updates

content/_index.md

@@ -1,6 +1,6 @@
 ---
 title: codegirl.games
-description: We build affordable games and teach the craft along the way — documenting every step in public.
+description: I build games and show the craft along the way, documenting every step in public.
 ---
 
-We aim to build affordable games and teach things we learn along the way by documenting our journey.
+I aim to build games and show things I learn along the way by documenting my journey.

content/about.md

@@ -0,0 +1,10 @@
+---
+title: About
+description: A place where I build games and share the development process in public.
+---
+
+**Codegirl Games** is a place where I build games and share how they're made: devlogs, experiments, and lessons from real projects.
+
+I'm not pretending to have all the answers. I'm learning in public: comparing languages on the same project, applying patterns from books like *Game Programming Patterns*, and building through trial and error.
+
+Whether you're starting out, leveling up, or just curious how games get made, you're in the right place.

content/posts/_index.md

@@ -0,0 +1,6 @@
+---
+title: Posts
+description: Devlogs, lessons, language comparisons, and book reviews from my game development work.
+---
+
+Devlogs, lessons, and experiments from my game development journey.

content/posts/colony-sim-kickoff.md

@@ -0,0 +1,117 @@
+---
+title: "Starting a colony sim, lessons I'm carrying forward"
+description: "Kicking off a new Odin prototype and the grid-game patterns I already trust: cameras, tilemaps, and how to lay out entity data."
+date: 2026-07-03
+type: lesson
+series: colony-sim-prototype
+series_order: 1
+languages: ["odin"]
+tags: ["odin", "colony-sim", "game-dev"]
+---
+
+I'm starting a second prototype, a colony simulation, in [Odin](https://odin-lang.org/) and Raylib. It's early days and there's not much game there yet. This post isn't a feature tour. It's the stuff I already know works because I learned it building the [tower defense prototype](https://github.com/Codegirl-Games/tower-defense-prototype) first.
+
+If you're starting a top-down grid game, these are the foundations I keep reaching for.
+
+## The 2D camera is just coordinate math
+
+A 2D camera doesn't need a library. It's two numbers and two functions:
+
+- **Offset**: which world point sits at the center of the screen
+- **Zoom**: how many world pixels map to one screen pixel
+
+Convert world → screen:
+
+```
+screen = (world - offset) * zoom + screen_center
+```
+
+Convert screen → world (for mouse input):
+
+```
+world = (screen - screen_center) / zoom + offset
+```
+
+That's it. Panning moves the offset. The scroll wheel clamps zoom between sensible min/max values. Pan speed gets divided by zoom so movement feels consistent when you're zoomed in.
+
+The lesson I keep re-learning: **every mouse click must go through `screen_to_world` before you do anything useful.** Selection, movement commands, building placement: all of it. Forget this once and your clicks drift when the camera moves.
+
+## Tilemaps are flat arrays with helpers
+
+A tilemap is a width, a height, and a flat buffer. Tile `(x, y)` lives at index `y * width + x`. Wrap access in two helpers and never think about the math again:
+
+- `tile_index(map, x, y)`: buffer lookup
+- `tile_in_bounds(map, x, y)`: guard every read and write
+
+World position to tile coordinate is just `floor(world / tile_size)` on each axis. I used the same pattern in both prototypes. The colony sim added a comment in `tile_in_bounds`, *"learning from another game, this will become handy"*, because I skipped it early in the tower defense project and paid for it later.
+
+Terrain type can start as a `u8` per tile. A switch or lookup table maps type → color. Don't over-engineer biomes on day one; get the grid drawing and the coordinate conversions right first.
+
+When drawing, multiply tile size by zoom and cull tiles that fall off-screen. An 80×60 map is 4,800 rectangles, fine for a prototype, but the cull pass is free and keeps the pattern honest for bigger maps.
+
+## Logical tiles vs visual position
+
+Grid games have two positions whether you plan for it or not:
+
+- **Logical position**: which tile the entity occupies (`Tile_Coord{3, 7}`)
+- **Visual position**: where the sprite actually renders (smoothly interpolated between tile centers)
+
+The colonist's grid cell updates one step at a time. The circle on screen lerps toward the next tile center each frame. Gameplay stays discrete; motion looks continuous. Mix these up and pathfinding, collision, and selection all get harder.
+
+I didn't need this in the tower defense game: enemies moved in continuous world space along a path. Colony sims live on tiles. Separate the two early.
+
+## Struct of arrays vs array of structs
+
+Both layouts show up in my code. Neither is always wrong.
+
+**Array of structs (AoS)**, what the tower defense prototype uses:
+
+```odin
+enemies: [MAX_ENEMIES]Enemy,
+```
+
+Each slot is a full `Enemy` struct: position, health, speed, active flag, all together. Natural to read: `enemy.health -= damage`. Good when you often touch most fields on one entity at once.
+
+**Struct of arrays (SoA)**, what the colony sim uses:
+
+```odin
+Entity_World :: struct {
+    active:          [MAX_ENTITIES]bool,
+    position:        [MAX_ENTITIES]Tile_Coord,
+    move_state:      [MAX_ENTITIES]Move_State,
+    visual_position: [MAX_ENTITIES]rl.Vector2,
+    // ...
+}
+```
+
+Each field is a parallel array across all entities. Good when you update one system at a time (move every entity, then draw every entity) and when entities are sparse (lots of inactive slots in a fixed pool).
+
+My rule of thumb so far:
+
+| Reach for AoS when… | Reach for SoA when… |
+|---|---|
+| Entities are few and always accessed whole | You iterate one component across many entities |
+| Struct fits in cache and you're touching most of it | Many slots are inactive (object pool) |
+| Code clarity matters more than layout | Systems are split (movement, render, AI) |
+
+Both prototypes use **fixed pools** with an `active` flag, no allocate/free per spawn. That pattern transferred directly from tower defense to colony sim regardless of AoS vs SoA.
+
+## Entity handles, not raw indices
+
+The colony sim returns `Entity`, a `distinct u32`, instead of passing array indices around. Internally it's `index + 1`, with `0` meaning invalid. Small thing, but it stops you from accidentally passing a tile coordinate or a mouse value where an entity ID goes.
+
+## Update and draw stay separate
+
+Both games follow the same loop shape:
+
+1. Read input
+2. Update simulation (`world_update`, `entity_update_movement`)
+3. Draw (`tilemap_draw`, `entity_draw`)
+
+Simulation code never calls draw functions. Draw code never changes game state. Obvious, but worth stating because it's the seam that keeps things readable as files multiply.
+
+## What I'm not writing about yet
+
+Pathfinding, job queues, resources, building: none of that exists in the repo yet. When there's a full week of work to show, I'll write a proper devlog. For now, the [repo](https://github.com/Codegirl-Games/colony-sim-prototype) is a camera, a tilemap, one colonist, and a right-click move command.
+
+The game will come. These patterns are the part I'm confident in.

content/posts/game-programming-patterns-review.md

@@ -0,0 +1,81 @@
+---
+title: "Game Programming Patterns, first impressions"
+description: "A book review of Robert Nystrom's Game Programming Patterns and how I use it in my tower defense prototype."
+date: 2026-06-28
+type: books
+book: "Game Programming Patterns"
+tags: ["patterns", "books"]
+---
+
+*Game Programming Patterns* by Robert Nystrom is the reference I'm using as I build. The patterns aren't abstract; I'm already applying several of them in my [Odin tower defense prototype](https://github.com/Codegirl-Games/tower-defense-prototype).
+
+Here is where the book shows up in real code so far.
+
+## Patterns in the tower defense game
+
+### Object Pool
+
+Enemies and projectiles are never allocated per spawn. Both use fixed arrays (`[MAX_ENEMIES]Enemy`, `[MAX_PROJECTILES]Projectile`) with `acquire_enemy` and `acquire_projectile` scanning for inactive slots. A spawned unit resets its fields and sets `active = true`; on death or impact, `active = false` returns the slot to the pool.
+
+This was the first pattern I reached for, in week one. Every wave can spawn dozens of enemies and towers can fire many projectiles per second. Avoiding allocate/free in the hot path keeps the update loop predictable.
+
+### Command
+
+Player input goes through a command layer, not straight into game logic. `gather_commands` and `poll_controls_command` read keyboard and mouse state and return a `Command` value: `Place_Tower`, `Start_Wave`, `Upgrade_Tower`, or `None`. `execute_command` dispatches to the right handler.
+
+UI buttons, hotkeys, and mouse clicks all produce the same command type. Adding a new input source does not mean rewriting placement or wave logic.
+
+### Event Queue
+
+Gold changes do not happen inside combat code directly. When an enemy dies or a wave is cleared, systems call `push_event` with `Enemy_Killed` or `Wave_Survived`. At the end of `update_world`, `process_events` drains the queue and applies gold rewards.
+
+Combat systems announce what happened; the economy system decides what it costs. That separation kept the update loop readable as towers, waves, and enemy types piled on.
+
+### State
+
+The game runs in explicit phases: `Build`, `Combat`, `Game_Over`, and `Victory`. Phase controls what input is accepted (you cannot place towers during combat), when waves can start, and when the simulation stops updating.
+
+This is a straightforward state machine, not a deep AI behavior tree, but the same idea: one variable drives which rules apply this frame.
+
+### Update Method
+
+Each system owns an update function called once per frame from `update_world`: `update_wave`, `update_enemies`, `update_projectiles`, `update_towers`, `update_phase`. No single giant function walks every entity type inline.
+
+The book's "one game loop, many systems" structure maps cleanly onto separate `.odin` files as the project grew.
+
+### Game Loop
+
+`app.odin` runs the classic loop: read input, call `update_world`, render the map, entities, overlay, and controls. Update and draw stay separate; simulation code never calls draw functions.
+
+### Data Locality
+
+Enemies and projectiles live in contiguous fixed arrays rather than scattered heap allocations. I iterate the full pool each frame but skip inactive slots. Not as aggressive as a struct-of-arrays layout, but the fixed-buffer approach gives similar benefits: no pointer chasing, no allocator pressure during combat.
+
+### Type Object (archetype tables)
+
+Tower and enemy stats live in shared lookup tables (`TOWER_ARCHETYPES`, `ENEMY_DEF`), not duplicated on every instance. A placed tower stores its kind, position, cooldown, and upgrade level; range, damage, cost, and footprint come from the table row.
+
+Adding a cannon or a tank enemy is mostly a new table entry plus a behavior branch, not a new class hierarchy.
+
+### Subclass Sandbox
+
+Tower types differ by `switch t.kind` in `update_towers`: archers spawn homing projectiles, cannons fire ballistic shots with splash, ice towers apply slow on hit. I skipped inheritance trees in favor of enum variants and explicit branches. Nystrom argues this is the right call when you only have a handful of types and the differences are behavioral, not structural.
+
+### Spatial partition (grid)
+
+The map is a 2D tile grid (`Blocked`, `Path`, `Build`). Tower placement queries `can_build_at` and footprint overlap against grid cells, not against every entity on the map. A full spatial hash would be overkill at this scale; the grid already gives O(1) tile lookups for build rules.
+
+## What I have not used yet
+
+Patterns I expect to need later but have not implemented in this prototype:
+
+- **Pathfinding** for dynamic routes when tower placement blocks the path
+- **Behavior trees** or a deeper **State** machine per enemy for complex AI
+- **Observer** beyond the simple event queue (e.g. UI reacting to stat changes)
+- **Component** or **Entity-Component-System** if entity types multiply significantly
+
+## Why I keep the book nearby
+
+Nystrom names the patterns, explains the tradeoffs, and shows when *not* to use them. That matches how I work: reach for Object Pool and Command early because the problem is obvious; hold off on ECS until the entity count justifies the complexity.
+
+Expect follow-up posts that go deeper on individual chapters as I apply more patterns to the tower defense game and the colony sim.

content/posts/td-odin-week-1.md

@@ -0,0 +1,56 @@
+---
+title: "Week 1: Tower defense in Odin, project kickoff"
+description: "First week building my Odin tower defense prototype: grid map, object pool, commands, events, and a playable build/combat loop."
+date: 2026-06-12
+type: devlog
+series: tower-defense-prototype
+series_order: 2
+languages: ["odin"]
+tags: ["odin", "tower-defense", "devlog"]
+---
+
+First devlog in the tower defense series. This week I kicked off the [Odin prototype](https://github.com/Codegirl-Games/tower-defense-prototype) and went from a blank `main.odin` to a playable loop: build towers, start a wave, watch enemies walk the path, lose base health when they leak through.
+
+![Early map with build zone and L-shaped path](/images/tower-defense/td-week-1-map.png)
+
+## What I built
+
+**Day 1: map and loop.** Switched the project to Odin + Raylib. Added a fixed 30×20 tile grid with three tile kinds: blocked grass, a build zone, and a hand-authored L-shaped path. The game window is exactly map-sized (`MAP_W * TILE_SIZE`), and the main loop separates `update_world` from `render_world`.
+
+**Enemies that move.** Enemies spawn at the path start and step toward waypoints extracted from path tiles on the grid. Movement is simple vector math, no steering, no physics engine.
+
+**Object pool from day one.** Enemies live in a fixed `[MAX_ENEMIES]Enemy` array. `acquire_enemy` reuses inactive slots instead of allocating every spawn. This is the first [Game Programming Patterns](https://gameprogrammingpatterns.com/object-pool.html) idea I reached for, and it fit Odin naturally: explicit memory, no hidden allocations.
+
+**Commands and events.** Input goes through a small command layer (`Place_Tower`, `Start_Wave`). Gold changes go through an event queue (`Enemy_Killed`, `Wave_Survived`) so combat logic does not touch the economy directly. That separation made the update loop easier to read even before I had much gameplay.
+
+**Towers, waves, and phases.** By the end of the week I had one tower type (Archer) backed by a `Tower_Archetype` table: range, damage, fire rate, cost in one place. The game alternates between **Build** and **Combat** phases. Press `N` to start a wave; a spawner drips out enemies with scaling health and speed. Towers pick the nearest target in range and apply damage directly. Survive the wave, earn gold, place more towers.
+
+## What worked
+
+- Odin's struct enums and fixed arrays made the object pool straightforward, no fighting the language
+- Data-driven tower archetypes: adding stats later did not require rewriting placement logic
+- Command + event split kept `update_world` readable as systems piled on
+- Raylib got something on screen fast; I spent the week on game logic, not boilerplate
+
+## What broke
+
+- **No projectiles yet.** Towers subtract health instantly. It plays, but it does not look or feel like a tower defense game yet
+- **Path is not pathfinding.** `build_path` scans the grid for path tiles in row order. Fine for a hand-drawn map, useless once I want procedural levels or tower placement that reroutes enemies
+- **HUD lives in the renderer.** Gold, base HP, wave count, and phase hints are drawn inline in `render_world`. Works for now, will get messy
+- **Tower placement during combat.** The command gatherer had a duplicate mouse handler that let you place towers mid-fight. Small bug, easy miss
+
+## Repo snapshot
+
+By June 12 the `game/` package had separate files for map, path, enemies, towers, waves, events, commands, and rendering: about 450 lines added across the week. Not pretty, but the skeleton for every port (C, C++) is visible: fixed pools, explicit phases, data tables instead of inheritance trees.
+
+Next: projectiles, split out the HUD, and more tower types.
+
+<figure class="post__video">
+  <iframe
+    class="post__iframe"
+    src="https://www.youtube-nocookie.com/embed/MKBSuTkClys"
+    title="Week 1: Tower defense in Odin, project kickoff"
+    allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
+    allowfullscreen
+  ></iframe>
+</figure>

content/posts/td-odin-week-2.md

@@ -0,0 +1,49 @@
+---
+title: "Week 2: Projectiles, footprints, and a UI detour"
+description: "Second week on the Odin tower defense prototype: homing projectiles, multi-tile towers, render split, and learning to stick with Raylib for UI."
+date: 2026-06-19
+type: devlog
+series: tower-defense-prototype
+series_order: 3
+languages: ["odin"]
+tags: ["odin", "tower-defense", "devlog"]
+---
+
+Week two on the [Odin prototype](https://github.com/Codegirl-Games/tower-defense-prototype). Last week ended with instant-hit archers and a keyboard-driven HUD. This week the game started to look like a tower defense: arrows fly, towers occupy real space on the grid, and I got my first proper UI, after a brief and expensive detour through third-party UI libraries.
+
+## What I built
+
+**Projectiles.** Archers no longer subtract health on the frame they fire. Towers spawn homing projectiles from a second object pool (`[MAX_PROJECTILES]Projectile`), track the target enemy by slot index, and deal damage on impact. Same pattern as the enemy pool from week one: acquire slot, reset fields, mark inactive when done.
+
+![Wave 1 combat with homing projectiles](/images/tower-defense/td-week-2-projectiles.png)
+
+**Render split.** `render_world` became a thin orchestrator: `render_map`, `render_enemies`, `render_towers`, `render_projectiles`. Each system owns its draw calls. Small refactor, big readability win as files grew.
+
+**Multi-tile towers.** Tower archetypes gained `footprint_w` and `footprint_h`. The archer is 1×2 tiles, taller than a single cell. Placement now checks whether the full rectangle fits on build tiles and does not overlap other towers. Rendering draws a gold rectangle sized to the footprint instead of a fixed 28×28 square.
+
+**Memory cleanup.** Added explicit `delete` calls for dynamic arrays (`path`, `towers`, `events`) when the game loop exits. Odin will not save you from leaking if you allocated with `append`.
+
+**Overlay bar.** Moved gold and base health out of scattered `DrawText` calls into `overlay.odin`, a top bar with consistent padding and colors. Wave count and combat hints stayed in `hud.odin` for now.
+
+**Control panel.** Bottom bar with clickable buttons: select Archer, start wave. `poll_controls_command` feeds into the existing command layer so UI clicks and keyboard shortcuts share one path. Added `utils.odin` with a reusable `draw_button` helper for centered labels.
+
+## What worked
+
+- Projectile pool mirrored the enemy pool: copy the pattern, ship faster
+- Footprint-based placement forced me to think in grid coordinates early; multi-tower-type layouts will need this anyway
+- Command layer absorbed UI input cleanly: buttons return `Command` values just like keyboard handlers
+- Rip-and-replace on Clay was painful but left me with a simpler codebase than I started with
+
+## What broke
+
+- **UI library detour.** Tried ImGui bindings, then [Clay](https://github.com/nicklockwood/clay) for two days. Gold moved to Clay, start-wave became a Clay button, then I deleted all of it and rewrote the overlay in pure Raylib. Lesson: for a small game HUD, immediate-mode Raylib is enough. Do not import a layout engine until you have a layout problem.
+- **Tower selection half-wired.** The Archer button sets `world.selected_tower`, but `execute_command` still hardcodes `.Archer` on placement. UI looks done; logic is not.
+- **Split HUD.** Gold and health live in `overlay.odin`, wave/enemy count in `hud.odin`, controls in `controls.odin`. Three files for one screen; next cleanup pass needed.
+- **Range bug (later fix).** A typo in the distance function made archer range longer than intended. Caught at the end of the week; fix landed June 26.
+
+## Repo snapshot
+
+June 13–19 added about 470 net lines across 18 files. New modules: `projectile.odin`, `overlay.odin`, `controls.odin`, `utils.odin`. Still one tower type, but the archer now shoots, occupies space, and has a shop button.
+
+Next: end screens, fullscreen, more tower types, and enemy variety.
+

content/posts/td-odin-week-3.md

@@ -0,0 +1,52 @@
+---
+title: "Week 3: Fullscreen, variety, and splash damage"
+description: "Third week on the Odin tower defense prototype: end screens, render-to-texture scaling, three tower types, enemy archetypes, wave recipes, upgrades, and cannon splash."
+date: 2026-06-26
+type: devlog
+series: tower-defense-prototype
+series_order: 4
+languages: ["odin"]
+tags: ["odin", "tower-defense", "devlog"]
+---
+
+Week three on the [Odin prototype](https://github.com/Codegirl-Games/tower-defense-prototype). Week two left me with one tower, one enemy, and a HUD split across three files. This week the prototype started feeling like a game: win/lose screens, resizable fullscreen, three tower types, three enemy types, wave recipes, and cannons that splash.
+
+## What I built
+
+**End screens and unified overlay.** Added `endscreen.odin` for game over and victory overlays. Moved wave count and enemy totals into `overlay.odin` and deleted the leftover `hud.odin` split. One top bar for economy and combat status; end screens dim the world and show the result.
+
+**Fullscreen via render texture.** New `display.odin` renders the fixed 960×640 game world into a `RenderTexture`, then scales it to whatever window size the player uses. `F11` toggles fullscreen. Mouse coordinates go through `game_mouse()` so clicks still map to grid tiles when the viewport letterboxes. The game logic stays pixel-fixed; only presentation scales.
+
+**Three tower types.** Archer (homing arrows), Cannon (2×2 footprint, ballistic arc), Ice (slow effect, smaller footprint). Each has its own archetype row: cost, range, fire rate, footprint, upgrade caps. Shop buttons in `controls.odin` grew to match: select tower, click the map, place.
+
+**Tower preview.** Hovering the build grid while a tower is selected draws a ghost footprint before you commit gold. Small UX win that made placement feel less like guessing.
+
+**Enemy variety.** Replaced the single grunt with archetypes: Grunt, Runner (fast, fragile), Tank (slow, thick). Each scales health and speed with wave number and gets its own color.
+
+**Wave recipes.** Waves are no longer `5 + wave * 2` of the same enemy. `build_wave_recipe` returns an ordered list of `(kind, count)` entries: wave 1 is six grunts, wave 2 mixes grunts and runners, later waves add tanks. The spawner drips through the recipe entry by entry.
+
+**Tower upgrades.** Towers track `upgrade_level`. Stats scale via `tower_stats_at_level`: extra damage and range per level, paid from gold during build phase.
+
+**Ballistic projectiles and splash.** Projectiles gained a `Projectile_Mode`: homing for archers, ballistic for cannons. Cannon shots fly in a fixed direction; on impact they call `apply_splash_damage` in a radius defined on the archetype. First time area damage changed how I thought about placement: kill zones, not just single targets.
+
+**World refactor.** Split `World` into nested structs for economy, combat, and wave state. Gold and enemies moved out of flat fields. Cleaner ownership before more systems land.
+
+## What worked
+
+- Render-to-texture scaling solved fullscreen without rewriting every coordinate
+- Enemy and tower archetype tables made adding Runner/Tank/Cannon mostly data changes
+- Wave recipes are easy to read and tweak; no code change to reshuffle wave 3
+- Projectile modes reused the same pool; ballistic and homing share acquire/update/render paths
+
+## What broke
+
+- **Half-finished tower rollout.** Cannon and Ice landed before projectiles and shop buttons caught up; several commits of "no projectiles, no button" in the log
+- **Wave recipe memory leak.** Forgetting to `delete` the old recipe before building a new spawner leaked dynamic arrays every wave start. Fixed same day
+- **Path length check inside the enemy loop.** A `return` on short paths aborted the entire update instead of skipping one enemy. Moved the guard outside the loop
+- **Range typo.** A bug in `distance` made archer range longer than intended, caught and fixed at the end of the week along with a range buff
+
+## Repo snapshot
+
+June 20–26 added about 650 net lines across 16 files. New modules: `display.odin`, `endscreen.odin`. Deleted: `hud.odin`. Three towers, three enemies, two projectile modes, and a win/lose loop.
+
+Next: tests, more refactors, and cleaning up the control panel as tower count grows.

content/posts/td-odin-week-4.md

@@ -0,0 +1,51 @@
+---
+title: "Week 4: Tests, refactors, and data tables"
+description: "Fourth week on the Odin tower defense prototype: splitting tower placement, table-driven defs, first tests, and undoing a nested world refactor."
+date: 2026-06-29
+type: devlog
+series: tower-defense-prototype
+series_order: 5
+languages: ["odin"]
+tags: ["odin", "tower-defense", "devlog"]
+---
+
+Week four on the [Odin prototype](https://github.com/Codegirl-Games/tower-defense-prototype). Week three shipped fullscreen, three tower types, enemy variety, and splash damage. This week was quieter on features and heavier on structure: splitting files, pushing logic into data tables, and writing my first tests.
+
+## What I built
+
+**Split tower logic.** `tower.odin` had grown to handle combat, stats, rendering, and placement. I pulled placement into `tower_placement.odin`: footprint checks, overlap tests, `try_place_tower`, ghost preview. Combat and archetypes stay in `tower.odin`. Placement is geometry; combat is timing and targeting. Keeping them separate made both files easier to navigate.
+
+**Data tables over switch statements.** Several `get_*` functions became lookup tables:
+
+- `TILE_COLORS` replaced a `tile_color` switch
+- `ENEMY_DEF` replaced per-kind switch logic with base stats, per-wave scaling, draw radius, and gold reward in one row per enemy
+- Tower archetypes dropped a redundant `kind` field and gained a `color` column for rendering
+
+Tweaking tank gold from 5 to 10 or archer cost became a one-line edit instead of hunting through switch arms.
+
+**Constants consolidation.** UI layout lived in helper functions that recalculated button rectangles every frame. Moved to `constants.odin`: `ARCHER_BUTTON_RECT`, `START_WAVE_BUTTON_RECT`, `STARTING_GOLD`, `STARTING_BASE_HEALTH`, overlay colors, control bar dimensions. `controls.odin` got shorter and the layout stopped drifting.
+
+**First tests.** Added `math2d_test.odin` (distance, `move_toward`) and `wave_test.odin` (start wave, clear wave, recipe progression). Small suite, but it caught a wave recipe memory leak during gold rebalancing: the kind of bug I'd fixed once by hand and reintroduced during a refactor.
+
+**Economy pass.** Gold rewards moved into enemy definitions. Tank kills pay more than grunts. Tower costs live in archetype rows. Balanced enough to play-test without obvious snowball or stall.
+
+**World struct simplified.** Flattened `base_health` and map init into cleaner helpers. Started the week by reverting the nested `Economy` / `Combat` / `Wave` sub-structs from week three, back to a flat `World` with direct field access.
+
+## What worked
+
+- Splitting placement from combat scaled better as tower logic grew
+- Table-driven defs made adding and tuning enemy/tower stats mostly data changes
+- Tests paid for themselves immediately on wave spawner edge cases
+- Flat `World` struct was easier to reason about than nested sub-structs for a game this size
+
+## What broke
+
+- **Nested world refactor, reverted.** Week three split `World` into `Economy`, `Combat`, and `Wave` sub-structs. Looked clean on paper. Every system suddenly needed `world.combat.enemies` instead of `world.enemies`, accessors multiplied, and nothing got simpler. Reverted on June 27. Don't refactor structure until the current shape is actually hurting you, and wait until you have tests.
+
+- **Test cleanup is manual.** Odin tests that allocate dynamic arrays need explicit `defer delete`. Forgot once; leak showed up in the test runner, not the game.
+
+## Repo snapshot
+
+June 27–29 touched 14 files, ~490 net lines. New modules: `tower_placement.odin`, `math2d_test.odin`, `wave_test.odin`. Three towers, three enemies, wave recipes, upgrades, splash damage, fullscreen, end screens, plus a test suite to build on.
+
+Next: pathfinding, more tower shop wiring, ice slow polish, and whatever breaks once I add a fourth tower type.

content/posts/theprimeagen-spire-deckbuilder.md

@@ -0,0 +1,125 @@
+---
+title: "Building a browser deckbuilder in vanilla JavaScript"
+description: "How a birthday gift became a 6,000-line roguelike deckbuilder: state machines, commands, data tables, and what the commit history looks like when you ship anyway."
+date: 2026-07-04
+type: lesson
+languages: ["javascript"]
+tags: ["javascript", "patterns", "deckbuilder", "web"]
+---
+
+Last year I built a parody *Slay the Spire* game in the browser as a birthday gift for [ThePrimeagen](https://www.twitch.tv/theprimeagen). It started as a joke. The [repo](https://github.com/codegirl-007/theprimeagen-spire) ended up at ~6,400 lines of JavaScript across 60 files, with two acts, dozens of cards, community-written birthday messages, and a full roguelike loop you can play without a build step.
+
+You can play it at <a href="https://theprimeagenbirthday.com" target="_blank" rel="noopener noreferrer">theprimeagenbirthday.com</a>.
+
+This post is not a feature tour. It is how the project was built, what the git history actually shows, and where *Game Programming Patterns* shows up outside my Odin prototypes.
+
+## What shipped
+
+The game is a static site: `index.html`, ES modules, CSS split by screen. No React, no bundler in production (npm was added briefly for tests, then removed).
+
+The loop matches StS closely enough to feel familiar:
+
+- Branching map with battles, elites, shops, rest sites, and events
+- Turn-based combat with energy, block, weak/vulnerable, and intents
+- Deck building: strike/defend staples plus dev-themed cards (`Terminal Coffee Rush`, `Production Deploy`, `Code Review`)
+- Relics with hook functions (`onTurnStart`, `onDamageTaken`, etc.)
+- Two acts with different enemy rosters and map layouts
+- Win/lose screens, mid-run saves to `localStorage`, and a pre-launch countdown that blocked play until September 9, 2025
+
+The flavor is extremely online. Enemies are stream/community in-jokes. Events quote Lewis and Tolkien. The victory screen unlocks birthday messages from people who sent notes for Prime. That part is personal. The architecture underneath is reusable.
+
+## What the commits look like
+
+There are 88 commits from August 30, 2025 to March 11, 2026. Rough phases:
+
+**Week one (Aug 30–Sep 2): gameplay exists.** The initial commit already added ~7,600 lines: map, battle engine, card UI, styling. After that it was iteration: card costs, enemy tuning, keyboard vs mouse fixes, double-tap to play cards, swipe sounds, acts, and a commit literally titled `think this is the final gameplay commit`.
+
+**Week two (Sep 3–Sep 10): content and polish.** Birthday messages landed in batches (`Birthday Messages`, `phpeepee!`, `casey!`, `DHH!`, `ken wheeler and AOP`). Bug fixes for deck initialization, event HP/energy leaks, block reset between turns. UI passes on the battle screen and welcome message. Balance commits: `nerf act 2`, `nerf dax`, `Un-nerf act 2w`.
+
+**Week two, structure (Sep 8):** `implement state machine`. Gameplay was already there; this commit extracted map, battle, shop, rest, event, victory, defeat, and relic selection into discrete states. That refactor made the rest of the project maintainable.
+
+**Quiet period, then March 2027 prep.** From September to March the repo sat mostly idle. Then a concentrated refactor week: split client vs shared code, moved data files, serialized shop/reward state for future networking, WebP assets, save behavior fixes (`fix block leak between enemy turn`), and a 1,700-line Cloudflare co-op design doc (`tutorial.md`) for authoritative multiplayer later.
+
+The history is not a clean agile epic. It is burst development, meme commit messages, and a second pass when you already know the game works.
+
+## Architecture that held up
+
+### State machine for screens
+
+`GameStateMachine` registers one class per screen: `MAP`, `BATTLE`, `REWARD`, `SHOP`, `REST`, `EVENT`, `VICTORY`, `DEFEAT`, `RELIC_SELECTION`. Each state implements `enter`, `exit`, `render`, and optional save/restore hooks.
+
+Battle mid-run resume works because `BattleState.getSaveData()` persists the enemy, flags, and `battleInProgress`. On load, bootstrap checks whether you were mid-fight and routes back into combat instead of dropping you on the map with orphaned state.
+
+This is the same *State* pattern I use for Build/Combat phases in the [tower defense prototype](https://github.com/Codegirl-Games/tower-defense-prototype), applied to UI flow instead of simulation phases.
+
+### Command pattern for input
+
+Player actions go through command objects (`PlayCardCommand`, `EndTurnCommand`, `MapMoveCommand`, etc.) executed by a `CommandInvoker`. Keyboard shortcuts, mouse clicks, and shop buttons all funnel into the same paths.
+
+That separation mattered when input got fancy: single number press raises a card, double press plays it. InputManager handles code review picks, shop purchases, and map navigation without duplicating game rules in event listeners.
+
+### Data tables for content
+
+Cards, enemies, relics, and map nodes live in plain JS objects:
+
+```javascript
+coffee_rush: {
+  id: "coffee_rush",
+  name: "Terminal Coffee Rush",
+  cost: 0,
+  type: "skill",
+  effect: (ctx) => { /* ... */ },
+}
+```
+
+Adding content means adding rows, not subclass trees. Enemy AI is mostly `(turn) => ({ type, value })` functions. Relics use optional hook objects. The `Code Review` card sets `pendingCodeReview` on the root; battle render and InputManager know how to show the pick-one-of-three overlay.
+
+Same idea as `TOWER_ARCHETYPES` and `ENEMY_DEF` in Odin: behavior stays in code, stats and identity stay in tables.
+
+### Shared vs client split (March refactor)
+
+Late refactors moved simulation-ish code under `src/shared/` (`engine/`, `data/`, `game/`) and kept DOM/render/input under `src/client/`. The goal was a future where a Cloudflare Durable Object owns authoritative state while the browser keeps rendering.
+
+Multiplayer never shipped. The split still made the codebase easier to reason about: battle math does not live beside `innerHTML` templates.
+
+## Bugs the commits keep fixing
+
+Roguelikes hide nasty state bugs. This repo is no exception:
+
+- **Block leaking between turns.** Fixed in separate commits for player and enemy turn boundaries. Block must reset at turn start; missing one side means silent damage inflation.
+- **Mid-battle saves.** Saving `_battleInProgress`, enemy HP, and hand state to `localStorage`, then restoring on reload. Easy to get wrong when most testing happens in one sitting.
+- **Event modifiers vs combat.** `fix health bug and energy bug inside events` shows how one-shot screens can corrupt run state if they touch player stats outside the battle engine's expectations.
+
+The `?screen=battle` and `?screen=shop` URL params plus mock player data in `bootstrap.js` were added so individual screens could be tested without playing to them every time. Worth copying for any UI-heavy browser game.
+
+## Performance passes
+
+Late commits focused on load and layout cost: WebP conversion, dropping aggressive image preload, caching swipe sound, reducing layout churn during battle animations. For a static birthday game, this was optional polish. For a public deploy on slow mobile networks, it matters.
+
+## How this connects to my other work
+
+I keep [*Game Programming Patterns*](https://gameprogrammingpatterns.com/) nearby while building the Odin prototypes. This JavaScript project applies several of the same ideas in a different shape:
+
+| Pattern | Here | Tower defense (Odin) |
+|---|---|---|
+| State | Screen flow (map, battle, shop) | Build / Combat / Game Over |
+| Command | PlayCard, EndTurn, MapMove | Place_Tower, Start_Wave |
+| Update method | Per-state `render()` + battle engine steps | `update_enemies`, `update_towers`, etc. |
+| Data locality | Plain objects, shuffle/draw on arrays | Fixed pools, archetype tables |
+
+Different language, same instinct: separate input from rules, separate screens from simulation, push content into data.
+
+## What I would do differently
+
+- **State machine earlier.** Gameplay landed first; the refactor on Sep 8 touched 15 files. Starting with states would have hurt day-one momentum but saved mid-project pain.
+- **Smaller render files from the start.** `render.js` was enormous before feature folders (`battleRender`, `mapRender`, etc.) split it up.
+- **Link the live site in the README.** The game runs at <a href="https://theprimeagenbirthday.com" target="_blank" rel="noopener noreferrer">theprimeagenbirthday.com</a>; the repo should say that up front next to the clone instructions.
+- **Move the repo under the studio org.** It still lives at `codegirl-007/theprimeagen-spire`; my prototypes now live under [Codegirl-Games](https://github.com/Codegirl-Games).
+
+## Worth a post?
+
+Yes, but as a lesson, not a weekly devlog. There is no neat week-by-week timeline after launch week. The value is architectural: how far vanilla JS gets you, what patterns transfer to Odin, and an honest commit log that includes `Ligma balls` next to `implement state machine`.
+
+Play at <a href="https://theprimeagenbirthday.com" target="_blank" rel="noopener noreferrer">theprimeagenbirthday.com</a>. To explore the code, start at `src/client/app/bootstrap.js` for the state registration, `src/shared/engine/battle.js` for combat rules, and `src/shared/data/cards.js` for content shape. Multiplayer design notes are in `tutorial.md` if you want to see the planned next step that never left the design doc.
+
+The game was a gift. The structure is the part worth stealing for the next project.