# Modeled level-65 equipment

Shield projections retain the level-65 references' **44 base Block** separately
from their allocated stats. The vendor tooltip line supplies this value;
`GetItemStats` and planner metadata omit it. Base Block does not consume the
discretionary stat budget, and it is not an inferred proc or bonus stat.

The current benchmark incorporates the [October 11 audit](october-11-audit.md) against client build **1.60.1.70338**: engine corrections, reselected Warrior rotations and Hunter consumables, and a confirmed gear re-search of all DPS profiles. All **201 race/build profiles across 31 builds** are replayed. The [October 8 update](october-8-overnight-update.md) records the earlier 70291 conditionality, guard failures, resource costs and unsupported features. The earlier 70009 full gear search remains historical. The [patch audit](forever-70009.md) separates verified effects from unmodeled mechanics and unresolved coefficients; these projected gearsets remain estimates, not observed endgame drops.

The equipment in the current benchmark is a **hypothetical, all-modeled scenario**, not obtainable loot. Every equipped item in each ranked profile has a v2 synthetic ID beginning at `920000001` and a name listing its numerical stats, armor and, where relevant, weapon damage and speed. A two-hand build naturally has an empty off-hand slot. Real enchants remain on the modeled items; no real item or v1 synthetic item is equipped. The generated catalog is [`forever_synthetic_gear_v2.json`](../assets/db_inputs/forever_synthetic_gear_v2.json). The verified source records remain separate in [`forever_gear_catalog.json`](../assets/db_inputs/forever_gear_catalog.json).

The complete hypothetical pool has **504 variants, 496 eligible for the benchmark**. All profiles are initially given a complete legal modeled loadout; the repeated slot-coordinate search then compares modeled alternatives using the simulator, with coupled weapon layouts considered together. This is neither an exhaustive combination search nor proof of optimal real-world gear. Candidate evaluations, independently seeded confirmation, complete profiles and class summaries are in [`artifacts/modelled_gear_search/current/`](../artifacts/modelled_gear_search/current/) and [`artifacts/modelled_gear/`](../artifacts/modelled_gear/). The earlier **mixed real/v1 modeled** benchmark is preserved at [revision `ff7b01f28`](https://github.com/gunba/wow-forever-sim/blob/ff7b01f28/artifacts/modelled_gear/forever_dps_5min.json), distinct from the separate real-item benchmark. Its 385 v1 definitions remain in [`forever_synthetic_gear.json`](../assets/db_inputs/forever_synthetic_gear.json) for historical replay and manual inspection, not for current selection.

The [proposed gear spreadsheet](../artifacts/modelled_gear/proposed_gear.xlsx) shows numerical item names, editable stat allocations, reference items, armor and weapon proofs, and coverage gaps. The pinned level-65 client stat-quantity, armor and weapon-damage tables in `assets/db_inputs/forever_model_tables/` reproduce 1,217 catalog armor values and 178 weapon ranges. They do **not** determine an item’s future allocation, drop location, proc or availability. Run `python3 tools/database/generate_forever_synthetic_gear.py`, `python3 tools/database/compile_forever_equipment.py` and `go run ./tools/database/gen_db -outDir=assets -gen=db` to regenerate the model and compiled database. The workbook generator needs the optional Python `XlsxWriter` package.

Body, jewelry and ordinary weapon alternatives share the same estimated slot capacity under an explicit, **unverified** nonlinear item-budget and special-stat price model. Modeled body armor spends at most an estimated 20% on Stamina; the 80%-capacity passive trinkets are eligible, while eight full-capacity versions are sensitivity cases only. A four-offensive-stat caster template mirrors the four-offensive-stat Strength/Agility/Crit/Hit template at the **assumed** armor spell-power price. Higher-hit caster and hybrid templates remain projected: some exchange Stamina for Hit, and their modeled costs do not establish Blizzard’s item-generation rule. Each mirror has the same estimated slot cost as its physical counterpart. This does **not** turn Intellect into spell power or prove equal DPS value. Realized class scaling, overlapping primary-stat effects, mana limits and the correct special-stat prices remain uncertain. Measured DPS changes in this scenario must not be read as verified class balance.

**Armor-material guard:** no modeled Leather or Cloth body piece has Strength. Druid gear selection is restricted to Leather and cannot equip Mail or Plate. Modeled Strength Mail draws on actual level-65 Mail patterns such as Outrider’s Mail Leggings and Defiler’s Mail Pauldrons; those are available to Mail wearers, not Druids. A separate hypothetical Mail variant swaps the observed MP5 for Hit at the **assumed** item-budget price; the reference Mail item does not have Hit. Projected Cloth offensive armor with Hit uses a caster **off-hand** allocation, not a verified Cloth armor item. Each source slot and uncertainty is listed in the spreadsheet. Jewelry and weapons are not armor-material templates and can have different allocations. These restrictions make the modeled Feral set more plausible, not an assertion about undiscovered drops.

There is no suitable verified level-65 crafted/PvP Haste allocation for these templates. The pool does not invent Haste or Expertise distributions; the +10%/+50% sensitivities scale its existing stats and weapon damage, not an unknown future mix.

Real level-65 caster wands are healing-focused and real class relics carry effects rather than comparable fixed stats. The current pool therefore contains clearly labeled **projected damage-stat wands and fixed-stat idols, librams and totems**. Since the [October 16 follow-up](october-16-follow-up.md) the benchmark no longer equips those fixed-stat relics: every real idol, libram and totem whose effect the simulator implements has a projected copy in [`forever_hybrid_relics.json`](../assets/db_inputs/forever_hybrid_relics.json) that runs the real item's effect and carries a quarter of the modeled relic's stats, and the search picks among them. A modeled thrown weapon likewise derives from a real ranged weapon table. Their reference item IDs identify the art, slot, and part of the projection, **not** a claim that a corresponding stat-only item exists or that an actual relic effect applies. Catalog effects, source, ordinary set, profession and unique flags are never copied into a model item. The exact projected stats are shown in each item’s name and the workbook.

The current modeled benchmark does **not** purchase or convert hit. The simulator uses only selected item, enchant, talent and racial hit. The selector starts with hit-bearing modeled equipment, then accepts substitutions only if no rotational hit requirement becomes less satisfied. Gear hit above the cap is wasted, never refunded as AP/SP; Tauren’s racial can reduce the number of hit-bearing items needed. Exact hit, any remaining cap shortfall and every selected item are disclosed per profile. The older paid-hit convention remains identifiable only in archived benchmark requests; it is not applied here.

The [selected-gear audit](../artifacts/modelled_gear/gear_equity.json) reports the raw physical and caster stat totals, item counts, modeled capacity points and effective hit for every race/build. Across these loadouts, the assumed capacity spans about 533–540 points: weapon layouts account for the small remaining difference. All equipped item levels are 65. This **does not** certify equivalence in offensive value or predict future drops; the item-budget formula and special-stat prices remain hypotheses.

The selector uses each class’s usual armor material to avoid identical lower-armor copies; this is a search-pool choice, not a new equipment rule. Talents, APLs, race availability, external buffs, consumables, the 300-second encounter, ordinary MP5 and full role-specific Tier 1 override remain unchanged. No real-item result should be silently compared to or relabeled as this model scenario.