puzzle-design-kb/puzzles/pattern-learning.md

Pattern Learning / Knowledge Transfer

Information Architecture: Game teaches a complete mechanical system in a low-stakes domain, then requires application of exact same rules in higher-stakes domain against different target. The tutorial IS the training—no explicit instruction, only pattern discovery through interaction.

Player Action Pattern: Engage with training instances exhaustively to discover full rule set. Internalize cause/effect mappings (insult A → retort B). Face final application where mechanics are identical but consequences escalate (more rounds, permanent failure state, critical plot progression).

Core Mechanic: The learning domain and application domain share IDENTICAL underlying rules. Player success depends on recognizing that Domain A was never a different puzzle—it WAS the solution manual for Domain B. This is NOT "practice makes perfect" through repetition; it's "this system you just learned operates elsewhere."

Variations:

• Combat dialogue systems (insult/retort pairs)
• Construction/crafting frameworks (ingredient categories → application recipe)
• Code/password mechanics where training example uses same algorithm as lock

Adventure Game Implementation:

• Tutorial or "optional" early-game puzzles that teach full system
• Standard TALK/USE actions reveal mechanical rules through trial/error
• Later puzzles invoke same mechanic with different flavor/text/skins
• Player must recognize: the framework hasn't changed, only the context

---

Example Structure

Swordfight Combat System (MI1)

Learning Phase:
→ Encounter 4 types of roaming pirate NPCs with combat mini-game
→ Each duel reveals subset of insult/retort rule mappings
→ Exhaust all 16 combinations: learn every insult → correct retort pair
→ Rule: If opponent uses wrong retort, you win point; if they use right one, they win

Application Phase:
→ Sword Master fight introduces NEW insult text (20+ different lines)
→ BUT: Same 16 retort rules apply to mapped insults
→ Pattern recognition critical: "My tongue is sharper" → maps to feather-duster rule
→ First to 5 points wins (vs training's first-to-3)

Voodoo Doll Construction Framework (MI2)

Learning Phase (Act I - Largo):
→ Voodoo Lady explains 4 ingredient categories for curse doll:
  1. Something of the Thread (clothing)
  2. Something of the Head (hair/hairpiece)  
  3. Something of the Body (bodily fluids)
  4. Something of the Dead (ancestor bone/remains)
→ Player gathers these 4 types for Largo doll successfully
→ Complete doll construction = working framework learned

Application Phase (Part IV - LeChuck):
→ After explosion, discover same Juju Bag mechanic
→ Framework from Act I applies: need Thread/Head/Body/Dead components
→ New instances of each category: Skull (Head), Beard Bits (Body), 
   Underwear (Thread), Used Hanky (fluid)
→ Apply same construction recipe → empowered doll works on LeChuck

---

Key Identifiers

1. Same mechanic, different skin: Not "I learned to make Largo's curse" but "I learned the voodoo doll system"
2. Exhaustible rule space: Finite complete system can be discovered (16 insults, 4 categories)
3. Domain transfer: Learning happens in Domain A, application in Domain B, mechanics identical
4. No new teaching: Application phase provides zero tutorials; assumes player recognized the system

---

Related Types

• Multi-Faceted Plan: Both gather across multiple steps, but MFP has different categories of requirement (key+code+distracter), not one unifying system
• Observation Replay: OR reproduces exact sequence verbatim; KT applies rules to new targets
• Environmental Storytelling: KT often includes world text, but puzzle is the mechanic transfer, not narrative connection

---

Common Misidentifications (NOT Knowledge Transfer)

Apparent Similarity	Why It's Different	Actual Type
Password game: learn finger-counting logic → apply to 3 doors	Single-domain application, no mechanical transfer	Pattern Recognition / Logic Puzzle
Parrot directions: feed crackers → get navigation clues	Information collection, not system learning	Multi-Faceted Plan (direction synthesis)
Spitting contest: watch wind timing → spit in window	Observation + execution of single puzzle	Environmental Timing Puzzle
Bone maze dream: song lyrics → corridor mapping	Cryptic message decoding	Metaphor-to-Literal Translation

Critical Test: Could you describe the solution as "I learned [SYSTEM] that applies to both [CONTEXT A] and [CONTEXT B]"? If not, it's not Knowledge Transfer.