stuff

IDEAS.md

@@ -0,0 +1,132 @@
+yes ai made this, no, i dont care.
+
+# Ideas & Future Vision
+
+Parking lot for things I want to build but aren't building yet. Nothing here is a
+commitment to a timeline — it's here so it's out of my head and safe.
+
+Rule for this file: an idea only graduates to "actually build it" once the things
+it depends on already exist. Don't build up the dependency chain. Build down it.
+
+---
+
+## The Big One: Emergent Quest Engine
+
+The dream. A quest system where quests aren't hand-written — they're generated by
+chaining several smaller generators together, each feeding the next. The goal is
+quests that feel *authored* because the pieces reference each other and reference
+the player's actual cat/state.
+
+### The chain
+1. Generate an **item** + what it does (effect).
+2. Generate a **character** who needs that item *because of what it does*.
+3. Generate a **location** the character (or the cat?) travels to.
+4. Generate an **encounter / fight**, scaled to a generated skill level, maybe
+   flavored by a custom name.
+5. Resolve it → bring the item back → gain **xp, money**, maybe **food/items found
+   on the way**.
+
+### Why it could feel special (not generic radiant-quest slop)
+- Quests reference the *specific cat*: its traits ("your chonky cat..."), its name,
+  its current stats. Two players get different quests because their cats differ —
+  not because of raw RNG.
+- The chain links cause to effect: the character needs the item *because* of what
+  the item does. The reward ties back to the chain. That's the "wait, this feels
+  written" magic.
+
+### The trap to avoid
+"Unique" and "meaningful" pull opposite directions. Pure randomization gives you
+infinite *technically unique* quests that all *feel identical* (Skyrim radiant
+quests: "go to random place, kill random thing, fetch random item" forever).
+
+The fix — same lesson as the name generator: **variety lives in the TEMPLATES,
+uniqueness lives in the SLOTS.** Ten hand-written quest templates with rich,
+state-aware slots beats one template randomized harder. Constrain to get quality.
+
+### The hard part most people underestimate
+- **Completion checking.** A quest is only meaningful if the game can tell when
+  it's *done*. Every template's win condition must be something the game actually
+  tracks. "Feed 3 times" needs a feed counter. "Happiness > 50" needs the happiness
+  stat. So quests are GATED on the systems they reference already existing.
+- **Procedural MECHANICS are way harder than procedural CONTENT.** A generated
+  *name* is just text — any string works. A generated *item effect* has to actually
+  DO something the game can execute. That means generated items pick from a FIXED
+  vocabulary of effects the game already knows how to run ("restore N hunger",
+  "add N happiness", "worth N money") — generated *parameters*, fixed *effect types*.
+  Do NOT try to generate brand-new mechanics; generate combinations of existing ones.
+
+### Quests must persist
+An active/in-progress quest has to survive save/load. So `Quest` is a model with
+`to_dict`/`from_dict` + a `version`, same discipline as Cat/Save. Active quests
+live on the Save.
+
+---
+
+## Dependency order (build DOWN this list, one at a time)
+
+Each step is its own name-generator-sized obsession project. Each one is gated on
+the steps above it. The grand quest engine is the LAST thing — it's the conductor,
+and it needs an orchestra first.
+
+0. **Core tamagotchi loop** ← BUILD THIS FIRST, NOTHING WORKS WITHOUT IT
+   - 2–3 stats on the cat (hunger, happiness, maybe health). Start with ONE.
+   - Decay over time (stats drop while away — uses elapsed-time logic).
+   - Actions that restore them (feed → hunger, pet/play → happiness). First real
+     `rules.py` verbs. Pure + testable.
+   - A reason actions aren't free → food item → money → a way to earn money.
+   - **Prove the core loop is fun before building anything on top of it.** If the
+     cat-care loop isn't satisfying, no quest engine saves it.
+
+1. **Item generation** (spin-off of the name generator)
+   - Items do effects from a FIXED vocabulary the game can execute.
+   - Generate the parameters/flavor, not new mechanics.
+   - Gated on: stats existing (effects need something to affect).
+
+2. **Character / NPC generation**
+   - Reuse the name generator for NPC names.
+   - Each NPC has a need (tied to an item effect).
+   - Gated on: item generation.
+
+3. **Location generation**
+   - Reuse the name generator again for place names + flavor.
+
+4. **Combat system**
+   - Its own whole system. Skill/level scaling, resolution.
+   - Gated on: stats, maybe items (gear?).
+
+5. **Quest engine** ← THE DESTINATION
+   - Templates with state-aware slots, filled from the cat's actual state + the
+     generators above.
+   - Orchestrates items + characters + locations + combat into a chain.
+   - Completion conditions that check real, existing systems.
+   - Rewards that pay out real money/xp/items.
+   - Gated on: literally everything above.
+
+---
+
+## Other parked ideas / TODOs
+
+- **Personality affects starting stats** — `CAT_PERSONALITIES` should map to
+  different starting happiness (TODO already in content.py). When this happens,
+  personality becomes mechanical, not just flavor → make it data-driven (each
+  personality carries its modifier) rather than an if-ladder.
+- **Save overwrite handling** — two cats named the same silently overwrite. Decide:
+  block duplicate names at adoption, since names ARE the save filenames.
+- **Saves → proper user-data dir** — currently `untitled/saves/` inside the package.
+  Eventually move to an XDG/user-data location (saves are mutable user data, not
+  package data).
+- **Delete save** flow (main-menu only, with confirm, never the active save).
+- **Web version** — pty/xterm.js bridge. Dead last, after the game is actually a game.
+- **Name generator polish** — order-3 vs order-2 experiment; soft length cap is in.
+- **A studio splash / fake boot sequence** easter egg (CatDOS-style).
+
+---
+
+## Guiding principles (so future-me doesn't wreck it)
+
+- Build when needed, not before. Park visions here; build the next small thing.
+- Variety in templates, uniqueness in slots. Constrain generators to get quality.
+- Generate parameters from a fixed vocabulary, never generate raw mechanics.
+- Keep the layers: model (data) ← rules/generation (logic) ← screens (I/O).
+- Pure logic stays testable. If you can't test it without a terminal, it leaked.
+- Prove each loop is fun before stacking the next system on it.