Three-Card Climb Lab: A Choice/Strategy Playground Where the Choice Space Is a Dial

Three-Card Climb Lab: A Choice/Strategy Playground Where the Choice Space Is a Dial

The One-Sentence Essence

This is a laboratory-grade minimal game instance built for the Choice / Strategy archetype: every round you pick 1 card from attack / shield / economy, your stacked stats auto-resolve against a guardian that grows stronger each round, and you win by surviving 12 rounds. It does not pretend to be a commercial game — no art package, no deck-building meta. It is a choice-space experiment rig: the archetype's two primary design axes (cards per round, enemy growth slope) are exposed as sliders, so you can turn them yourself and watch the difficulty change.

Behavior Loop

1. Read the board: check your HP/ATK/DEF/gold, the current guardian's HP/ATK, and the next-round guardian preview printed right on the canvas.
2. Deal: the system deals N cards (N = 2/3/4, adjustable) with weights ATK 35% / DEF 30% / Economy 35%.
3. Pick 1: the chosen card applies immediately (ATK +X / DEF +X / gold +X); the rest are lost — opportunity cost lands at once.
4. Auto-resolve: hits to kill = ceil(foe HP / your ATK); damage taken = (hits − 1) × max(1, foe ATK − your DEF). The formula is printed above your hand.
5. Camp: +3 gold kill bounty, then every 10 gold auto-converts to 3 HP (heals only what is missing, cap 30) — economy is stored blood.
6. Advance: HP at 0 means defeat; surviving round 12 means victory; otherwise the enemy grows by the slope and you loop back to step 1.

Primitive Annotations

• Choose(cardId) — pick 1 of N cards; the rest are discarded (choose)
• Compare(card value, enemy threat) — weigh "what is this card worth now" against "how much will I get hit this round" (compare)
• Accumulate(ATK/DEF/gold) — card values permanently stack onto your panel (accumulate)
• ResolveCombat() — auto-resolve with a public formula; every report line is traceable
• ConvertGold(10g→3HP) — economy converts into sustain at camp
• AdvanceRound() — the enemy grows by the slope; 12 rounds is the cap

Parameter Cheat Sheet

| Parameter | Meaning | Default | Range / Formula | |-----------|---------|---------|-----------------| | slope | Enemy growth curve slope (slider) | 1.2 | 0.6 ~ 2.0, step 0.1 | | cards | Cards per round (slider) | 3 | 2 / 3 / 4 | | Enemy HP | Grows each round | 6 (round 1) | round(3 + round × slope × 2.2); 35 on round 12 at default slope | | Enemy ATK | Grows each round | 2 (round 1) | round(1 + round × slope × 0.8); 13 on round 12 at default slope | | Player start | HP / ATK / DEF / gold | 24 (cap 30) / 5 / 1 / 0 | fixed | | ATK/DEF card value | Appreciates slowly | — | 2 + ⌊round/4⌋ + random 0~1 | | Economy card value | Appreciates noticeably | — | 5 + ⌊round×1.2⌋ + random 0~3 | | Kill bounty | Per guardian killed | +3 gold | fixed | | Camp ratio | Gold to HP | 10 gold → 3 HP | heals up to the cap of 30 | | Win condition | Rounds survived | 12 | fixed; final score = 300 + gold + remaining HP |

Design Trade-offs

1. The two sliders are the archetype's two primary axes. cards tunes the size of the choice space: fewer cards means the type you want is more likely to be missing, and decisions degrade from "pick the best" to "make do" — the choice space itself is the difficulty. slope tunes the speed of risk accumulation: the steeper it is, the sooner a one-dimensional strategy (all attack, or all savings) gets punished. Difficulty is not hidden in a spreadsheet; it is handed to the player to feel.
2. The formula is printed on the canvas. The damage formula sits right above your hand, and every round's combat report (hits to kill, hits taken, HP recovered at camp) is traceable line by line — delivering the archetype checklist item "outcomes must be explainable": the player can say "I lost because of that card on that round."
3. Economy is stored blood. Gold does nothing on its own; it only converts to HP at camp at 10:3, healing just what is missing. That makes economy a delayed-defense card, so the three card types genuinely trade off against each other instead of collapsing into a fake choice with converging net value.
4. Next-round enemy preview. Information is fully public, pushing the challenge from "guessing" back to "weighing"; the residual uncertainty lives in the random card supply — neither pure guessing nor pure arithmetic.
5. Laboratory honesty. The cross-session record is just two W/L counters, and the math model is visible at a glance. The demo deliberately does one thing: let you turn the dials and watch how "choice space × growth slope" rewrites the answer to "can I survive 12 rounds?"

Implementation Notes

• Pure static single file: built on the AtomCore engine (Canvas 2D, portrait 2:3 auto-fit), zero dependencies and zero external requests; ?lang=zh switches the copy to Chinese.
• Two-phase round machine: phase is only choose / resolve; a 550ms timer advances the round after a pick, guarded by a token against the "stale timer advances a fresh run" race.
• Dealing and resolving are separated: dealCards rolls the weighted random hand, while pick does exactly four steps — apply, resolve, camp, report. Combat is fully automatic: the interaction cost is zero, so the decision cost is everything.
• Sliders as live config: game.params feeds enemyFor and dealCards in real time — change the slope and the next guardian is immediately stronger; change the card count and the next deal immediately changes.
• Debug hook: window.__atomGame exposes a read-only game object for automated verification and numeric replay.