Item power (⚡) redesign: bound affix variance, score god items, and account for the full 50-zone ilvl range

[stphung]

Background

Supersedes #272 (god items should outscore top Legendaries) and #332 (Legendary power variance too high from affix type weighting). Both are symptoms of the same formula, and both were written against a Zone-10 world that no longer exists.

Current formula (src/items/types.rs Item::power(), weights in src/items/scoring.rs):

power = Σ attribute_values + Σ (affix_value × affix_weight)     // weights 0.5x–2.0x

Two structural problems:

1. Affix values live on incommensurate scales. At the same ilvl/tier/rarity, an HPBonus roll is ~50–80 base, DamagePercent is 6–10, CritMultiplier is 0.2–0.35, AttackSpeed is 2.0–3.3 (src/items/generation.rs). The weight table tries to correct for scale and impact simultaneously and does neither — a CritMultiplier 1.3 (a strong roll) contributes ⚡2 while an HPBonus 125 contributes ⚡63.
2. The formula only sees attributes and affixes. God item passives (Giant's Might 150% dmg, Windborne 100% AtkSpd, Divine Bulwark 30% multiplicative DR) and non-combat bonuses are invisible, so all three god items score ⚡100.

What changed since #272/#332: there are 50 zones

ilvl_for_zone(zone_id) = zone_id × 10 is uncapped and process_item_drop passes current_zone_id (up to 50) straight through. The stat multiplier 1 + (ilvl − 10)/30 therefore spans:

Zone	ilvl	Multiplier
10	100	4.0x
20	200	7.3x
30	300	10.7x
40	400	13.7x
50	500	17.3x

Consequences:

• God items should score higher than T9 Legendaries in item power #272's acceptance criterion ("all god items > best Zone 10 Legendary, ~356") is obsolete. A perfect Zone 50 T9 Legendary scores ~2,040 under the current formula. God items are fixed at ilvl 100 / ⚡100 and get buried by roughly Zone 14 onward.
• Legendary item power variance too high due to affix type weighting #332's variance problem scales with zone: the absolute power gap between lucky and unlucky affix types grows 17x by Zone 50.
• Any redesign must answer a question the old issues never had to: what should ⚡ mean across a 50-zone world, and where do fixed-stat god items sit on that scale?

Reference items

Hand-rolled from the actual generator ranges in generation.rs (all values are legal rolls). Used for every comparison below.

#	Item	Rolls
A	T6 Legendary Ring, ilvl 40 (Zone 4) — the "Evaa ring" from #332; near-perfect rolls of low-weight types	CON 6; HPBonus 125, DR 14%, CritMult 0.53, DR 15%, HPBonus 121
B	T9 Legendary Armor, ilvl 40 (Zone 4) — mediocre defensive rolls	CON 10, WIS 8; HPBonus 102, HPRegen 13, Reflect 13%, XPGain 12%
C	T9 Legendary Weapon, ilvl 100 (Zone 10) — hot offensive rolls	STR 22, DEX 18, CON 16; Dmg 38%, Dmg 36%, Dmg 33%, CritChance 30%, CritMult 1.30
D	T9 Legendary Weapon, ilvl 300 (Zone 30) — typical good rolls	STR 53, DEX 43; Dmg 96%, CritChance 75%, HPBonus 693, DR 85%, XPGain 75%
E	T9 Legendary Weapon, ilvl 500 (Zone 50) — theoretical perfect roll	3× attr 104; 5× Dmg 173%
F	God items (Asprika / Sleipnir / Megingjord) — ilvl 100, T9, fixed stats	60 attrs; XPGain 40%; passive (30% DR / 100% AtkSpd / 150% Dmg)

Current-formula scores: A=167, B=104, C=317, D=933, E≈2,040, F=100 each.

The A-vs-B pair is the #332 tier inversion (T6 beats T9 at the same ilvl by 1.6x). F-vs-everything is #272, now much worse than the issue described.

Candidate designs

Design 1 — Base power × roll quality ("budget" model) — recommended

power = rarity_base × ilvl_mult × tier_mult × (1 + 0.25 × roll_quality)

• rarity_base: Common 10, Magic 20, Rare 35, Epic 55, Legendary 80, Mythic 120
• roll_quality ∈ [0,1]: mean of each attribute/affix roll normalized against its own max possible roll at that ilvl/tier/rarity — HPBonus 125 is judged as "a 98% HP roll", not as 125 raw points
• Deterministic progression inputs (ilvl, tier, rarity) form the base; RNG only modulates ±25%
• God items: Mythic base plus passives scored as oversized normalized rolls

Properties: tier is near-monotonic (worst T9 = best T6 at same ilvl/rarity); affix-type luck bounded at 1.25x instead of ~4x; every input already exists in generation.rs. Open detail: affix-count luck (Legendary rolls 4–5 affixes) — fold into roll_quality or into rarity_base.

Design 2 — Minimal patch: normalize affixes in place, feed passives through as affixes

Keep the formula shape. Per affix: quality × budget × weight where quality = value / max_roll(type, ilvl, tier), budget = 20 × ilvl_mult × tier_mult, and weights narrowed to 0.8x–1.5x. God item passives enter as pseudo-affixes at true magnitude (150% Dmg normalizes to 3.75× a perfect Dmg roll) with a uniform "unique passive" weight of 2.0.

Properties: smallest diff, kills the per-type scale artifact (CritMult 1.30 goes from ⚡2 to ⚡89). But tier inversions shrink rather than disappear (see table: A still beats B), and Asprika fails — 30% multiplicative DR normalizes small against a 40% DR affix roll even though it's far stronger in the damage pipeline. Fixing that requires a hand-tuned premium, which is #272's rejected option 3 wearing a hat.

Design 3 — Impact model: marginal DPS/EHP on a fixed reference hero

power = k × (ΔDPS% + ΔEHP% + Δutility%) computed against a fixed reference build (character-independence preserved). Every affix and passive gets an exact conversion instead of a guessed weight: Windborne = 2× DPS, Divine Bulwark 30% multiplicative DR = +43% EHP, flat HP judged against the reference HP pool.

Properties: most principled, player-meaningful number, new affixes scored for free. Highest effort: needs conversions mirroring combat/, revalidation when combat formulas change, and it immediately surfaces that late-zone % affix rolls are balance-broken in their own right (see "Related" below).

How the reference items score under each design

Designs 1–2 computed exactly; Design 3 is coarse (depends on reference-build assumptions).

Item	Current	Design 1	Design 2	Design 3 (~)
A — Z4 T6 ring, lucky types	167	154	161	~700
B — Z4 T9 armor, mediocre	104	184 ✅ beats A	109 ❌ still loses to A	~250
C — Z10 T9 weapon, hot	317	388	538	~1,400
D — Z30 T9 weapon, good	933	999	1,001	~3,500
E — Z50 T9 weapon, perfect	~2,040	1,733	2,912	~9,000
Megingjord	100	~2,300–2,600 (pinned*) / ~575 unpinned	740	~1,600
Sleipnir	100	~2,300–2,600 (pinned*) / ~575 unpinned	1,352	~1,200
Asprika	100	~2,300–2,600 (pinned*) / ~575 unpinned	~260 ❌	~550

* "Pinned" = god items scored at the max-zone ilvl multiplier (see decision below).

Reading the table:

• Legendary item power variance too high due to affix type weighting #332's tier inversion (A vs B): only Design 1 fixes it outright. Design 2 narrows the gap ~1.7x → ~1.5x but quality+count luck still outweighs the T6→T9 budget difference.
• God items should score higher than T9 Legendaries in item power #272 (god items): under Design 1 pinned, all three clear the Zone 50 ceiling with no hand-set values (Mythic base 120 × 17.3 = 2,076 > max Legendary 1,733). Design 2 handles Megingjord/Sleipnir but fails Asprika without arbitrary tuning. Design 3 scores passives truthfully but leaves god items mid-scale.
• Zone progression: all three designs keep median power rising with zone, matching the 1.6x/1.25x enemy scaling story.

Decision needed: where do god items sit on a 50-zone scale?

God items have fixed stats (ilvl 100) while drops scale to ilvl 500. Auto-equip already never replaces a Mythic (scoring.rs), so this is purely about what the ⚡ number communicates. Options:

• (a) Pin god items to the top of the scale — score them at the max-zone ilvl multiplier. Honors the spirit of God items should score higher than T9 Legendaries in item power #272 ("god items are the ceiling"), consistent with the never-auto-replaced rule. Recommended.
• (b) Zone-relative — let them score at ilvl 100 and drop God items should score higher than T9 Legendaries in item power #272's criterion; ⚡ becomes purely informational and god items are honestly outstatted by Zone 20+ drops.
• (c) Scale god item stats with progression — a real buff, not a scoring change; out of scope here but worth a separate issue if (b) feels bad.

Acceptance criteria (zone-aware replacements for #272's)

• At the same ilvl and rarity, a T9 with worst-case rolls scores ≥ a T6 with best-case rolls (tier monotonicity)
• Within the same ilvl/tier/rarity, max/min power ratio from affix luck ≤ ~1.3x (was ~4x)
• Median Legendary power strictly increases with zone bracket across all 50 zones
• All 3 god items score above the max possible drop from any zone (if option (a) above)
• Power stays character-independent and computed on the fly (no save migration)
• Unit tests: god-item ceiling test, statistical tier-ordering test over N generated items
• src/items/CLAUDE.md and wiki updated (both document the current 0.5x–2.0x weights)

Related observation (probably its own issue)

While hand-rolling item D: generic % affixes scale linearly and uncapped with ilvl, so a single Zone 30 Legendary affix can roll +85% DamageReduction and Zone 50 +173% Damage — independent of how we score them, these raw magnitudes look balance-broken (DR stacking past 100%?). Whatever design wins here, the generator's late-zone affix ranges deserve their own audit.