settled-reach/docs/workshops/wiki-review/round2-nigel.md

Round 2 — Nigel (Sandbox & Replayability)

Cross-Pollination Synthesis

I've read everyone's Round 1. Here's what I'm seeing: there's more convergence than the tension framing suggests. Let me work through the four assigned tensions, then flag some connections nobody else has made.

---

Tension 1: FRIEND Pool Depth — How Many, At What Cost, When?

Gore and I agree. Let me make that explicit.

Gore said: "THE FRIEND cannot be procedurally generated." I never said generate. I said draw from a pool of fully authored candidates. Gore then said: "Write 4-6 FRIEND-eligible NPCs at full Tier 1 depth per district. The randomizer selects which one is THE FRIEND for this playthrough. The content is authored; the selection is procedural."

That's exactly my proposal. We're in violent agreement. The question isn't whether, it's how many and when.

The authoring cost, honestly

Gestalt mapped the bidirectional knowledge requirement. Paula mapped the PC upgrade fields. Let me put real numbers on what a FRIEND candidate costs:

Component	Lines/Effort	Notes
Full 10-axis profile	2 wiki pages	Same as Kael/Sera
Authored dialogue	25-35 lines	No generation expansion
Trust-gated dialogue	15-20 lines	Tier 1 depth
Unprompted lines	5-8 lines
Monologue (per PC)	10-15 lines	x2 characters = 20-30
Tell observation lines	5-8 lines
Contradiction discovery	3-5 lines
PC-FRIEND relationship stubs	1-2 paragraphs per PC	Bidirectional — Paula's "reversed dual-lens"
Total per FRIEND candidate	~70-100 authored lines + 2 pages prose	No shortcuts

That's serious. Each additional FRIEND candidate is roughly equal to building another Kael or Sera from scratch. And Gore is right — you can't cut corners. The tell progression, the contradiction arc, the five phases, the retroactive contamination of warmth — all of it requires Tier 1 authored depth.

My revised proposal: phased rollout

v0.1: 1 FRIEND per character (2 total). Kael for smuggler, Sera for detective. Ship what we have. The architecture supports pool selection, but the pool has exactly one entry per character. This is fine for the vertical slice — nobody replays a 30-minute demo five times.

v0.2: 2 FRIENDs per character (4 total). Add one new FRIEND candidate per PC archetype. Now the randomizer draws from {Kael, NewFriend1} for the smuggler and {Sera, NewFriend2} for the detective. Four possible FRIEND pairings. The second playthrough IS genuinely different.

v0.3+: 3 FRIENDs per character (6 total). Nine possible FRIEND pairings. Metagaming effectively impossible. Two players comparing notes will have had different FRIENDs.

What the architecture needs NOW (even for v0.1's single-entry pool)

This is the part I want to fight for. Even if v0.1 ships with Kael-only and Sera-only, the content files should be structured AS IF they're pool entries:

# content/districts/sova-transit/friends/pool.yaml
smuggler_friends:
  - npc_id: kael-davan
    contradiction_type: spatial_deviation     # B-7 meeting
    tell_type: behavioral                     # looks left when lying
    discovery_method: single_observation       # one dramatic moment

detective_friends:
  - npc_id: sera-venn
    contradiction_type: avoidance_pattern     # leaves when Torek arrives
    tell_type: behavioral                     # topic deflection + overcompensation
    discovery_method: cumulative_pattern       # three observations

When v0.2 adds a second FRIEND, it adds a second entry. No restructuring. The randomizer reads the pool. The content was always pool-shaped. Cost of this structural decision: near zero. Value: enormous later.

The anti-metagaming constraint that matters

Gore's insight about dramatic irony is crucial here. The FRIEND pool doesn't just prevent "I know Kael's tell" metagaming. It creates the second-playthrough experience Gore described: "Kael is just a dock worker. You know his secret from last time, but your character doesn't."

If Kael is always THE FRIEND, this irony is interesting once. If Kael is sometimes THE FRIEND and sometimes just a colleague, the irony is DIFFERENT each time. One playthrough: "I know Kael's secret and he's my FRIEND." Next playthrough: "I know Kael's secret but my FRIEND is Renn and Kael is just... there. Nervous. And I can't do anything about it."

That's not metagaming prevention. That's a whole new emotional texture that the pool structure creates for free.

---

Tension 2: Content Architecture Convergence

Adopting Tyre's framework, extending it with pool semantics

I'm dropping my parallel mod manifest format. Tyre's three-layer stack with ADD/REPLACE/MERGE is the right architecture. My mod.yaml was reinventing what Tyre already solved more cleanly. Let me map my replayability requirements onto Tyre's structure.

How pool-based randomization fits into Tyre's directory structure

Tyre's structure:

content/
  districts/
    sova-transit/
      npcs/
        kael-davan.yaml
      templates/
        logistics-hub.yaml    # Social site: roles, NPC assignments

My extension: the template file declares role slots that the randomizer fills from the NPC pool:

# content/districts/sova-transit/templates/bar.yaml
template_id: bar
roles:
  - slot: owner
    assigned: lera-sessik        # Fixed — Lera IS the bar
    pool_eligible: false

  - slot: regular
    count: 2-4                   # Randomizer picks 2-4 regulars per seed
    pool:                        # NPCs eligible for this slot
      - torek-lintar
      - harek
      - olin
    pool_eligible: true          # Modders can add NPCs to this pool

  - slot: bartender
    assigned: sess               # Fixed for v0.1
    pool_eligible: false

  - slot: friend_detective       # THE FRIEND slot (detective's)
    pool:
      - sera-venn                # Only one candidate in v0.1
    pool_eligible: true          # v0.2 adds more

This maps cleanly onto Tyre's overlay operations:

• ADD: A mod drops a new NPC YAML into npcs/ and adds the NPC's ID to a role slot's pool list. Tyre's MERGE on the template file concatenates the pool arrays.
• REPLACE: A mod replaces an NPC's YAML entirely (Tyre's entity-REPLACE).
• MERGE: A mod adds dialogue lines to the pool (Tyre's line-pool MERGE). The randomizer selects which NPCs are active, and only active NPCs' lines enter the runtime pool.

Gestalt's global/_districts split is compatible

Gestalt proposed _global/ for schema/factions/technology and districts/ for local content. Tyre proposed global/ and districts/. Same structure, different underscore convention. I don't care about the underscore — pick one and ship it. The important thing for replayability:

Global content is NOT randomized. Factions, technology, FactId definitions — these are fixed vocabulary. The randomizer operates on district-level content: which NPCs fill which role slots, which FRIEND is selected, which NPC is compromised.

District content IS the randomization surface. The template role-slot system I described above lives entirely within districts/. A mod that adds a new district is adding a new randomization domain. A mod that adds NPCs to an existing district is expanding the pool.

My concrete ask for Tyre

One thing Tyre's structure doesn't yet include: a seed configuration file that the randomizer produces at game-start. Something like:

# runtime/seed-state.yaml (generated, not shipped)
seed: 847291
district: sova-transit
assignments:
  friend_smuggler: kael-davan
  friend_detective: sera-venn
  compromised_inspector: drin
  wavering_member: pell
  bar_regulars: [torek-lintar, harek, olin]
  contraband_primary: lattice-components
  contraband_secondary: medical-grade
  torek_ring_connected: true      # Gestalt's T3 randomization flag

This is the record of what the randomizer decided. It's saved with the game state. It's what makes two players' games different. Tyre doesn't need to design the randomizer — but the content architecture needs a place for this output to live.

---

Tension 3: Contraband Shuffle — Does It Break Cultural Coherence?

I'm walking this back. Partially.

Gore is right that the game is about complicity, and the specific contraband type is load-bearing for the moral texture. Miri's Krenn brief is beautiful and culturally specific — the Commission regulation of lattice components isn't arbitrary, it's locally embedded. "The Commission regulates lattice tech" → "people can't afford upgrades" → "the ring exists because the system fails" → "pragmatic Krenn culture doesn't moralize about how needs get met." That chain only works for lattice components.

If I shuffle the primary contraband to severance tech, the cultural logic breaks. "The ring is smuggling counter-surveillance equipment" doesn't connect to "affordable healthcare" or "the system failed working people." Severance tech is political, not economic. It changes the ring from "healthcare providers" to "political resistance" — and that's a different game in a different setting.

Revised proposal: fix primary, vary secondary

Primary contraband is fixed per district. Sova Transit District smuggles lattice components. Period. That's culturally grounded, morally load-bearing, and tied to the specific Commission-regulation dynamics of Krenn System. It's what makes the detective's job morally ambiguous. It's what makes the smuggler's justification honest. Gore's complicity thesis depends on this.

Secondary contraband varies per seed. What ELSE is in the crates?

Secondary	Moral Shift	Gameplay Shift	Cultural Fit
Medical-grade replacements (seed A)	Amplifies sympathy — "they're saving lives"	Detective feels worse about busting the ring	High — same Commission regulation, different end-users
Severance tech (seed B)	Adds political dimension — "this is bigger than Sova"	Detective realizes the ring connects to something larger	Medium — political, not economic, but still anti-institutional
Enhanced capability mods (seed C)	Adds luxury dimension — "some of this IS just vanity"	Detective finds moral clarity easier; smuggler justification weakens	High — same supply chain, different market segment
Nothing extra (seed D)	Pure lattice components	Clean moral frame, no complications	Maximum — baseline experience

The secondary contraband doesn't change the ring's identity or the cultural logic. It adds a WRINKLE. "We're smuggling lattice components for people who need them. ...And also some stuff I don't ask about." The smuggler's relationship with the secondary cargo is itself a moral question: do you look? Do you ask Devra what's in the sealed containers?

This preserves Gore's complicity thesis while adding per-seed texture. The detective who discovers severance tech in the crates has a different "oh no" moment than the one who discovers medical-grade replacements. But both are investigating the same ring, in the same culturally grounded setting, with the same NPC relationships.

What this means for the fact catalog

Paula flagged that FactIds are currently discovery-shaped (Suspects → KnowsDetails). The secondary contraband needs its own FactId per type, gated behind the primary:

contraband.secondary_cargo:
  Suspects: "Some of these containers don't match the usual profile."
  KnowsOf: "There's more than lattice components moving through here."
  KnowsDetails: # varies by seed:
    # medical-grade: "Medical-grade replacements. For people whose lattices are degrading."
    # severance: "Severance equipment. Counter-surveillance. Someone upstream is political."
    # enhanced: "Enhanced capability mods. Not medical. Luxury market."

The FactId exists globally. The KnowsDetails text is selected by the seed configuration. This fits Tyre's content pipeline — the fact definition is in global/knowledge/facts.yaml, the seed-specific text variant is runtime data, not content.

---

Tension 4: PC-as-NPC Scope — v0.1 or v0.2?

Two different things called "PC-as-NPC"

I've been conflating two concepts. Let me separate them.

Concept A: The other PC exists as an NPC in your playthrough. When you play the smuggler, the detective is an NPC in the world. When you play the detective, the smuggler is an NPC. This is what Ozzie described: "Someone ELSE reacts to you like they already know you." Kael says "There you are" because the smuggler PC has a pre-existing life.

Concept B: PC identity is drawn from a randomizable pool. Instead of always being "the same smuggler," you could be a DIFFERENT dock worker each playthrough. Different name, different relationships, different starting knowledge.

Concept A is v0.1. Concept B is v0.2+.

Concept A (v0.1): The smuggler and detective each need an NPC-format brief (Paula's proposal #7). When you play the smuggler, the detective NPC follows their routine — arrives at the bar, talks to Sera, asks questions. The smuggler can observe this. When you play the detective, the smuggler NPC works their shifts, runs ring operations, hangs out with Kael.

This is low additional authoring cost because both PCs already have most of what's needed:

• The smuggler's routine is defined by their ring membership and dock work
• The detective's routine is defined by their investigation pattern and bar visits
• Both characters' relationship stubs exist in the NPC profiles that reference them

What's MISSING (and what we need to author for v0.1):

1. NPC-mode behavior scripts for both PCs: what do they do when no player controls them? 5-7 routine entries each.
2. Reverse dual-lens notes: how does each NPC see the smuggler NPC? How does each NPC see the detective NPC? Paula's "reversed dual-lens" — these are the SAME data inverted from the NPC profiles, but they need to be compiled explicitly.
3. Observation monologue for when the player notices the other PC-as-NPC: 3-5 lines per character. "Commission. Has to be. Why are they here?" (smuggler seeing detective NPC). "Dock worker. Efficient, unremarkable." (detective seeing smuggler NPC).

That's maybe 20-30 authored items total. Manageable.

Why it matters for v0.1: Ozzie's "ultimate payoff" depends on this. The player finishes the smuggler path, starts the detective path, and sees the smuggler NPC working the docks. "That was ME." The second-playthrough wow moment IS the PC-as-NPC. Without it, the second playthrough is just "different dialogue." With it, the second playthrough is "I'm watching my previous character from the outside and they don't know what I know."

Concept B (v0.2+): PC identity randomization requires multiple PC-viable NPCs per archetype, each with full relationship stubs against all other NPCs. The combinatorial authoring cost is real (Gestalt mapped it). Each new PC candidate needs:

• Full NPC brief
• Relationship stubs with every FRIEND candidate (bidirectional)
• Starting knowledge inventory
• Orientation monologue pool
• Reverse dual-lens from every NPC who knows them

For 2 PC candidates per archetype x 2 FRIEND candidates per archetype, that's 4 PCs needing relationship stubs against 4 FRIENDs = 16 authored relationship pairs. Not impossible, but not v0.1.

Recommendation summary

Concept	Scope	Cost	Payoff
Other PC exists as NPC (Concept A)	v0.1	~20-30 authored items	Second playthrough wow moment
PC identity randomization (Concept B)	v0.2+	16+ relationship pairs per district	True replay variety in PC identity
FRIEND pool (1 per character)	v0.1	0 additional (Kael + Sera exist)	Architecture supports future expansion
FRIEND pool (2 per character)	v0.2	~140-200 authored lines + 4 pages	Metagaming defeated, FRIEND surprise
FRIEND pool (3 per character)	v0.3+	~140-200 more lines + 4 pages	Comparison test passes

---

Connections Nobody Else Made

1. Gestalt's T3 randomization + my network shuffle = the same system

Gestalt proposed making Torek's ring connection randomizable per seed. I proposed shuffling WHO is compromised (Drin vs. Resha vs. Maret). These are the same mechanism: a per-seed flag on NPCs that determines their entanglement status.

We should generalize this into a district entanglement map in the seed configuration:

entanglement_overrides:
  torek-lintar:
    ring_connected: true          # seed A: Torek is ring
    # ring_connected: false       # seed B: Torek is mundane
  drin:
    compromised: true             # seed A: Drin has gambling debt
    # compromised: false          # seed B: Drin is clean
  resha:
    compromised: false            # seed A: Resha is just new
    # compromised: true           # seed B: Resha is being blackmailed

This is the "network shuffle" I proposed, but expressed as Gestalt's per-seed flag system. Same mechanism, same content files, different instantiation per seed. The detective can't memorize "Drin is always compromised" because Drin isn't always compromised.

The D-029 entanglement ratio (30/50/20 flat/mundane/entangled) constrains the randomizer: no seed can make everyone entangled or everyone clean. The ratio must hold within its variance band.

2. Miri's cultural brief + my replay variation = cultural tells vary too

Miri's Krenn brief gives us "perk" (Finnish curse), "kylm" (cold), "vakt" (lookout shift — criminalized by usage). Beautiful. Now consider: what if heritage vocabulary usage is ITSELF a randomizable social tell?

In seed A, Drin uses "vakt" casually — it's just a word for shift, nobody reacts. In seed B, the word "vakt" has been heat-mapped by Devra as ring-specific slang, and anyone who uses it in front of the detective gets a nervous correction from Renn. The same word has different social charge per seed.

I'm NOT proposing we randomize Miri's language work. I'm saying: the cultural substrate she built has replayability potential that's free to exploit. Different seeds can give different social weight to the same cultural markers. The detective who learns "vakt means they're involved" in seed A can't use that knowledge in seed B where "vakt" is just normal speech.

3. Paula's smuggler-specific attributes + my network shuffle = smuggler replayability

Paula identified the biggest content gap: the smuggler's mechanical vocabulary is underdeveloped compared to the detective's. Her proposed attributes (operational_reliability, exposure_risk, loyalty_assessment, leverage_held, social_debt) are exactly what the network shuffle needs.

If WHO is compromised varies per seed, then the smuggler's exposure_risk assessment of each NPC ALSO varies per seed. In seed A, the smuggler's knowledge includes {drin: {exposure_risk: "critical", leverage_held: "gambling_debt"}}. In seed B, Drin is clean and the smuggler knows {drin: {exposure_risk: "low", leverage_held: "none"}} — but Resha has the entry instead.

Paula's smuggler attributes aren't just content parity. They're the data model for what varies per seed in the smuggler's playthrough. The detective's investigation varies because evidence is in different places. The smuggler's experience varies because the risk map is different. Same mechanism, different character lens.

4. Mellanie's environmental text + seed variation = the news ticker gut-punch scales

Mellanie cataloged ~65-90 environmental text items. Most of these are fixed per district (signs, menus, safety notices). But the news ticker (15-20 headlines per cycle) is the perfect seed-variation surface.

The news ticker is already identified as wow moment #5 (D-039): "Same ticker, opposite monologue reactions per character." Now extend that across seeds: different seeds surface different ticker headlines, which trigger different monologue reactions, which create different "oh" moments.

Seed A ticker: "Commission announces expanded lattice certification program." Smuggler monologue: "Expanded certification. More hoops. More people priced out." Detective monologue: "Good. Tighter regulation means fewer off-market mods."

Seed B ticker: "Medical-grade lattice shortage reported in outer Reach systems." Smuggler monologue: "Shortage. That's why the price is up. That's why we move what we move." Detective monologue: "Supply disruption. Watch for increased black market activity."

Same game system. Different ticker seed. Different emotional texture. And it costs Mellanie almost nothing — she writes a pool of 40-50 ticker headlines, the randomizer selects 15-20 per seed, the monologue reactions are already per-character.

---

Summary: What I'm Asking For

Decisions to record

1. FRIEND pool architecture adopted now, populated incrementally. v0.1 = 1 per character, v0.2 = 2, v0.3 = 3. Content files are pool-shaped from day one.
2. Primary contraband fixed per district, secondary varies per seed. Lattice components are always the core of Sova's ring. Secondary cargo (medical/severance/enhanced/none) varies.
3. PC-as-NPC Concept A (other PC as NPC) is v0.1 scope. Requires ~20-30 authored items. PC identity randomization (Concept B) is v0.2+.
4. District entanglement map as part of seed configuration. Per-seed flags on NPCs determine ring connection, compromise status, and role assignments. D-029 ratio constrains.
5. Adopt Tyre's content architecture (three-layer stack, ADD/REPLACE/MERGE) with template role-slot extension for pool-based randomization.

New tickets for SI

1. Create friends/pool.yaml structure in content directory (even if single-entry for v0.1)
2. Create NPC-format briefs for both PCs (smuggler + detective) — Concept A, v0.1 scope
3. Write observation monologue for PC seeing other-PC-as-NPC (3-5 lines per character)
4. Define seed configuration schema (what the randomizer produces at game-start)
5. Add pool_eligible and pool fields to template schema
6. Define entanglement override schema for per-seed NPC flags
7. Write secondary contraband variant text for contraband.secondary_cargo FactId (3 variants)
8. Add news ticker headline pool (40-50 entries) to Mellanie's environmental text scope

Style guide additions

1. Template role-slot definitions should be part of every social site spec
2. FRIEND-eligible NPCs need a friend_metadata section in their profile: contradiction type, tell type, discovery method
3. PC-upgradable NPCs need the fields Paula specified: starting EntityKnowledge, starting FactId inventory, access tier map, reverse dual-lens

---

The test hasn't changed. Two friends play the game. They compare notes. They played different games. Not because we randomized everything into mush — but because we randomized the RIGHT things (FRIEND identity, NPC roles, secondary contraband, evidence locations) while keeping the authored emotional core (Tier 1 FRIEND arcs, primary contraband, cultural substrate, triangle structures) rock-solid.

Gore's complicity thesis holds across all seeds. Ozzie's wow moments fire in every configuration. Miri's cultural substrate grounds every variation. And the second playthrough — EVERY second playthrough — surprises you.

That's not a feature request. That's the game.