What it is. A table maps a roll to a reward. Weight each entry, sort into rarity tiers, and pull from it with a deterministic RNG. The whole design of “what you get and how badly you wanted it” lives in these numbers.

Player fantasy / why it’s fun. The pull of the unknown chest, the spike when the gold-bordered thing drops, the story of the run that hinged on one lucky roll. Loot turns repetition into anticipation — but only when the odds feel authored, not hostile.

When to use / when NOT

• Use when acquisition should feel earned-but-surprising: roguelike drops (genre-roguelike), deckbuilder card offers (genre-deckbuilder), horde-survival upgrade picks (genre-horde-survival), any system-reward-schedules with a variable payoff.
• Use to gate build variety (system-build-diversity) — the table decides which pieces a player can even assemble.
• NOT for deterministic progression. If the fourth boss should always give the double-jump, that is system-progression, not a table. Randomizing a required unlock is a trap.
• NOT as a substitute for content. A wide table over three items is a antipattern-content-desert wearing a rarity costume.
• NOT where the player expects a promise kept. Quest rewards, story beats, tutorial gifts — hand-place those.

Variants

Twist seams

• Loot but the table is visible and you influence it (perspective). Surface the odds and let the player nudge weights — a “luck” stat, a reroll, a slot they lock. Balatro shows every card’s tag and lets you tag-hunt; Hades shows the boon rarity before you pick. Transparency converts blind gambling into a decision. See pattern-meaningful-choice and anchor-balatro.
• Drops but rarity is set by how you killed, not luck (mechanic-swap). Skill sets the tier: a no-hit kill, a parry-finish (mechanic-parry), a hazard-assisted death rolls on a richer table. Now variance rewards play, not patience — the pattern-risk-reward curve becomes the loot curve. Compare anchor-shadow-of-mordor nemesis kills and system-spawn-directors tuning stakes.

Tuning levers

• Rarity spread — the ratio between common and rarest weight. Tight (5:1) reads as “everything’s useful”; wide (1000:1) manufactures chase items and highlight-reel moments. Decide the emotional gap you want, then set the numbers.
• Pity threshold — how many dry rolls before the guarantee fires. Set it just past the point where players start to notice the drought. Too low and the rare feels handed-out; too high and you’ve already lost them (antipattern-grind-wall).
• Dupe handling — dead dupes rot a table fast. Convert to shards/currency (system-economy), remove-until-refill, or upgrade the owned copy. Never let the jackpot roll be a thing they already have.
• Table scope — global vs per-source. A shared table is simple; per-enemy or per-biome tables let you place meaning (this boss is the only orange source). More scope, more authorship, more upkeep.
• Floor & ceiling — guarantee a minimum-quality drop so a chest is never nothing, and cap how many jackpots stack so a hot streak doesn’t trivialize the run (system-difficulty-and-dda).
• Level-gating the pool — swap or reweight entries by depth/act so early tables can’t spit end-game power (antipattern-power-creep, system-progression).

How it wires to Hayao

• Every roll goes through world.rng. No Math.random, no Date.now-seeded noise — determinism is sacred, and a seeded table means a run replays identically and a level’s drops are provable. Weighted pick = one rng draw into a prefix-sum over the weights.
• Ordered iteration. Build the weighted list in a stable order before you draw, or two machines disagree on the same seed. Sort keys, don’t rely on insertion whim.
• Pity is state, not vibes. The dry-streak counter lives in sim state and is part of world.hash() — it saves, loads, and replays with the run (system-save-and-checkpoint).
• Cosmetic is the sparkle, not the roll. The gold beam, the rarity chime, the item-name flourish are pure view — mark them cosmetic so they never touch the hash. The drop is logic; the dazzle is decoration (pattern-juice-choreography, see design/JUICE.md).
• For a randomness primitive in isolation, read the procgen lab under sandboxes/; for the reward-cadence half, pair with system-reward-schedules.

Fails when…

• Variance erases agency. If the run’s outcome is 90% which orange dropped and 10% how you played, you’ve built a slot machine. Loot should amplify skill expression, not replace it. See antipattern-rng-frustration and pattern-fairness-and-trust.
• The dry streak has no floor. No pity, wide spread, long grind — the tail of the distribution is where players quit. Cap the worst case (pattern-anti-frustration).
• Junk drowns signal. A table that’s 95% vendor-trash makes every drop a shrug. Trim, or route junk straight to currency (antipattern-currency-spaghetti).
• The choice is fake. Three drops that are strictly worse than the one you have is a non-decision (antipattern-fake-choice).
• The metagame solves it. If there’s one correct farm and one correct table, the loot is theater over a spreadsheet (antipattern-solved-metagame, antipattern-boring-optimal).

Verify

Odds are a claim you can prove. Roll the table thousands of times against a fixed seed and assert empirical frequencies match the authored weights within tolerance; assert the pity counter always fires by its threshold; assert no roll can exceed the level-gated ceiling. Feel-check the drought and the spike against the roguelike track in verify-with (design/FUN.md#10-·-traditional-roguelike). The reward lands elsewhere — see design/JUDGE.md for whether the drop reads as loot at a glance.