, enums correct	Every generator run
Cross-reference validation	All NPC references resolve, all FactIds exist, all location slugs valid	Every generator run
Triangle integrity	All triangle members present, all forks produce distinct outcomes	Every generator run
Pool coverage	Every archetype has viable starting position, every social site has minimum population	Every generator run
Name collision check	No duplicate names within a district, no phonetic collisions per Paula/Miri rules	Every generator run
Dialogue coverage	Every NPC has minimum line count per tier, every situation has at least 1 line	Every generator run
Diversity metrics	Cultural distribution across 300 worlds matches target, no cultural group over/under-represented	Per-batch generation

Human quality gates

Gate	What It Checks	How Often
Spot-check review	Random sample of generated districts reviewed by writer for voice quality, cultural coherence, emotional resonance	5-10% of generated districts per batch
Benchmark comparison	Generated districts compared to hand-authored benchmarks on defined quality axes	Every generator version change
Playtest feedback	Player reports of "this felt off" or "this NPC was flat" → fed back to generator tuning	Continuous
FRIEND/MIRROR review	Every generated FRIEND and MIRROR instance reviewed by writer before shipping	100% — these are too important to skip

The FRIEND/MIRROR bottleneck at 300 worlds

If each world has 1 district with 2-3 FRIEND candidates and 1 MIRROR candidate, that's 900-1200 Tier 1 NPCs across the game. Even with the skinner model (authored template + cultural skin), each skinned instance needs human review for emotional integrity.

At 10 minutes per review: 150-200 hours of review. That's 4-5 weeks of a single reviewer's time. Manageable — but it's the one task that can't be automated. Budget for it in every release that adds worlds.

---

8. Revised Release Cadence

Release	Worlds	PC Archetypes	Generator Status	Content Focus
v0.1	1	2	N/A (hand-authored)	Prove interaction model
v0.2	3-5	3	Prototyping (NPC gen, naming algo)	Test generators vs hand-authored quality
v0.3	10-15	4	Alpha (most generators working)	Generator iteration, quality tuning
v0.4	30-50	5	Beta (all generators working)	Scale test, automated QA pipeline
v0.5	80-100	6	Production (quality gates passing)	Content pipeline at cruising speed
v0.6-v0.8	150-200	7	Stable	Profane content expansion (new cultural groups, FRIEND templates)
v0.9	250-280	8	Stable	Near-complete, polish pass
v1.0	300	8	Locked	Final validation, hand-elevation of key worlds

Critical milestones

1. v0.2: Generator quality benchmark. Can the NPC generator produce a Tier 2 NPC that passes blind review against hand-authored? If not, generators need more development time before scaling.
2. v0.4: Scale test. 30-50 worlds generated in a single batch. Automated QA catches structural issues. Spot-check catches quality issues. The pipeline either works or it doesn't.
3. v0.6: Cruising speed. Generators produce worlds faster than writers produce generator inputs. The pipeline is net-positive. From here, the constraint is "what new content do we feed the generators" not "can the generators produce enough."

---

9. What I Got Wrong in Round 3

Round 3 Claim	Correction
"20 districts = 10-15 months of authoring"	300 worlds = 10-15 months of generator development + ~1 year of authored generator inputs. Total calendar time is similar, but the work is completely different.
"3-4 parallel writers for v0.3+"	Writers needed: 2-3 for authored seed content + benchmark districts. Engineers needed: 3-5 for generators. The team composition shifts toward engineering.
"Per-district cycle: 2-3 weeks"	Per-generator cycle: 4-8 weeks to develop, then generates districts in minutes. The investment is front-loaded.
"Tier 1 content is the bottleneck"	Generator quality is the bottleneck. Tier 1 content is the most expensive authored input, but if the generators can't produce good Tier 2/3 output, the game fails regardless of Tier 1 quality.
"4 PC archetypes for v1.0"	8 archetypes for v1.0 per lead direction. The generator-based model actually makes this more feasible — once the archetype is defined, the generator handles per-district instantiation.

---

10. Ticket Implications (Not Creating Yet — Flagging for When Approved)

The ticket structure I drafted in si-ticket-changes.md was for v0.1 implementation. The 300-world reframe means the LONG-TERM ticket structure looks different:

New epics needed (long-term backlog)

Epic	Scope	Team
Cultural Ingredients Menu	Define 30-50 ingredients across all categories. Generator-readable format.	copy
NPC Generator	Tier 2 + Tier 3 NPC generation from templates + cultural params	server
FRIEND/MIRROR Skinner	Authored template pool + cultural skinning system	server + copy
District Composer	End-to-end district generation from parameters	server
Naming Algorithm System	Per-cultural-group name generation	server
Dialogue Line Expander	Seed lines → expanded pool matching voice + culture	server
Gate Topology Generator	System connectivity with Sacred/Profane/Middle Kingdom rules	server
PC Archetype Framework	8 archetype definitions + transition mechanics + per-district instantiation	copy + server
Content QA Pipeline	Automated validation + spot-check workflow + quality metrics	ci
Benchmark District Suite	5-8 hand-authored districts as quality benchmarks	copy

These are strategic backlog items. They don't go into sprints yet — but they should exist so we can plan against them.

---

SI out. The 300-world reframe is the right call. The generator-based model produces more content with less authoring than the hand-crafted model — but only if we invest in generator quality first. Tooling isn't just first. Tooling is the game.