official-dnd-ranking-review

name: official-dnd-ranking-review description: Use for official, partner, UA, third-party published, or legacy-compatible D&D 5.5e/2024/5R ranking and review work in this project. Enforces local DND5e_chm source verification, Rankings/changelog.md criteria, version matching for community comments, full rule/resource/action/spell/feat interaction analysis, and prevents using model memory for class/subclass feature text.

Official D&D Ranking Review

Use this skill for any work under root Rankings involving official / partner / UA / third-party published / legacy-compatible D&D class, subclass, build, rule, spell, feat, item, or balance review.

Do not use this skill for homebrews unless the user explicitly asks to compare a homebrew against official standards; homebrew output uses homebrews\Rankings\changelog.md.

Do not treat raw creator-homebrew entries discovered from external 5etools-homebrew class/subclass files as public official / UA / partner / third-party checklist material by default. If the source is not an official / UA / partner / clearly public third-party published resource and is instead Patreon, personal-repo, Discord/forum, or otherwise creator-homebrew material, keep it out of official-review-checklist.md. Place any private calibration checklist or review under homebrews\Rankings External and use the homebrew workflow unless the user explicitly promotes that exact source into the public published-resource scope.

DPR-Next Backup Maintenance

Experimental six-category DPR standards are archived in SKILL.dpr-next.md beside this file. The active workflow currently uses legacy damage aspects. DPR-next remains worth maintaining as a calibration candidate and difficult-damage audit tool, but it is not ready to replace the active review standard. When this active skill changes for reasons unrelated to damage-aspect taxonomy or DPR calculation workflow, sync the same non-DPR change into SKILL.dpr-next.md. Future changes to six-category DPR aspects or calculator workflow should continue to go to the backup first unless the user explicitly re-enables DPR-next for active reviews.

Non-Negotiable Rule

Never write or revise a class/subclass feature review from model memory. First verify the current feature names, levels, source tags, and wording in 5etools-src / 5etools-cn JSON when available, then confirm the Chinese name and local wording through DND5e_chm or, for missing legacy 5e-era material, DND2014_legacy.

If community comments mention a feature that does not match the verified 5etools / local Chinese-source version being reviewed, treat that comment as historical or mismatched and do not use it as evidence for the current review.

Example: for 2024 奉献之誓 Oath of Devotion, do not mention 2014 Purity of Spirit if DND5e_chm\玩家手册2024\角色职业\圣武士\奉献之誓.htm has the 5.5e feature 卫护斩 Smite of Protection.

Every review must be completed as an individual evaluation of the latest verified source. The formal review may use only:

• the latest verified source text from 5etools-src / 5etools-cn, with Chinese names and local phrasing confirmed from DND5e_chm, DND2014_legacy, or the recorded effective project source for special mappings;
• this skill and any directly applicable companion skill;
• version-matched community discussion after local text verification.

Do not let the user prompt, prior assistant wording, previous review text, file-change comparison, requested direction, or surrounding discussion determine the ranking, design score, or public-facing review conclusion. Those materials may identify what to inspect, but the final judgment must be rebuilt from the latest source, project criteria, and verified community calibration. If the prompt gives a new rule, new standard, or additional factual premise that should affect future reviews, update the relevant skill or changelog first, then apply it as project criteria. If the prompt merely argues for a result, treat it as a hypothesis to test, not as evidence.

The target audience is Chinese. When naming any official / partner / UA / third-party published class, subclass, feature, spell, feat, item, monster, rule, action, condition, weapon mastery, or other rules object, use the Chinese translation from DND5e_chm whenever it can be found. You may include the English name in parentheses on first mention for searchability, but the displayed name must not be English-only unless no Chinese translation can be located locally.

Common abbreviations already understood by Chinese D&D communities do not need to be translated or expanded into forced Chinese wording. It is acceptable to write abbreviations such as DPR, RAW, RAI, DM, PC, NPC, DDB, UA, T1-T4, ASI, BA, OA, AoE, source abbreviations such as PHB / DMG / MM / TCE / XGE / EGW, and build shorthand when that abbreviation is the normal community term. If an abbreviation may be ambiguous in context, explain it once, but do not replace a familiar abbreviation with an awkward Chinese translation.

Known verified terminology note: Origin Feat(s) is 起源专长, as found in DND5e_chm\玩家手册2024\专长\起源专长.htm. Do not translate it as 出身专长.

Class and subclass display names, directory names, and review filenames must also match the verified Chinese name in DND5e_chm. If an existing ranking file uses a stale or unofficial translation, rename the file and update headings/links to the local Chinese name before continuing the review. Example: if the local source names Oath of the Watchers as 守望者之誓, use 守望者之誓.md, not 守望之誓.md, even if an older path or previous review used the shorter wording.

Filename source markers are mandatory for non-first-party official material. If a review file is for UA, partner, or third-party content, append a full-width bracket marker after the verified Chinese class/subclass/build name:

• UA: 子职中文名（UA）.md; for 5e-era UA content specifically, use the resource marker 5E UA, so the filename should be 子职中文名（5E UA）.md.
• Partner resource: 子职中文名（资源缩写）.md, for example 回音骑士（EGW）.md if a separate marked file is needed.
• Third-party resource: 子职中文名（资源缩写或资源中文简称）.md, using the abbreviation/name found in DND5e_chm where possible. Unmarked filenames are reserved for current official first-party published or official legacy resources. Do not let a UA replacement review occupy the unmarked official filename.

Required Workflow

1. Identify scope.

   • Official / partner / UA / third-party published / legacy-compatible content writes to root Rankings.
   • Homebrew writes to homebrews\Rankings and follows homebrew archiving rules, not this root workflow.

2. Read project criteria.

   • Before substantial official ranking work, consult Rankings\changelog.md.
   • Use the current power/function/overall rank order exactly: EX > S+ > S > S- > A+ > A > A- > B+ > B > C > D > E > E- > F.
   • Preserve EX restraint, Tier definitions, and “overall is not an average” rule.

3. Locate the rules text.

   • For mechanical discovery and rules-object navigation, use the two 5etools submodules as the main searchable resources because their JSON structure is easier to navigate reliably than compiled CHM HTML:
     • 5etools-src: primary mechanics source for current English rules data, latest 2024 / partner updates, class/subclass/spell/feat/item structure, source tags, and cross-source lookup.
     • 5etools-cn: parallel Chinese 5etools cn2.0 mirror. Use it as a main search resource when a Chinese JSON counterpart exists, and as an EN/CN alignment aid.
     • 5etools-homebrew: 5etools external homebrew/partner JSON repository. Use it as the local source for 5etools loadexthomebrew partner / third-party material when the content is not present in 5etools-src\data. Treat this repository as partner / third-party published data, not first-party official rules, unless the JSON source metadata and project review scope prove otherwise.
     • 5etools-unearthed-arcana: 5etools prerelease / Unearthed Arcana JSON repository. Use it as the local source for 5etools loadextprerelease UA and playtest material when the content is not present in 5etools-src\data. Treat this repository as UA / prerelease material, not official published material, unless a later official source has republished the same option.
   • Use DND5e_chm and DND2014_legacy primarily for translation confirmation, local Chinese naming, existing project wording, and legacy-source confirmation. DND5e_chm remains the main authority for verified Chinese display names and Chinese rules phrasing in review output; do not replace established local Chinese names merely because 5etools has a different or missing translation.
   • Use DND2014_legacy only when dealing with legacy / 5e-era material that is missing from the current 5etools / DND5e_chm path, especially 5e-era UA content. Do not use it to override current 2024 / 5.5e rules unless the review is explicitly a legacy-compatible evaluation.
   • Practical search order for most official / partner / UA / third-party reviews:
     1. Search 5etools-src\data and, when useful, 5etools-cn\data for the rule object, source, level, feature text, spell list, item data, or structured relationship.
     2. Confirm Chinese name, translation, and local review terminology in DND5e_chm; also check existing Rankings when a naming convention has already been established.
     3. For UA / prerelease content missing from those sources, resolve the source through 5etools-unearthed-arcana\_generated and open the referenced JSON file.
     4. For partner / third-party external content missing from 5etools-src\data, resolve the source through 5etools-homebrew\_generated and open the referenced JSON file.
     5. For 5e-era UA / legacy content missing from those sources, confirm the Chinese legacy text in DND2014_legacy.
     6. If 5etools-src / 5etools-unearthed-arcana / 5etools-homebrew has a newer English rule object that lacks a verified Chinese translation in DND5e_chm, DND2014_legacy, 5etools-cn, or existing Rankings, keep the English official name and clearly mark any Chinese wording as a temporary / descriptive translation rather than a verified local name. Do not fabricate polished Chinese names as if they were official.
   • 5etools JSON search guidance for agent work. Prefer JSON under data/, not the page HTML:
     • Search indexes: 5etools-src\search and 5etools-cn\search contain generated search indexes that can be much faster for locating resources before opening the underlying data/ JSON. Check relevant index-*.json files when direct rg over data/ is noisy or when the source/category is unclear. Current useful files include index.json for the broad site index, index-item.json for items, index-alt-spell.json for alternate spell search, index-foundry.json when Foundry-style object lookup helps, and index-partnered.json for partnered-resource counterparts. 5etools-cn\search\index-partnered.json may already contain the Chinese name for a partnered resource. Use these indexes to locate the object, source, category, and candidate CN name, then still open the underlying data JSON and confirm final Chinese wording against DND5e_chm, DND2014_legacy, 5etools-cn, and existing Rankings as appropriate.
     • CN 5etools online loader references, verified in 5etools-cn\js\utils.js: the CN site uses https://github.com/tjliqy/unearthed-arcana as its prerelease / UA repository and https://ua.kiwee.top/ as its raw UA root. It uses https://github.com/tjliqy/homebrew/ as its homebrew / partner repository, https://homebrew.kiwee.top/ as its raw homebrew root, and https://gh-proxy.org/https://raw.githubusercontent.com/tjliqy/homebrew-img/main/ for homebrew images. Use these exact links when diagnosing CN 5etools loadextprerelease / loadexthomebrew behavior.
     • CN 5etools online raw roots are useful even though the GitHub repos may not visibly contain the full translated JSON. Verified examples: https://ua.kiwee.top/class/%E7%A0%B4%E8%A7%A3%E5%A5%A5%E7%A7%982025%20-%20%E7%81%B5%E8%83%BD%E4%BD%BF%E6%9B%B4%E6%96%B0.json returns full translated 灵能使 JSON with ENG_name fields; https://homebrew.kiwee.top/collection/Avantris%3B%20%E6%AD%AA%E6%9B%B2%E4%B9%8B%E6%9C%88.json returns full translated 歪曲之月 partner JSON. The root https://homebrew.kiwee.top/index.json is not a valid index and may return 404; do not use that failure as evidence that CN partner resources are missing.
     • For CN online UA / partner file discovery, query generated indexes directly:
       • https://ua.kiwee.top/_generated/index-sources.json
       • https://ua.kiwee.top/_generated/index-props.json
       • https://homebrew.kiwee.top/_generated/index-sources.json
       • https://homebrew.kiwee.top/_generated/index-props.json Use index-sources.json to map source IDs such as XUA2025PsionUpdate, CrookedMoon24, IllriggerRevised, ValdaGunslinger, or Pugilist2024 to exact CN JSON paths, then URL-encode that path and fetch it from the same raw root. Use index-props.json when the source is unknown but the entity category is known. The online CN raw roots can therefore be a better Chinese-text source than local 5etools-cn\prerelease / 5etools-cn\homebrew, whose local index.json files may only be empty development placeholders.
     • When both EN local external JSON and CN online external JSON exist, compare them if the review depends on exact wording. Prefer the CN online JSON for Chinese display text and feature/spell/item names, but use the EN local JSON as a fallback and as a sanity check for source identity, edition, level mapping, and missing fields. If a CN online file is accessible but an individual translation field, entity name, feature entry, spell, item, or rule text is missing, first check DND5e_chm for current 5.5e / 2024 or locally maintained partner translations, then DND2014_legacy for 5e-era / legacy translations. If the CN online file is inaccessible or clearly stale compared with EN, use EN mechanics and mark any Chinese wording as provisional unless confirmed by DND5e_chm, DND2014_legacy, or existing Rankings.
     • Partner / third-party external JSON: 5etools-src\search\index-partnered.json is a search index only. It can locate a partner object's name, hash, source, and page, but it is not the rules text. For the actual partner / third-party source data, open 5etools-homebrew. Its directory layout mirrors 5etools entity properties: class, subclass, spell, feat, item, baseitem, optionalfeature, race, subrace, background, charoption, creature, book, adventure, boon, cult, reward, vehicle, variantrule, table, trap, hazard, and similar folders. Use the category folder matching the rules object, then open the JSON file named by publisher/title. Some partner books are bundled as large multi-property files under collection, such as 5etools-homebrew\collection\Avantris; Crooked Moon.json; if a partner item appears in index-partnered.json but not in its obvious property folder, search collection before concluding the content is missing.
     • 5etools-homebrew\_generated contains the remote-loading indexes used by 5etools homebrew loading: index-sources.json, index-props.json, index-meta.json, index-timestamps.json, and index-adventure-book-ids.json. Use index-sources.json to map a source abbreviation such as crookedmoon24 or pugilist2024 to its JSON file path; use index-props.json to map entity properties to files; use index-meta.json for displayed source names, abbreviations, edition marker, readiness status, and partner marker. In index-meta.json, a file entry with p: 1 is partnered content; if absent, treat it as non-partner third-party/homebrew unless a more specific project rule says otherwise.
     • The 5etools URL fragment loadexthomebrew:a~b~c corresponds to loading external homebrew source IDs through these generated indexes. Do not assume the comma/hash URL itself contains content. Resolve each source ID through 5etools-homebrew\_generated\index-sources.json, then read the referenced JSON file in 5etools-homebrew.
     • UA / prerelease external JSON: loadextprerelease:a~b~c is the prerelease analogue of loadexthomebrew. Resolve each source ID through 5etools-unearthed-arcana\_generated\index-sources.json, then read the referenced JSON file in 5etools-unearthed-arcana. The repository uses the same generated-index pattern: index-sources.json maps source IDs to files, index-props.json maps entity properties to files, index-meta.json stores source display names / abbreviations / edition and readiness metadata, and index-timestamps.json records update timing. Its category directories include class, subclass, spell, feat, item, race, subrace, vehicle, variantrule, and collection. Use collection when a UA article bundles multiple object types.
     • UA release-date fallback: when a UA row/source lacks _meta.sources[].dateReleased, use 5etools-unearthed-arcana\_generated\index-sources.json to map the UA source ID to a JSON file, then use the mapped file's p timestamp in 5etools-unearthed-arcana\_generated\index-timestamps.json as the release date. This fallback is currently for UA / prerelease material only; do not use partner / third-party generated timestamps to force-fill their 未知 dates unless the user explicitly opens that pass. If a UA source is referenced locally but no source file exists, use a narrow manual date only after source-specific verification; current manual case: UATheMysticClass = 2017-03-13.
     • Version precedence for same-name classes/subclasses is mandatory. Use these three cases:
       1. If official published and UA versions are both 2014 / 5e-era, or both 2024 / 5.5e-era, use the latest official published version as the primary review. Do not let a same-era UA replace an official version unless the user explicitly asks for a UA-only or historical comparison review.
       2. If the official published version is 2014 / 5e-era and a separate UA version is 2024 / 5.5e-era, evaluate both as standalone current-relevant options. Keep the 2014 / 5e official version in the unmarked official filename, and create a separate 2024 / 5.5e UA file with the proper （UA） marker. Do not let the 2024 UA erase the 2014 official review, and do not let the 2014 official status suppress the 2024 UA review when the user asks for complete current coverage.
       3. If no official published version exists, use the latest available UA / prerelease version. If both old UA and newer UA exist and no official version exists, use the newest UA unless the user asks to compare historical UA versions.
     • Checklist duplicate cleanup must apply the same version precedence. Remove same-era UA rows when an official counterpart exists, including obvious renamed playtest lineages such as Love Domain / Unity Domain -> Peace Domain, Oath of Heroism -> Oath of Glory, Psionic Soul / Aberrant Mind -> Aberrant Sorcery, Clockwork Soul -> Clockwork Sorcery, The Lurker in the Deep -> The Fathomless, and The Undying Light -> The Celestial. Keep UA-only options with no official successor, and keep 2024 / 5.5e UA versions beside older 2014 / 5e official versions. When the lineage is ambiguous, keep the UA row and flag it for human review rather than deleting it.
     • Partner / third-party version precedence is different from the official-vs-UA exception. If a partner or third-party class/subclass has both 2014 / 5e-era and 2024 / 5.5e-era versions, use the 2024 / 5.5e version as the active review/checklist target and treat the older version as historical unless the user explicitly asks for a legacy comparison. Do not create duplicate 2014+2024 partner/third-party rows merely because both exist. If both partner/third-party versions are from the same rules era, use the newest published version.
   • Local conversion reference: converting-to-srd-5.2.1.pdf is an official SRD 5.1-to-SRD 5.2.1 conversion guide kept as an ignored local reference. Use it as a systematic checklist when converting legacy 5e / SRD 5.1 / 2014 / older partner / older UA material into this project's 5.5e / 2024 / SRD 5.2.1 review environment. Do not quote or cite it in review files as a rules source, and do not commit the PDF. The actual reusable conversion rules must be recorded in this skill, then verified against 5etools-src, 5etools-cn, DND5e_chm, or DND2014_legacy.
   • SRD 5.2.1 conversion workflow:
     1. Apply renamed terms first. Examples that matter for review wording include D20 Test, Heroic Inspiration, Armor Training, Species, Character Origin, Origin Feat, Magic action, Utilize action, Study action, Influence action, Hit Point Dice, Speed, Malnutrition, Dehydration, and Magical Contagions where relevant. Use verified Chinese names from local sources when available.
     2. Check the content type before scoring. Subclasses must be checked against the current class's subclass level structure, class chassis, spell preparation, always-prepared spells, Fighting Style feats, Weapon Mastery access, Epic Boon endpoint, and obsolete subclass chassis patches. Spells must be checked for 2024 action, slot, Ritual, preparation, concentration, component, target, and list changes. Items must be checked for current Utilize / Bonus Action / Magic action wording, rarity/value/crafting/scroll rules, and attunement economy. Monsters must be checked against the 2025 stat-block structure, including Initiative, gear, saving throw presentation, spellcasting action placement, Bonus Actions, Reactions, omitted/replaced stat blocks, and revised CR expectations.
     3. Normalize capitalization / term identity in English source comparisons, but keep review output in Simplified Chinese and follow this skill's Chinese-name rule. Do not mistake a capitalization or name-only change for a mechanical buff or nerf unless the verified local text changes function.
   • SRD 5.2.1 rules-drift audit. When a legacy feature, spell, feat, item, monster, or subsystem depends on any of the following, re-read current local text before scoring: D20 Test; Heroic Inspiration as reroll rather than old Inspiration Advantage; tool + skill proficiency Advantage; Attack action weapon / Unarmed Strike scope, draw/stow per attack, and movement between attacks; Help action proficiency/proximity limits; Hide/Search/Study/Influence action boundaries; Magic action replacing Cast a Spell and interacting with magic items / magical features / Antimagic Field; Utilize replacing Use an Object; Surprise as Initiative Disadvantage; Prone blocked at Speed 0; movement through allied / incapacitated / Tiny spaces; underwater Piercing-weapon rule; saving-throw damage rolling; knockout to 1 HP and Short Rest; Armor Training terminology; background-based ability score increases; Species no longer granting ability scores/languages/alignment; character language choice; multiclass prerequisites/proficiencies being class-specific; AC calculation choice; potion drinking/administering as Bonus Action; spell scroll and crafting rules; Ritual casting without class-specific feature gate; one spell with a spell slot per turn; identifying ongoing spells; Concentration save DC maximum 30; voluntary failed saving throw; Exhaustion / Grappled / Incapacitated / Invisible / Stunned / Opportunity Attack / Unarmed Strike revisions; Emanation and other AoE terms; Bloodied / Ally / Enemy / Attitude / Magical Effect / Target / Teleportation definitions; poison, curse, contagion, travel, trap, vehicle, and magic-item value/crafting revisions when a feature references those systems.
   • Class-structure conversion audit. Before reviewing a legacy class/subclass on a 2024 chassis, check the current class for new or revised baseline features that can replace, obsolete, or recontextualize old subclass text. Current high-impact examples from the conversion guide include: Weapon Mastery and Epic Boon added to many martial/half-martial classes; Bard 魔法奥秘 / spell list / Font changes; Cleric 神圣秩序, 受祝击, 神圣干预, and Greater Divine Intervention changes; Druid 原初秩序, Wild Shape, Wild Companion, Wild Resurgence, Elemental Fury, and Beast Spells changes; Fighter Action Surge / Second Wind / Tactical features; Monk Focus, Stunning Strike, Deflect Attacks, Self-Restoration, and Disciplined Survivor changes; Paladin smite / Channel Divinity / aura / Weapon Mastery / Faithful Steed / Abjure Foes changes; Ranger spellcasting, Hunter's Mark ecosystem, and Weapon Mastery changes; Rogue Weapon Mastery, Cunning Strike, Reliable Talent, and Devious Strikes changes; Sorcerer Innate Sorcery, Sorcery Incarnate, and spell-list changes; Warlock Pact Magic / invocations / Mystic Arcanum / subclass timing changes; Wizard Scholar, Memorize Spell, Arcane Recovery, Spell Mastery, Signature Spells, and spell-list changes. Use current local class text, not this checklist alone, for final scoring.
     • Publish-date audit is mandatory for every same-name or same-role conflict across official, UA, partner, and third-party resources. First search local structured metadata before using web search:
       • 5etools source fields on the entity itself (source, classSource, subclassSource, edition, released, published, date, and similar fields when present).
       • 5etools-src\data\generated / data\class\index.json / source metadata files, when the object is first-party.
       • 5etools-unearthed-arcana\_generated\index-meta.json, index-sources.json, index-timestamps.json, and the referenced UA JSON file for UA / prerelease material.
       • 5etools-homebrew\_generated\index-meta.json, index-sources.json, index-timestamps.json, and the referenced partner / third-party JSON file for external material. Treat source names ending in 24 / 2024 or metadata edition markers as strong local evidence of a 2024 / 5.5e revision, but still open the referenced JSON if the exact mechanics matter.
       • DND5e_chm, DND2014_legacy, existing Rankings, and source filenames / headings as translation and historical-version confirmation.
     • If local metadata cannot establish which same-name resource is latest, and the latest status affects whether an entry is reviewed, browse/search the web for the resource publication date or publisher release page before deciding. Record the date/source in the review reasoning or checklist-generation notes when it resolves a conflict. If neither local metadata nor web search can establish the date, keep the candidate entry only when it is mechanically distinct and mark the uncertainty instead of silently discarding it.
     • Checklist rows must include a Resource release date column. Fill it from 5etools _meta.sources[].dateReleased when available; otherwise use project-maintained manual metadata, source files/headings, DND5e_chm, DND2014_legacy, or web-verified release pages. Use 未知 only when no reliable date is available after that audit. When a row has combined sources, such as Theurgy mappings, list the component dates in source order.
     • Checklist dedupe must normalize same-lineage naming, not merely exact source strings. Labels such as (UA), (Revised), (Revisited), v2, or similar test-version markers do not by themselves create a standalone class/subclass identity. Treat them as historical versions of the same option unless their mechanics are intentionally preserved by a project rule. Example: Artificer, Artificer (Revised), and Artificer (Revisited) are the same class line for checklist selection. Use the latest effective source under the version-precedence rules instead of listing all of them as separate base classes.
   • Do not hard-code "current latest source" conclusions in the skill when a checklist can record them. For each checklist rebuild, verify the latest source through release dates, same-lineage dedupe, and current 5etools metadata; record the chosen source/date in the checklist row rather than relying on an old example in this skill.
   • When both official and UA are reviewed under case 2 or by explicit user request, keep them as separate files with source markers and make the version status explicit in the filename/reasoning.
   • Classes and subclasses: 5etools-src\data\class\class-*.json and 5etools-cn\data\class\class-*.json; each class file contains class, classFeature, subclass, and subclassFeature style data for that class. Use data\class\index.json and data\generated\gendata-subclass-lookup.json to locate source/class/subclass relationships when direct rg is noisy.
   • Species / races: data\races.json plus data\fluff-races.json. 5etools still uses the legacy file/key name race even for 2024 species; do not treat the filename as a rules-era decision.
   • Feats: data\feats.json plus data\fluff-feats.json; feat spell grants, prerequisites, category, repeatability, and 2024 source tags can appear as structured fields.
   • Spells: data\spells\index.json maps source files; individual spell files are data\spells\spells-*.json, such as spells-xphb.json, spells-frhof.json, and spells-egw.json. Use data\generated\gendata-spell-source-lookup.json when checking which classes/lists own a spell, but verify list/preparation rules in the actual class source before scoring.
   • Weapons and weapon mastery: base weapons are in data\items-base.json under baseitem. Weapon entries use item type codes such as melee/ranged weapon and may include a mastery array for 2024 mastery properties. Magic items and special items are in data\items.json; weapon fluff is in data\fluff-items.json. If a mastery property's rules text is needed, search {@itemMastery ...} references and the XPHB book/rules text in 5etools-src\data\book or verify against DND5e_chm first.
   • Optional features / fighting styles / invocations / similar selectable features: data\optionalfeatures.json plus data\fluff-optionalfeatures.json.
   • Background-style character creation options: data\charcreationoptions.json plus data\fluff-charcreationoptions.json; ordinary backgrounds may also be in data\backgrounds.json.
   • When using 5etools JSON, read the actual object's source, page, edition, reprintedAs, _copy, and nested entries/entriesHigherLevel/additionalSpells/optionalfeatureProgression fields. A search hit alone is not enough to decide whether the object is 2014, 2024, UA, prerelease, partner, reprint, or legacy.
   • For every class-wide pass, read DND5e_chm\不全书.wcp and locate that class's topic block in [TOPICS]. Do not assume the existing Rankings files are complete.
   • In 不全书.wcp, class topic blocks can contain ————合作内容分割线————. Entries after that divider and before the next class topic are partner / third-party content for that class unless a more specific local source classification says otherwise. List these entries before deciding scope, filenames, source markers, and leaderboard exclusion.
   • Entries before the divider are official first-party / official legacy / UA / partner items according to their own source tag and path; do not treat all old 5e books as partner merely because they are not 2024 PHB.
   • Prefer the newest verified source when names collide inside the same source category:
     • 2024 PHB / 5.5e / 5R official paths over 2014 official paths.
     • DND5e_chm\其他\新UA before older UA when a UA appears to overwrite old UA content.
     • FRHoF / current partner source over older same-name partner material when it is clearly newer.
   • Do not let a 5.5e / 2024 UA update erase an older official 5e / 2014 published subclass. If an old official subclass has a newer 5.5e / 2024 UA update, evaluate both: keep the latest official published version in the unmarked filename, and put the latest UA version in a separate （UA） file.
   • When converting any 5e-era subclass to the 5.5e / 2024 class chassis, including 2014 official subclasses, 5e-era UA subclasses, partner subclasses, and third-party subclasses, treat every subclass feature or subclass spell that the old source grants below level 3 as being gained at level 3 instead. This reflects the 2024 subclass-entry level and prevents Tier 1 over-crediting from old level-1/2 subclass timing. Example: 5E UA 城市领域, 科技魔法, and 机械之灵 do not receive their old level-1/2 subclass package at character level 1 or 2 in this project’s 5.5e compatibility evaluation; those benefits are evaluated as coming online at level 3.
   • For 5e-era UA / legacy-compatible content that references the modern/urban firearm rules, use the CN translated legacy source DND2014_legacy\其他\UA\城市规则.html as the baseline. When that file references 轻型枪械, treat it as firearms (枪械) that belong to 简易远程武器 Simple Ranged Weapons; when it references 重型枪械, treat it as firearms that belong to 军用远程武器 Martial Ranged Weapons. Referencing this file also means all firearm options in 枪械时代 Firearm Eras are enabled for that review context, unless the reviewed source, user prompt, or review file explicitly narrows the assumption. For 牧师-城市领域, 异界宗主：机械之灵 Ghost in the Machine, and 法师-科技魔法, default to enabling 城市规则 and all modern-style spell / technomagic spell content in DND2014_legacy\其他\UA\新法术.html when evaluating their functionality and balance.
   • For 5e / 2014 Cleric subclasses converted to the 5.5e / 2024 Cleric chassis, ignore subclass features that grant 重甲 / Heavy Armor training, 军用武器 / Martial Weapon proficiency, 神圣打击 / Divine Strike, or 强力施法 / Potent Spellcasting when assigning balance/functionality ranks. Those benefits are covered or overwritten by the 2024 Cleric base class through 神圣秩序 and 受祝击; they may be mentioned only as legacy-source notes or design/compatibility issues, not as subclass power gains.
   • For 5e / 2014 Druid subclasses converted to the 5.5e / 2024 Druid chassis, treat old subclass armor-training or bonus-cantrip style chassis patches cautiously. The current Druid base class supplies 原初秩序 Primal Order and other revised Wild Shape / Elemental Fury / spellcasting features, so do not double-credit an old subclass for a benefit now covered by the base class. Mention retained legacy flavor or compatibility friction separately from power scoring.
   • Special 5e-era UA Wizard subclass rule for 神圣奇术 Theurgy: use .codex/skills/theurgy-ua-ranking-review/SKILL.md whenever reviewing or ranking Theurgy or a Theurgy-to-Cleric-domain combination. The effective project wording is stored in Rankings\法师\神圣奇术\README.md; the original UA wording is historical context only. When a new non-homebrew Cleric-domain review is added under root Rankings\牧师, also create or update the corresponding 神圣奇术（5E UA） - [牧师领域] review under Rankings\法师\神圣奇术. Do not map homebrew Cleric domains to Theurgy unless the user explicitly requests it.
   • Partner / third-party 5e-era base classes are not automatically modernized, but apply explicit project compatibility patches when recorded. Current default compatibility patch: 血猎手 Blood Hunter and 邪狱使 Illrigger count as gaining 武器精通 Weapon Mastery at level 1 when reviewed in the 5.5e / 2024 environment. They also use the 5.5e / 2024 属性值提升 Ability Score Improvement structure, replacing a level-19 ASI with 传奇恩惠 Epic Boon. This is a baseline martial-compatibility patch, not a full 2024 redesign; do not invent other modern features unless the local source grants them or this recorded patch explicitly says so. Treat the level-19 传奇恩惠 part as parity with the 2024 endgame feat baseline, not as a unique class advantage: it may prevent these 5e-era classes from being penalized for an obsolete ASI endpoint, but it should not by itself raise a Tier 4 rank. When creating or updating the class-level README.md for 血猎手 or 邪狱使, clearly tell readers that this default compatibility patch is applied, including both level-1 武器精通 and level-19 传奇恩惠, and that 传奇恩惠 is scored as baseline parity rather than an exclusive strength.
   • Compatibility patch for legacy 战斗风格 Fighting Style grants: when a confirmed legacy option grants a named Fighting Style and that Fighting Style has a corresponding 2024 战斗风格专长 Fighting Style Feat in DND5e_chm\玩家手册2024\专长\战斗风格专长.htm, evaluate it as granting the corresponding 2024 feat version instead of the obsolete 2014 rules text. If the named Fighting Style was not rewritten in the 2024 Player's Handbook, keep the local legacy wording and state that it is not modernized. Current confirmed targets: 剑舞学院 College of Swords and 血猎手 Blood Hunter features that grant Fighting Style choices. This patch updates rules compatibility only; it does not grant extra ASIs, extra feat choices, extra Weapon Mastery, or unrelated 2024 class features. Example: a pure 剑舞学院 Bard does not gain 武器精通 Weapon Mastery from this patch. If a review discusses full dual-wielding or mastery synergy for Sword Bard, mark it as a multiclass or external-build note, such as a 1-level dip into Fighter / Paladin / Barbarian / Rogue / Ranger, and do not let that multiclass ceiling inflate the pure-subclass rank unless the file is explicitly reviewing that build.
   • Blood Hunter damage-loop calibration: when reviewing 血猎手 Blood Hunter, count 猩红仪式 Crimson Rite as a real per-hit, normally pre-combat, no-concentration damage increment, but verify the current die timing, HP cost, activation requirements, and number of weapons affected. For subclasses that add spells or other riders, recalculate the actual hit loop and slot ceiling from source text instead of importing old Paladin or Warlock comparisons.
   • 5e-era UA base-class compatibility patch for 秘术师 Mystic: use the 5.5e / 2024 属性值提升 Ability Score Improvement endpoint structure by replacing only the level-19 ASI with 传奇恩惠 Epic Boon. Do not grant 武器精通 Weapon Mastery to Mystic unless a future local source or explicit user instruction says so. Treat level-1 Mystic subclass/order features as being gained at level 3 in the 5.5e / 2024 evaluation, matching the project’s legacy-subclass conversion rule. Mystic reviews must analyze the full 灵能灵术与灵能天赋 Psionic Disciplines and Talents system before scoring: read the local discipline/talent source text, identify the best and most common discipline picks for each tier and order, and judge the class by actual psi-point economy, concentration-equivalent focus pressure, action economy, save types, damage/control/defense scaling, and known community optimization. As a community reference, consult Giantitp Mind Over Matter: A Guide to the 5e Mystic (https://forums.giantitp.com/showthread.php?542539-Mind-Over-Matter-A-Guide-to-the-5e-Mystic) when accessible; if direct access is blocked, use search-indexed snippets and other accessible community guides only as secondary calibration, with local source text remaining authoritative.
   • Special published-third-party assumption for 铳士 Gunslinger only: when reviewing the 瓦尔达的秘密尖塔 Gunslinger class or its subclasses, assume 文艺复兴枪械 Renaissance Firearms are enabled by default, because this best aligns with the 5.5e / 2024 PHB firearm baseline. Do not assume 工业时代枪械 Industrial Age Firearms or 现代枪械 Modern Firearms by default; treat those as DM-permission / campaign-tech assumptions and state how they would change the score only in analysis. For Gunslinger only, also treat TCE 枪手 Gunner as an available and normally optimized 通用专长 General Feat by Tier 2 unless the review explicitly describes a feat-starved or no-firearm route. Local TCE text grants firearm proficiency, +1 Dexterity, ignores the firearm 装填 property, and removes disadvantage for ranged attacks within 5 feet of hostile creatures. This means default Renaissance firearms should not be scored as if 装填 permanently suppresses the Gunslinger’s 额外攻击 after the feat comes online, but the feat slot remains a real opportunity cost. This special assumption does not apply to other classes unless the user explicitly says so.
   • When reviewing 铳士 Gunslinger, remember that all Gunslingers have access to the class's common 战技选项 Maneuver Options; subclass-marked maneuvers are additional subclass-specific maneuvers, not the whole maneuver system. Also verify spell access from both DND5e_chm\第三方\瓦尔达的秘密尖塔\铳士\法术.htm and the broader DND5e_chm\第三方\瓦尔达的秘密尖塔\法术.htm when a Gunslinger subclass prepares or learns spells from a normal class list such as 法师 spells.
   • Gunslinger damage-stacking calibration: do not treat 铳士 Gunslinger as if it had Fighter attack scaling. Verify its actual attack count, firearm dice, maneuver/risk mechanics, feat pressure, subclass riders, and any spellcasting subclass slot table before scoring DPR. If a Gunslinger subclass gains Wizard-list rider spells, apply the general per-hit-rider and high-slot-accounting rules above, especially concentration conflict and 1/3-caster slot ceilings; do not reuse old Eldritch Knight or full-caster math without recalculation.
   • Warlock short-rest and pact-unit calibration: do not score 魔契师 Warlock like either a normal prepared full caster or a normal martial. Count the actual 契约魔法 Pact Magic table by tier and by expected short rests: with no short rest the class is slot-starved, with one short rest it can rival or exceed same-tier casters in high-impact slot repetitions, and with two reliable short rests it can spend far more high-level slots than a normal full caster at T1-T2. This remains campaign-pacing sensitive; do not assume infinite short rests or no short rests unless the review says so. Also count 邪魔活力 Fiendish Vigor as a real level-2 defensive floor: it casts 虚假生命 False Life without a spell slot and uses the maximum temporary hit point result, but it is not upcast because no spell slot is spent. Count 链之魔契 Pact of the Chain as more than scouting: it is a cheap no-slot 寻获魔宠 Find Familiar route with special forms, combat/utility bodies, touch-spell delivery, Help/positioning/trigger bait, and sometimes meaningful attacks or special traits. Do not pre-credit later Chain-focused invocations unless the tier and file actually evaluate those invocations, but do not underrate the base pact as a ribbon. Apply the same check to builds that take Warlock in Tier 1, such as The Celestial Generalist, Sorlock, Paladin/Warlock, and Valor Bard/Warlock packages: credit the early Warlock defensive/companion/resource floor, but also charge the delayed main-class subclass, Extra Attack, Aura, Metamagic, Magical Secrets, or spell-level breakpoints.
   • Record enough local context in your own notes before writing: source path, class/subclass name, feature names, feature levels, spell list, and any special action/resource wording.
   • Record the Chinese translation for every official / partner name you plan to mention. If the local source only exposes an English name or garbled text prevents reliable Chinese extraction, state that the Chinese translation could not be localized and keep the English name. This does not require translating common D&D community abbreviations; keep familiar abbreviations when they are clearer for CN readers.
   • If Get-Content, Select-String, or rg displays mojibake for Chinese source text, do not guess the translation. Re-read the file with the correct local encoding, commonly GBK/CP936 for this repository, or inspect another local index/list page. Example PowerShell pattern: [System.Text.Encoding]::GetEncoding(936).GetString([System.IO.File]::ReadAllBytes($path)). Only write the Chinese name after this check.
   • Verify the class/subclass Chinese name before choosing the output directory, filename, and top-level heading. The filename must be [verified subclass Chinese name].md for first-party official content, or [verified subclass Chinese name]（source marker）.md for UA / partner / third-party content; class summaries must stay as README.md inside the verified class-name directory. If a previous file has the wrong Chinese name or lacks a required source marker, move/rename it rather than creating a duplicate.

4. Version-lock the review.

   • Only analyze features present in the verified source text.
   • If a previous review file contains old-version features, remove or correct them.
   • If a feature cannot be reliably found locally, explicitly say the source could not be localized and keep the analysis conservative. Do not invent feature names.

5. Build the actual rules context before scoring.

   • Identify which core or supplement rules the feature text relies on. Do not treat a feature sentence in isolation when PHB / DMG / MM / XGE / TCE / FRHoF or another local rule changes the real effect. Example: under 2024 每回合仅能施展一道消耗法术位的法术 One Spell with a Spell Slot per Turn, old bonus-action spell limits may be redundant or less meaningful than in 2014.
   • Community calibration for 2024 spell timing: discussions on the 2024 one-slot-spell rule commonly distinguish "spending a spell slot" from casting without a slot through a feature, item, scroll, or other special rule. When ranking action-economy compression, do not automatically apply the old 2014 bonus-action spell restriction. Verify through the source JSON/local text whether each spell actually expends a slot, when the slot is paid, whether a long casting time is already consuming the Magic action and concentration each turn, and whether the character still has the needed action / bonus action / reaction. No-slot or feature-cast access can be valuable, but only after checking uses, recharge, components, concentration, list/preparation legality, and opportunity cost.
   • Spell-level comparison wording for light/darkness, suppression, dispelling, and similar effects must be checked against official local wording before calling a homebrew unclear. 2024 昼明术 Daylight uses the pattern "any 3rd-level or lower spell that created Darkness" (任何三环或更低环阶法术创造的黑暗区域) and does not say "the spell's original/base level." Do not automatically demand that a homebrew write "use the original spell level" when it follows this official pattern. First verify the 2024 rules for casting a spell with a higher-level slot and the specific effect text. Unless a feature explicitly says "original/base spell level" or another local rule changes it, treat a spell-level comparison as referring to the effective level of that spell effect in the current casting context, including upcasting where the general spellcasting rules make the spell a higher-level spell. If the official pattern itself leaves a table-sensitive edge case, record it as an official wording / RAW-boundary issue rather than a mandatory homebrew defect.
   • Open-ended or unbounded-looking wording is an audit trigger, not a conclusion. Do not automatically penalize a feature merely because it has no explicit target cap, compresses action economy, copies spells, loosens the 2024 spell-slot-per-turn rule, stores spells, grants broad ally benefits, or could become scary with future hypothetical resources. First list the strongest currently legal executable routes under verified rules: level and tier, action sequence, resource cost and recharge, trigger control, number of meaningful targets in a normal playable encounter, concentration and component limits, counterplay, official / UA / partner / third-party horizontal baselines, and whether the route needs extreme NPC/army/summon density or permissive DM setup. Distinguish an expanded disclosed scope from true unboundedness: broadening player character ally to ally, or removing a secondary long-rest cap from an effect already gated by spell slots, class resources, reaction, active state, positioning, and trigger timing, does not by itself create open-ended power. Score the actual extra executable cases it adds, such as mounts, summons, hirelings, story NPCs, or extra uses across a long adventuring day. In normal high-tier play, NPC allies and summons should be evaluated by their real table role: whether they add meaningful actions, steal PC spotlight, create bookkeeping, or need protection, not by assuming a sane DM will let them overshadow the party. Evaluate only the resource range that is currently published and enabled for the review; a normal human campaign can decide its allowed official books, UA, partner/third-party sources, adventures, and optional rules. Do not require a feature to be future-proof against all possible unreleased resources. If the danger exists only in future unreleased content, unusual mass-combat assumptions, or an already-table-sensitive official RAW pattern, describe it as boundary risk or publication guidance rather than stable power or a required fix. If the current executable route is actually stronger, cheaper, earlier, broader, or less counterable than ecosystem baselines, then score that honestly and recommend a boundary fix.
   • Encounter/story-specific counters, immunities, and always-on detection require DM-design-space calibration, not a reflexive "OP" or "weak/niche" conclusion. Community encounter-design discussion commonly recognizes that human DMs adjust encounters around the party: sometimes a feature hard-counters a prepared situation as a reward, spotlight scene, or emotional payoff, while later encounters challenge different axes. Ask whether the feature invalidates an occasional scene, a broad combat/narrative family, a campaign theme, a monster ecology, or a secrecy/social/investigation tool; whether it requires adversarial hard counters; and whether it rewards investment or erases play. For immunity-like features, separate raw frequency from reliance: use 5etools monster JSON / spell data to count how often the damage type or condition appears at the reviewed tier, then ask whether it is the main threat, a recharge/signature action, a rider, or an optional spell. Keep official baselines as calibration, not templates: Paladin 勇气灵光 Aura of Courage is an accepted encounter-shaping class payoff against Frightened; late self-only immunity such as Forge Cleric 火与锻造圣徒 Saint of Forge and Fire is not a blanket precedent for party-wide damage immunity auras. For legacy features that grant resistance to "nonmagical" bludgeoning/piercing/slashing, do not silently convert that into full 2024 B/P/S resistance; record it as a legacy-conversion decision unless the reviewed source explicitly modernizes it.
   • Long-history open-ended spells and rules patterns are normal D&D design, not abnormal defects. 祈愿术 Wish has existed as an open-ended wish mechanic since early D&D and has appeared across multiple editions; do not treat a feature as poorly designed merely because it grants, references, or resembles this kind of historically proven open-ended spell. Apply the same principle to other long-running spells/features whose identity is partly DM-adjudicated or table-sensitive. Spell duplication through 祈愿术 is normally resolved by the spell's own text plus tabletop DM adjudication; do not write or demand subclass-specific "Wish rules" that accidentally imply every official 祈愿术 should follow that subclass's interpretation, or that the subclass uses a private replacement version of 祈愿术, unless the source deliberately creates such a replacement. Still verify the exact current local wording and edition version. Criticize only the extra risk introduced by the reviewed option: removing the spell's normal drawback, creating a new no-cost loop, laundering the cost through proxies, changing who pays the stress, bypassing a cooldown, compressing action economy beyond the spell's own verified rules, or combining with version-specific rules in a way that exceeds historical ecosystem baselines. Do not add regulator text merely to satisfy LLM-style anxiety when the official spell pattern itself intentionally leaves DM-facing openness.
   • Do not automatically transfer 祈愿术's historical-design tolerance to a different feature that grants arbitrary spellcasting such as "cast any spell of 5th level or lower." A non-Wish arbitrary spell pool is its own rules object and must be audited separately: spell-list boundary or lack thereof, maximum spell level, action timing, no-slot wording, material/component waiver, concentration, target/range/duration inheritance, excluded triggers such as smite-like follow-up spells, public lookup burden, and whether a fixed spell list or narrower class-list restriction would improve public readability. This audit is not an automatic penalty; a feature can remain strong but acceptable when its level, cooldown, resource gate, state-ending cost, and inherited spell limits are real.
   • Read the base class again when reviewing a subclass. Record the exact resource amount, recharge timing, and feature conflicts for class resources such as 动作如潮, 回气, Channel Divinity, Rage, Focus/Ki, Bardic Inspiration, Wild Shape, spell slots, pact slots, prepared spells, and similar class currencies.
   • Multiple options that spend the same class resource are not a design flaw by themselves. Paladin subclasses, especially legacy / 2014-era oaths, often have two Channel Divinity options at the same subclass level, and current reviews should not penalize that pattern merely for existing. Audit the actual relationship instead: whether the options are cleanly separated, compete for scarce uses in a meaningful way, stack into a refund / loop / action-compression engine, or make the feature hard to run because a core state and a resource-conversion trigger both pull from the same class currency.
   • Action-economy conflict can affect design scores, but only when the conflict materially harms the intended play pattern. Meaningful choices are healthy design when each option is clear, separately valuable, and the player understands the tradeoff. Lower 设计质量 or 能力设计质量 when the reviewed option's own core features repeatedly block each other, require too many setup turns before the subclass feels online, hide a mandatory sequence behind several Bonus Actions / Reactions / Concentration gates, or make a granted feature frequently unusable because the same action/resource is already claimed by the class chassis. For always-prepared identity spells, separate thematic fit from routine-use expectation: a spell can strongly define the subclass fantasy while remaining a situational button because it costs a Bonus Action, spell slot, concentration, or a competing class resource. Record whether the conflict is purposeful tension, normal opportunity cost, or accidental self-sabotage; do not penalize merely because a powerful or flavorful feature has a real action cost.
   • Separate resource ceiling from practical throughput. A feature that looks extreme may be limited by short-rest/long-rest recharge, action cost, concentration, spell slots, prepared spell count, limited uses, saving throws, positioning, or competition with the class’s normal best turn.
   • Do not overrate post-roll or post-failure correction by treating it as automatically stronger than a constant bonus. For a single attack roll, a feature that adds +X only after a miss is mathematically equivalent to a constant +X to that same roll, assuming both follow the same natural-1 / automatic-miss and impossible-AC boundaries. Its practical value is salience and tactical targeting: it is not visibly wasted on rolls that already hit and may be saved for a later attack in the same turn if the earlier attack hits. The real comparison must count coverage and throughput: once-per-turn post-miss correction affects at most one attack, while a constant bonus that applies to every attack can scale with Extra Attack, bonus-action attacks, reaction attacks, and multiattack routines. Extra Attack can make a once-per-turn correction more likely to find a near-miss window, but it does not multiply the correction across every attack; do not write that multiattack "amplifies" or "scales" the correction unless the reason explicitly states the once-per-turn cap and the actual expected-hit opportunity being counted. When comparing features like a Paladin oath's miss correction against Sacred Weapon-style constant bonuses, record activation action, duration, end conditions, number of attacks covered, weapon / melee / ranged / unarmed restrictions, and whether the correction can target a high-value rider attack. Do not accept conservative "post-failure is dangerous" reasoning unless the actual expected-hit gain, coverage, resource cost, and action economy exceed ecosystem baselines.
   • Treat official same-pattern post-roll / post-failure timing as clear enough. If a current 2024 feature uses a structure like "when the creature fails a D20 Test / 当其在一次 D20 检定中失败时" plus "add a die or modifier to that roll/check, possibly turning failure into success," then a reviewed feature using the same structure should be read as a post-failure trigger. Do not require extra wording such as "after you know the check failed" unless the feature lacks a failure trigger, changes who chooses to spend the resource, creates a reaction/timing conflict, or needs to distinguish "after the roll but before the result" from "after failure." Example baseline: 2024 诗人激励 Bardic Inspiration uses the failed-D20-Test plus possible-success structure.
   • Treat concentration as a real resource, not a parenthetical cost. For any feature or spell that requires concentration, ask what it is replacing at that level and whether it is worth concentrating on compared with the class's best spells/features. Example: 重力魔法 调节密度 and 事件视界 have real tactical value, but they compete with Wizard concentration staples; do not score them as if they can run alongside the same caster's best control, summon, wall, or defense concentration spell. Conversely, no-concentration features that duplicate a concentration spell may be stronger than they first look.
   • Community calibration for concentration-bound class identity: the 2024 Ranger discussion repeatedly centers on 猎人印记 consuming concentration and bonus actions while several other Ranger combat spells have lost concentration. When a class feature is tied to a concentration spell, compare the whole package against the class's actual no-concentration alternatives and its best concentration alternatives. A feature can be strong as an attrition/default mode while still being bad design if it makes higher-level spell slots, subclass commands, or capstones feel locked behind a low-level concentration effect. If a Ranger subclass instead spends 宿敌 Favored Enemy uses as an alternate resource, audit it separately: this usually avoids the concentration lock of 猎人印记, but still competes with the free 猎人印记 uses, Bonus Action setup/transfer, and other Bonus Action plans such as dual-wielding, companion commands, or smite-style spells. Do not penalize this automatically; score whether the alternate use creates a clearer subclass engine, an awkward split resource, or both.
   • Be especially vigilant with melee / short-range / multi-hit damage-increment resources. This is a general mechanism check, not a closed list of spell names. Whenever an option adds damage or riders per hit, per attack, per ray, per projectile, per weapon hit, per spell attack, per creature hit, per turn with many triggers, through a short-range aura, or through a companion/summon/partner attack loop, calculate the actual throughput before changing ranks. Verify at least: who receives the rider, what attack types qualify, whether it triggers once per turn or on every hit, casting or activation action, bonus action / reaction pressure, concentration, range, duration, target access, pre-cast feasibility, spell-slot or class-resource cost, scaling, and whether the build can realistically produce enough hits while surviving at that range. Do not treat these resources as limited to the examples below; apply the same audit to spells, feats, magic items, class features, subclass features, maneuvers, companions, summons, partner / third-party rules, and homebrew analogues.
   • For multi-hit DPR math, write the number of expected hit opportunities before writing the rank. Check amplifiers such as 魔能爆 beams, 灼热射线 rays, 吉姆的魔法飞弹 projectiles, 额外攻击, 动作如潮, dual-wield / bonus-action attacks, reaction attacks, companion attacks, summon attacks, and any third-party multiattack loop. Also check whether the rider applies to spell attacks, weapon attacks, melee attacks, ranged attacks, or only attacks against creatures within a short range. If a class/subclass/build has access to 魂灵环绕 Spirit Shroud, 咒唤微元素群 Conjure Minor Elementals, or a similar per-hit rider and also has any meaningful multi-hit route, explicitly evaluate that interaction. This is mandatory, not optional color text: count weapon hits, spell-attack hits, stored/released spell-hit events, bonus-action hits, reaction hits, familiar/companion hits if the rider applies, and whether the relevant attacks occur after the rider is active. Then compare that DPR line against the same character's normal concentration/control/support uses, spell-slot limits, setup tempo, range requirement, and survivability. Do not assume the rider package is available merely because the spell is on a list, and do not ignore it when the subclass itself creates extra hit events or safer short-range delivery. High theoretical DPR should not automatically raise 综合评分: the review must still account for accuracy, defenses, concentration protection, enemy reach, flight, teleportation, resistance/immunity, action setup, and whether the character gives up stronger control/support concentration.
   • Conditional extra attack opportunities are not a separate damage aspect. Treat them as a DPR correction layer inside existing aspects. If the option grants the reviewed character an extra attack opportunity, score it under the relevant damage aspect (近战伤害者, 远程伤害者, or 法术伤害者) as a conditional upper edge rather than as baseline DPR. If it grants an ally or the party the attack opportunity, score the main value under 团队增益, then note in the relevant damage aspect and 综合评分 reason how much team DPR it can convert into. Always record: trigger frequency, who controls the trigger, whether enemy behavior can avoid it, reaction / Bonus Action / resource competition, range and positioning, whether it can carry per-hit riders, whether it can enable high-value ally attacks such as off-turn Sneak Attack or heavy weapon bursts, and whether the feature still has deterrence value when it does not trigger. Only adjust a damage or overall rank when the trigger is common or controllable enough, reaction competition is acceptable, and the added attack can reliably carry meaningful damage; otherwise describe it as conditional threat or deterrence rather than stable DPR.
   • Dual-wielding is a concrete 2024 hit-count amplifier and must be counted precisely. Local 迅击 Nick says that when you make the extra attack provided by the 轻型 Light property, you can make that attack as part of the 攻击 action instead of as a Bonus Action, but you still make only one extra attack from the 轻型 property each turn. Local 双持客 Dual Wielder says that after you take the Attack action and attack with a Light weapon, you can use a Bonus Action later that turn to make one extra attack with another melee weapon that lacks the Two-Handed property; this 双持客 attack is not the same attack as the 轻型 property attack. Therefore a normal Extra Attack character can commonly reach four weapon-hit opportunities in a turn after setup: two attacks from 额外攻击, one 迅击 / Light-property off-hand attack inside the Attack action, and one 双持客 Bonus Action attack. This is most valuable for two-attack martial or half-caster builds that have meaningful every-hit riders. Fighter-style attack scaling changes the comparison: because Fighters have more attacks and more feats, 长柄武器大师 + 巨武器大师 can often beat dual-wielding through stable proficiency-bonus damage per hit, bonus-action attack access, and better feat synergy. Fighter 11/20 can still use dual-wielding for five/six attacks before 动作如潮, and 动作如潮 can add another full Attack action but not another 迅击 or 双持客 attack beyond their once-per-turn / Bonus Action limits. When scoring this route, check the feat cost, weapon restrictions, draw/stow logistics, whether a Fighting Style or class feature adds the ability modifier to off-hand damage, competing 长柄武器大师 / 巨武器大师 damage, and whether the Bonus Action is already needed by Rage, Hunter's Mark transfer, Cunning Action, Smite casting, Channel Divinity, companion commands, or subclass features.
   • Paladin dual-wield calibration must be conditional, not a class-wide default. The main value is not the off-hand weapon die by itself; it is turning every-hit riders into slot-efficient sustained damage. At level 5+, 迅击 + 双持客 can give a Paladin four hit opportunities after setup, and 神恩 Divine Favor adds 1d4 radiant per weapon hit without concentration. This includes ranged weapon attacks, so 神恩 improves bow/crossbow/thrown backup turns as well as melee turns. Re-read the 2024 战斗风格 wording before judging ranged Paladin: local Fighter / Paladin / Ranger class text grants a 战斗风格专长, and local 箭术 Archery is a 战斗风格专长, so Paladins can legally choose 箭术. However, do not overrate this as a Fighter/Ranger-equivalent ranged-DPR package: 箭术 competes with 防御, 对决, 巨武器战斗, 防护 / 援护, and 受祝福的勇士; 至圣斩 is melee weapon / unarmed only; 光耀打击 applies to melee weapon attacks / unarmed strikes; and ranged turns still compete with the class's aura-frontline job. Ordinary Paladins are also feat-starved: 双持客 competes with ability scores, 巨武器大师, 长柄武器大师, defensive feats, and species / origin-feat plans, while 妖精触碰 for 猎人印记 or 脆弱诅咒 is often not a viable default. Therefore do not credit a Paladin or Paladin subclass with the full dual-wield rider package unless the local spell list, subclass, setting feat, or build route gives a real per-hit rider with acceptable opportunity cost. Clear positive cases include 复仇之誓 with oath 猎人印记, 惩戒之誓（歪曲之月） when the same resource's 教团新进者 Cult Initiate origin feat is chosen for 脆弱诅咒 instead of another origin-feat package such as 魔法学徒 for 护盾术 / 吸收元素, and 巨灵贵族之誓 after it gains 魂灵环绕 at level 9 and especially 咒唤微元素群 at level 13. For other Paladins, sword-and-shield, two-handed 巨武器大师, or 长柄武器大师 plans may be more realistic; do not raise their melee, ranged, or overall ranks merely because the class could theoretically spend scarce feats on rider support.
   • Paladin smite comparison must separate sustained rider math from one-hit burst, rider effects, and Bonus Action pressure. Local 2024 至圣斩 Divine Smite is a 1st-level Paladin spell with casting time Bonus Action, taken immediately after you hit a creature with a melee weapon or an Unarmed Strike; it adds 2d8 at 1st level and +1d8 per higher slot, with another +1d8 against Fiends or Undead. It remains valuable for immediate burst and for spending a slot only after a hit, but it is not the 2014 no-action class feature and cannot be treated as free action economy. When a class/subclass feature triggers from casting 至圣斩, score the trigger together with the Bonus Action cost and its conflict with 神恩, 猎人印记, 脆弱诅咒, dual-wield Bonus Action attacks, Channel Divinity activation, subclass commands, or other Bonus Action plans. Local 炽焰斩 Searing Smite starts at 1d6 on the hit and 1d6 at the start of each target turn until it ends, and all of that damage increases by 1d6 per higher slot; its value depends on target Constitution saves and how many upkeep ticks occur. Smite spells such as 雷鸣斩, 愤怒斩, 致盲斩, 震慑斩, and 放逐斩 should be scored for their riders as well as damage. Dual-wield rider stacking is mainly a substitute for low-level damage-only slot spending and a sustained-damage plan when the Bonus Action is free; it is not a blanket replacement for control smites, high-level smites, or emergency burst.
   • Paladin features that trigger from 至圣斩 Divine Smite or other smite spells must be scored with the current 2024 cost stack, not as free follow-up riders. Check the hit requirement, Bonus Action timing, spell-slot expenditure, one-slot-spell-per-turn limit, Channel Divinity or subclass-resource competition, and whether the Paladin first has no Heroic Inspiration / no mark / no existing state if the trigger depends on that. A repeatable trigger can still be fair when these costs are real; conversely, if a feature removes several of these costs or refunds them, treat that as a major action/resource-economy gain.
   • Paladin concentration correction: local 2024 神恩 Divine Favor does not require concentration, and most 2024 Paladin smite spells also do not require concentration. Do not write or score them as if they automatically compete with 祝福术, 妖火, 催眠图纹, 拂晓, 魂灵环绕, or other concentration plans. Verify each smite locally before scoring; current 2024 exceptions recorded in the local spell index are 闪耀斩 Shining Smite and 放逐斩 Banishing Smite, which still require concentration. No-concentration does not mean no cost: these spells still compete through Bonus Action casting / trigger timing, spell slots, hit requirements, saving throws, duration, target access, and the one-spell-with-a-slot-per-turn rule.
   • Track setup tempo for stacked riders. Under the 2024 每回合仅能施展一道消耗法术位的法术 rule, a character usually cannot cast 咒唤微元素群 and 神恩 in the same turn if both expend spell slots. If one of the spells is cast without a slot, pre-cast before initiative, or otherwise paid by a feature/item that avoids slot expenditure, the combined Action + Bonus Action setup can still consume a whole turn before normal attacks. Otherwise the full stack takes at least two turns. 魂灵环绕 plus 神恩 also normally takes two Bonus Actions across two turns; the character can still use the Action to attack during those turns, but the fully stacked state arrives later. This matters for ranking: a one-round boss fight, a mobile enemy, or a concentration break can make the theoretical dual-wield ceiling irrelevant, while a long melee attrition fight can make the same package extremely efficient.
   • 法术反制 Counterspell trigger standard: verify the local 2024 reaction condition before deciding whether a spell release / feature-cast spell is counterspellable. The current local trigger is a Reaction when you see a creature within 60 feet casting a spell that has a Verbal, Somatic, or Material component (当你看见60尺内一名生物施展一道具有言语、姿势或材料成分的法术时). If any one of those components is present in the current casting / release context and the caster is seen within range, the spell can satisfy 法术反制's trigger. If a feature explicitly removes all observable V/S/M components, hides the casting, or says the release is not casting, then evaluate that specific wording; otherwise do not treat no-slot casting, stored-spell release, or attack-linked casting as automatically uncounterspellable.
   • Stored-spell timing and caster-identity standard: when a feature or item uses 储存法术 / stored-spell wording comparable to 储法戒指 Ring of Spell Storing or 奥法暂滞 Arcane Abeyance, default to this project reading unless the local source explicitly says otherwise: the spell slot and any paid casting resources are expended when the spell is stored, and the later release/casting of that stored spell expends no spell slot, no matter who releases or casts it. Separate resource payment from caster identity. For 储法戒指, the wearer casts the stored spell, but the spell uses the original caster's slot level, spell save DC, spell attack bonus, and spellcasting ability unless the item text differs. For 奥法暂滞, the creature holding the bead releases/casts the spell and is treated as the caster for other purposes, while the spell uses the Chronurgy Wizard's spell attack bonus and save DC. Therefore, when reviewing any stored-spell feature, identify all of these separately: who paid the spell slot/components at storage time, who takes the release/casting action, whose DC/attack/spellcasting ability is used, who concentrates, whose position/range/line of effect is used, whether the release is observable/counterspellable, and whether the source says the releasing creature is the caster "for all other purposes." This matters for 2024 spell-timing review: the release turn is a no-slot casting/release for 每回合仅能施展一道消耗法术位的法术, but the stored spell is not free because the slot was already paid at storage time and may be wasted if the storage expires or is replaced. Do not apply this standard to ordinary "cast a spell as part of an attack/action" wording unless the text actually creates a stored spell.
   • Stored attack-spell replacement and per-hit riders: if a stored-spell or delayed-release feature says a stored spell that originally requires a melee/ranged spell attack uses a weapon attack's hit result instead of making its own spell attack roll, count that release as an additional spell-hit event for damage riders and on-hit consequences, but do not count it as another Attack-action weapon hit unless the feature says so. Check the original spell's range, components, concentration, target, damage scaling, and whether it is a cantrip or leveled spell. Example pattern: a stored 2024 冻寒之触 Chill Touch is a touch cantrip with a melee spell attack; if a later dagger hit releases it and the feature says the spell attack uses the dagger attack result, the spell can effectively hit without a separate spell attack roll. If 咒唤微元素群 Conjure Minor Elementals is already active and the target is within its Emanation, that spell-hit can receive the CME per-hit rider because CME keys off attack rolls the caster makes that hit creatures in the Emanation. This is an actual DPR / hit-count interaction, separate from whether the stored spell consumed a spell slot earlier or whether another leveled spell is cast later in the same turn. Also check resource substitution, not just action compression: if cantrips, weapon hits, stored cantrips, or other no-slot hit carriers can replace a high-level follow-up attack spell under a high-slot rider, the build may save major spell slots while keeping a large portion of the burst. Example: a normal high-end Wizard CME nova may spend one 9th-level slot on 咒唤微元素群 and another high slot on 灼热射线; a subclass that turns cantrips and weapon hits into extra CME carrier hits may still spend the 9th-level CME slot but avoid the 8th/7th-level follow-up ray slot. Score that as real resource efficiency if the positioning, concentration, hit chance, and setup are credible.
   • Per-hit rider access should be verified from spell-list / subclass-list JSON each time instead of memorized from examples. Critical recurring guardrails: do not treat a feat that grants only a 1st-level spell, such as 妖精触碰 Fey-Touched, as access to a 3rd-level rider such as 魂灵环绕 Spirit Shroud; do not assume a dip learns spells above the dip's actual class level; and do not rank a class as if every expanded-list spell is prepared, known, or worth the concentration. If an old example matters, re-open the spell and class/subclass JSON before using it.
   • Wizard-specific close-range rider calibration: 魂灵环绕, 咒唤微元素群, and similar per-hit riders become much more credible when the Wizard subclass can actually survive near the target and keep concentration. Count this more favorably for 剑咏者, 战争魔法, 防护师, and other subclasses with real AC, reaction defense, warding, saving-throw, or concentration-stability improvements. For ordinary fragile Wizard subclasses, short-range ray/rider bursts remain available but should be described as high-risk nova lines rather than default DPR. Do not score every Wizard subclass as if it safely stands within 10-15 feet for multiple rounds.
   • High-slot rider-slot accounting: when calculating 咒唤微元素群 / 魂灵环绕 plus multi-hit spells such as 灼热射线, 焰火迸射, 吉姆的魔法飞弹, 魔能爆, or similar attack sequences, track exactly which spell slot is spent on the concentration rider and which spell slot remains for the follow-up multi-hit spell. Do not assume both the rider and the ray spell are upcast with the same top-end slot. Example: if a Wizard spends an 8th-level slot on 咒唤微元素群, the later 灼热射线 must usually use a different lower or remaining slot; if the same character spends an 8th-level slot on 魂灵环绕, the same caution applies in reverse, because the high slot is now locked into the rider and not available for the ray spell. This matters most for limited 6th-9th-level slots, pact-slot ceilings, half-casters, and 1/3 casters. Count the burst only after checking actual slot tables, action timing, concentration, range, and whether the high-level slot would normally be better spent on a true high-level spell.
   • For 咒唤微元素群 Conjure Minor Elementals, 魂灵环绕 Spirit Shroud, 繁彩球 Chromatic Orb, 灼热射线, and similar DPR-sensitive spells, verify the current spell level, action, range, concentration, slot scaling, target cap, and class/subclass access from JSON/local text before using old calculations. Keep only reusable math principles in the skill: per-hit riders scale with actual qualifying hit count; high-slot riders consume the high slot and leave different slots for follow-up multi-hit spells; 1/3 casters and half-casters must be checked against their actual slot tables; attack-roll chain/AOE spells need both bounce/target mechanics and hit chance; and any special build case should be recalculated through dpr or a local scratch calculation rather than copied from an old example.
   • Movement-triggered damage and hazard calibration: for spells or features that trigger when an aura/emanation/area enters a creature's space, when a creature enters the area, when a creature moves within the area, or when it ends its turn there, read the exact 2024 wording before scoring. Check whether the area is centered on the caster, whether it moves with the caster, whether the caster can see the target, whether the trigger is once per turn / once per round / once per creature / no cap, whether forced movement counts, and whether the effect is concentration. Local 2024 咒唤林地卫士 Conjure Woodland Beings is a self-centered 10-foot 光环 Emanation, Action, concentration up to 10 minutes, 5d8 force damage plus 1d8 per slot above 4th, and the creature makes the save only once per turn. Moving the caster can make the Emanation enter enemies' spaces, and different creatures can be tagged on the same turn, but a single creature is not hit repeatedly on the same turn. Evaluate Moon Druid and other mobile/durable casters by actual movement, Disengage access, Wild Shape / mount / teleport options, concentration protection, enemy reach, opportunity attacks, line of sight, and whether the character gives up stronger concentration spells.
   • Table-operation burden is a real design axis, separate from wording density. Community play discussion commonly tolerates one-time AoE saves such as a Fireball-style resolution, and can tolerate ongoing areas when the area is small, visually easy to track, or limited to a few targets. The burden rises sharply when a feature creates a large or multi-origin aura/emanation, repeatedly forces every affected enemy to save on its own turn or at round boundaries, applies a short-duration condition that must be rechecked each turn, or requires the DM to remeasure many creatures every round. Audit the expected number of creatures, radius, origin count, trigger timing, save frequency, state duration, condition severity, whether the effect is instantaneous or persistent, whether it can be represented with a clear template, and whether it stacks with normal monster turns. High table workload can lower 能力设计质量 or 设计质量 even when the mechanic is balanced and clearly worded; conversely, replacing repeated mass checks with a persistent state, fixed aura benefit, once-per-round trigger, or smaller/clearer area can improve design quality without changing raw power much.
   • 荆棘丛生 Spike Growth forced-movement calibration: local 2024 text deals 2d4 piercing for every 5 feet a creature moves into or within the area and does not restrict the movement to voluntary movement. Community discussion commonly treats dragging, pushing, pulling, 推斥魔爆, 荆棘鞭, 心灵遥控, grappling, Swarmkeeper movement, and similar forced movement as valid damage triggers unless a table house-rules otherwise. Credit this strongly only when the class/subclass/build can both access 荆棘丛生 and reliably create forced movement or lock enemies in the terrain; if it only needs teammates to move targets, write it as a team-combo value rather than raising the solo rank. Count difficult terrain, camouflage, concentration, friendly positioning, creature size, flying/teleporting enemies, resistance to piercing, and map chokepoints.
   • For class-native lower-ceiling rider spells such as 月光涌泉 Fount of Moonlight, check the exact class/subclass access and hit engine before comparing them to high-slot riders. A lower-DPR spell can still matter when it is easier to access, uses the class's natural action loop, or adds defensive/control riders. Do not turn one Bard/Druid subclass example into a generic rule; re-read the feature that creates the extra hit or melee loop.
   • Compare these resources against 加速术 Haste carefully. If the same character casts both, they usually compete for concentration; if an ally casts 加速术, it is not paid from the same concentration budget and should not be used to erase the value of the character's own rider spell. 加速术 mainly grants one extra attack plus defense/mobility, while per-hit riders can exceed it when the character has many hit opportunities, especially through beams, rays, projectiles, 动作如潮, or bonus/reaction attacks. Conversely, per-hit riders can underperform in short fights, against mobile or ranged enemies, when the setup action loses a critical round, or when the character cannot remain within the required 10-15 feet.
   • Evaluate unwanted interruption consequences. If a feature can end early because concentration breaks, the caster is incapacitated, the caster dies, range/line/position is lost, a required per-turn action is missed, a companion disappears, or a maintained effect is dispelled/suppressed, ask what immediately happens to affected creatures, objects, terrain, summons, flight, restraints, shared attunement, temporary forms, or delayed costs. Verify the relevant local rules rather than assuming the effect simply “turns off safely.” Example: if a concentration feature grants allies a fly speed and then concentration breaks while allies are airborne, 2024 飞行 says a creature falls when it loses its fly speed unless it can hover or another effect keeps it aloft, and 坠落 can cause 1d6 bludgeoning damage per 10 feet, up to 20d6, plus prone on landing unless all falling damage is avoided. This risk affects practical strength, action planning, and design wording even if it does not always change the rank.
   • Compare “on par with” designs to the closest official feature or spell before scoring ceiling and design quality. Similar scale, action economy, duration, concentration, spell level, target type, and resource frequency matter. Examples: a 360-foot storm should be compared to 复仇风暴, including that 复仇风暴 is a 9th-level concentration spell; a fire-damage maximization feature should be compared to 塑能师 超限导能, including that 超限导能 has spell-level limits while an unrestricted maximizer could interact with 流星爆. If an option removes an official limiter, raise balance risk and possibly ranking; if it adds tighter frequency/target/turn limits, do not overstate the ceiling.
   • For any feature that maximizes, rerolls, upgrades, copies, doubles, adds, or otherwise manipulates a number of damage dice, count both the number of dice and the die faces. A cap like “12 damage dice” is not the same as “12d6”: 12d6 averages 42 and maximizes to 72, while 12d8 averages 54 and maximizes to 96, 12d10 averages 66 and maximizes to 120, and 12d12 averages 78 and maximizes to 144. If the design is meant to match a specific official benchmark, decide whether the wording actually caps the benchmark or only caps dice count. Example: if a fire maximization option is meant to sit near 塑能师 超限导能 and 5th-level 灼热射线 (six rays at 2d6 each, 12d6 total), “up to 12 fire damage dice” still allows larger die faces from spells, monster actions, magic items, partner/third-party content, or homebrew. In official reviews, evaluate the first-party default environment separately from open-content / partner / third-party / monster-ally assumptions, and mark whether the wording would need an equivalent-d6 cap, a maximum total die-face cap such as 72, a fixed expected-value bonus, or a spell-level/source restriction to preserve the benchmark.
   • Evaluate severe costs by asking what can actually pay or undo them in that tier. Death, disintegration, lost bodies, exhaustion, expensive material components, 9th-level spell slots, once-per-long-rest spell access, and treasure economy all matter. Example: if an option kills and ashes the target after a 24-hour super-buff, verify whether only 完全复生术 or 祈愿术 can restore the target, the required spell level and material component cost, who can cast it, and whether a Tier 4 party can reasonably afford that cost repeatedly. A cost that is payable once for a final boss fight is not the same as a free drawback, and it should still affect design quality and repeated-use ranking.
   • Audit abuse paths, not only intended use. For any feature that targets 友方, 盟友, 自愿生物, 非敌对生物, a controlled creature, an incapacitated creature, a summoned/created creature, or a creature whose attitude can be changed, ask whether the party can turn an enemy into a legal target and then weaponize the effect or downside. Write the possible procedure, required spells/features, saves, concentration, duration, action timing, resource cost, death/revival cost, DM adjudication, and failure points before deciding whether it changes balance or mostly damages design quality. Grade abuse severity by payoff: if the only abuse is protective and does not create significant extra damage, control, forced movement, action denial, instant kill, resource theft, or other offensive payoff against the affected target or its side, worry less and usually treat it as wording cleanup rather than a major balance issue. Example: protecting a controlled enemy is usually much less concerning than using that controlled enemy for maximized damage, lockdown, resource extraction, or no-save death. If a feature grants a 24-hour super-buff and kills/ashes the target when the caster dies, check whether 支配怪物, 支配类人, 暗示术, social coercion, charm effects, or another table-permitted control can make an enemy count as a friendly creature, then trigger the lethal drawback intentionally. Such a path may be unreliable because of target-legality rulings, failed saves, legendary resistance, concentration, timing, caster-death cost, 完全复生术 economy, and DM scrutiny; but if the wording enables it cleanly, it should affect balance and design scores.
   • Treat 暗示术 and 群体暗示术 as real but DM-sensitive abuse tools, not generic domination. Local 2024 暗示术 is a 2nd-level concentration spell lasting up to 8 hours; the suggestion must be achievable, must not involve obvious harm to the target or its allies, requires the target to hear and understand, fails against charm-immune creatures, and ends if the caster or allies damage the target. Community examples of pressure points include forcing an NPC to answer questions truthfully for hours, making a shopkeeper sell goods for 1 copper, asking enemies to leave combat and find harmless work, or telling guards/soldiers to put down weapons and perform another nonharmful task. For scoring, this means:
     • Do not count 暗示术 as reliable mind-control access to every friendly-target exploit; the target must be charmed, the wording must be table-reasonable, and direct self/ally harm is blocked.
     • It can still support abuse when the requested behavior is framed as nonharmful, achievable, and socially plausible, such as "stop fighting and leave", "bring me the key", "cooperate with this ritual", "stand here while we prepare", "lend your help to this ally", or "come with us for questioning". If that behavior makes the target a legal 友方 / 盟友 / 非敌对 target for another feature, record the exact wording and the DM ruling dependency.
     • Economic/social coercion is not free utility. Community discussion commonly treats magical swindling, forced testimony, and coerced bargains as crimes or campaign-level consequences. Rankings may credit the spell's control/social ceiling, but design notes should flag table consent, legal consequences, hostile witnesses, anti-charm law enforcement, 解除魔法, and later retaliation.
   • Include high-level abuse frameworks in that audit, but do not automatically treat every theoretical chain as default play. Typical real abuse appears when a feature's intended limiter is resource ownership, duration, body risk, action economy, or punishment, and another effect moves that limiter elsewhere. Check at least these frameworks when relevant:
     • Summon / created creature plus 异界誓缚 (Planar Binding): 2024 异界誓缚 is available to Bard, Cleric, Druid, Warlock, and Wizard; it targets Celestial, Elemental, Fey, or Fiend creatures, costs a 1000+ GP consumable gem, takes 1 hour, requires the target to remain within 60 feet during casting, and extends the duration of a summoned/created target to match the binding. If a class/subclass summons one of those creature types, ask whether the party can bind it into a 24-hour, 10-day, 30-day, 180-day, or 366-day servant and whether that turns a combat-duration companion into long-term action economy, scouting, spell delivery, damage, or utility. If the subclass only creates a short-lived nonmatching creature type, or if the table cannot hold the target in range through the casting, do not over-credit the abuse.
     • 魔魂壶 (Magic Jar) plus transformation/body access: 2024 魔魂壶 is a Wizard spell. The caster keeps their game statistics but uses the host body's hit points, Hit Dice, Strength, Dexterity, Constitution, speed, and senses. Abuse exists when class/subclass features were balanced around the caster's fragile body, low physical scores, death risk, or limited senses, but the caster can operate from a stronger Humanoid body. Audit target type, protection from 防护善恶 / 防护法阵, Charisma save, jar range, jar destruction, host death, and whether transformation or other body-changing effects create a valid or better host. Treat it as high-risk table tech, not baseline unless the build/class can reliably supply it.
     • 拟像术 (Simulacrum): 2024 拟像术 is a Wizard spell, creates a Beast or Humanoid construct copy with half hit point maximum, no leveling, no short rests or long rests, cannot cast 拟像术, is costly to repair, and a new casting destroys the old simulacrum. Abuse exists when the simulacrum can spend spell slots, class/subclass resources, reactions, once-per-rest features, item charges, or accept backlash/punishment instead of the original character. This can change the evaluation of features whose limiter is “you pay the resource / you take the consequence,” including exhaustion, backlash damage, dangerous spell use, or limited high-impact reactions. Also check whether the feature text says “you,” “a creature,” “a copy,” “a spell you cast,” or otherwise blocks proxy use.
   • 祈愿术 (Wish): 2024 祈愿术 is on the Sorcerer and Wizard lists, and may also be accessed by other characters through verified class features, subclass features, feats, magic items, or build routes. Apply the same 祈愿术 duplication rule to every option that actually casts 祈愿术; do not treat a homebrew, subclass, Bard route, Cleric route, or item route differently unless that access feature adds its own restriction. The verified local text says that when the basic use duplicates another spell, the caster does not need to satisfy any prerequisite required by the target spell, including costly material components, to make that spell take effect (你不需要满足目标法术所需的任何前提条件（包括有价值的材料成分）就可以让该法术生效). In this project, treat "any prerequisite" as including the target spell's own casting time, components, ordinary spell-list access, slot expenditure, and other casting prerequisites unless the duplicating feature or the target spell preserves a specific downstream limit. Non-duplication uses impose real stress, including spellcasting backlash, Strength 3 for 2d4 days, and 33% chance to lose 祈愿术. Abuse exists when that stress is carried by a simulacrum, bound creature, monster, magic item, or other proxy, or when 祈愿术 bypasses the material/time/target restrictions that normally keep another spell from enabling a class/subclass feature. Separate spell-duplication use from open-ended reality reshaping; the former is rules-constrained, the latter is DM-adjudicated and should not define default rankings unless the review explicitly assumes permissive high-level play.
     • Before using "has 祈愿术" or "lacks default 祈愿术" as a high-tier ranking limiter, verify the current local spell list for that class and any class feature that expands spell preparation. Do not rely on memory from older editions or informal community shorthand. In particular, PHB2024 Sorcerer has 祈愿术 on the Sorcerer spell list, so Sorcerer Tier 4 reviews must not cite lack of default 祈愿术 as a weakness. PHB2024 Bard can choose 祈愿术 through 魔法奥秘 once it has 9th-level Bard spell access, so Bard Tier 4 reviews must treat 祈愿术 as a viable locked-in Magical Secrets choice, not as absent. Bard can still be limited by prepared-spell pressure, level-up replacement pace, scroll limitations, lack of Wizard spellbook / ritual / scroll-scribing economy, lack of daily full-list preparation, and subclass increment quality. Sorcerer can still be limited by known/prepared spell pressure, lack of Wizard spellbook / ritual / scroll-scribing economy, lack of daily full-list preparation, and subclass increment quality.
   • Action-economy compression through replicated spells: 2024 神圣干预 lets a Cleric use a Magic action to choose a 5th-level or lower Cleric spell with casting time other than Reaction and cast it as part of that action without a spell slot or material components. 进阶神圣干预 lets that same feature choose 祈愿术; 祈愿术 itself is an action and can duplicate any 8th-level or lower spell without the target spell's prerequisites. Because those prerequisites include the target spell's casting time under this project standard, 祈愿术 can turn many long-casting-time duplicated spells into the action used to cast 祈愿术, unless a more specific local rule says otherwise. Whenever a class/subclass/build relies on a spell's long casting time, material cost, location setup, or downtime footprint as the limiter, check whether 神圣干预, 进阶神圣干预, or 祈愿术 compresses it into an action or bypasses the setup. Examples include, but are not limited to, 守卫刻纹 (1 hour), 治疗祷言 (10 minutes), 圣居 (24 hours), 克隆术 (1 hour plus 120-day maturation remains relevant), and 海啸术 (1 minute). Do not assume every duplicated spell loses all downstream limits: read the replicated spell and preserve any duration, target, maturation, placement, concentration, area, trigger, or post-casting requirement that still applies. Community calibration: public discussion around 2024 神圣干预 often centers on whether this wording truly compresses long casting times, especially for 圣居, 治疗祷言, and 守卫刻纹. The project default is the verified local wording: the selected spell is cast as part of the Magic action. Still, if a rank depends on instant 圣居, action 治疗祷言, free resurrection/restoration, action 守卫刻纹, or action 异界誓缚 setup, write the dependency explicitly and mark it as table-ruling-sensitive. Also check counterplay: 解除魔法, 法术反制 if components/casting are observable and table rules allow it, creature-type limits, range, line of effect, and encounter relocation can reduce the practical ceiling.
   • 守卫刻纹 (Glyph of Warding) spell glyph abuse: 2024 守卫刻纹 can store a prepared spell, trigger it later, and if the stored spell requires concentration, the effect lasts for its maximum duration without anyone concentrating. The glyph is destroyed if the warded object/surface is moved more than 10 feet from the casting location, so the real abuse is not “carry a glyph anywhere” by default; it is prebuilding a fixed buff/trap room, stacking multiple paid glyphs, converting concentration buffs or area spells into no-concentration triggers, and then using teleportation, 半位面, base defenses, or planned encounter locations to bring targets to the glyphs. Audit spell level, upcast slot, 200+ GP diamond dust per glyph, 1-hour casting time unless compressed, trigger wording, target legality, whether the stored spell can target the triggering creature or area, and whether the reviewed feature makes glyph setup easier or more rewarding. Community calibration: 守卫刻纹 + extradimensional storage is commonly discussed as a "buff room", "god box", or demiplane staging exploit. Do not turn that into a blanket S rank for every class that can eventually cast the spells. Treat it as high-level infrastructure that needs downtime, money, spell preparation, slot expenditure, legal triggers, a stable location, and encounter access. It increases the ceiling of prepared full casters and specific builds; it should change a subclass score only when that subclass materially improves the setup, access, trigger reliability, spell storage payoff, or ability to force enemies/allies through the room.
   • 半位面 (Demiplane) storage and staging abuse: 2024 半位面 creates or reconnects to a persistent 30-foot room through a 1-hour door; objects left inside remain there, creatures can remain unless they choose ejection when the door disappears, and the caster can reconnect to a known demiplane. Abuse exists when a party uses it as a portable fixed location for 守卫刻纹 stacks, stored bodies, 克隆术 containers, bound creatures, dangerous objects, prebuilt terrain, or delayed-cost features. This does not automatically make every stored setup combat-reliable: check access action, who knows the demiplane, whether enemies can follow, whether glyphs/objects were ever moved more than allowed, whether creatures can survive/wait there, and whether the encounter gives time to open and trigger the room.
   • 涅斯图魔法灵光 (Nystul's Magic Aura) type/aura manipulation: 2024 local text says the 伪装 effect can make spells and other magical effects treat a willing creature as another creature type, and the 虚假灵光 effect can alter an object's apparent magic aura. Community discussion often brings this up as counterplay or support for 圣居, creature-type triggers, detection, glyph conditions, and some binding/control assumptions. Before crediting it as an exploit, verify target willingness, 24-hour duration, 30-day permanence setup, dispel vulnerability, whether the relevant effect checks actual type or magically perceived type, and whether the table treats this as affecting targeting restrictions or only detection/trigger processing. This is a real rules interaction, but it is table-sensitive enough that it should usually be a caveat or design-risk modifier unless the reviewed option directly enables it.
   • Combined chains: 拟像术 + 祈愿术, 魔魂壶 + transformation, summon + 异界誓缚 + subclass aura/buff, or any combination with magic items, monster allies, or partner/third-party content can magnify a feature far beyond its printed class boundary. For each chain, state who can access each piece, whether one character can self-enable it or needs a party member, what level and material cost it requires, how many saves/concentration checks/DM rulings it needs, what failure does, and whether it is stable enough to affect balance ranking or should mainly affect design risk notes.
   • When applying abuse analysis, separate four environments: (1) default single-class/default party assumptions, (2) ordinary party cooperation with official spells, (3) magic item / monster ally / story reward assumptions, and (4) open partner / third-party / homebrew assumptions. Stable, repeatable abuse in environments 1-2 can change balance rankings. Environment 3-4 abuse usually belongs in caveats or design-quality risk unless the reviewed content itself grants the enabling piece.
   • For every caster or spell-granting subclass/build, check spell-slot progression, prepared/known spell count, preparation change rules, subclass bonus spells, spell-list scope, combat spell value, noncombat spell value, and whether the feature’s delivery mode actually improves the spells worth casting. Do not assume “more spell access” is strong without naming the spells and the use case.
     • When a feature casts a named spell for free, without a spell slot, with a fixed resource, or through a nonstandard delivery mode, verify the spell's actual casting level and scaling text before scoring it. If the feature does not say it is cast at a higher level, expends a higher-level slot, or otherwise scales the spell, default to the spell's base level. Do not credit high-tier scaling from the character's available spell slots unless the feature actually uses those slots or says the spell is upcast. This matters especially for summon spells, multiattack/HP-scaling stat blocks, upcast damage riders, and "free once/for points" spellcasting features: separate the value of the spell's rider or delivery mode from the value of an upcast spell.
   • Verify each class's actual 2024 spell-preparation model before scoring any spell-list expansion. The word "prepared" is not enough by itself: check the class table's prepared-spell count, when that count increases, whether the class can rebuild its list on a long rest, replace only one spell on a long rest, replace only on level-up, prepare from a spellbook, or receives always-prepared spells. A feature that expands "the spells you can prepare" usually expands the menu, not the number of prepared spells, unless it explicitly says otherwise. Do not turn expanded access to a whole class list into "all spells from that list are prepared" without text support.
     • Current Paladin calibration: local 2024 圣武士 prepares a limited number of spells shown in the class table. When that number increases, the Paladin chooses additional spells of levels for which they have spell slots; after each Long Rest, the Paladin can replace one prepared spell with another Paladin spell for which they have spell slots. Other features may grant always-prepared spells that do not count against this number. Therefore a Paladin subclass or homebrew that adds Wizard / Cleric / Druid / Bard / Warlock spells to the Paladin preparation menu should be scored as expanded locked/slow-swapped choices unless it explicitly grants extra prepared spells, all-spells-known preparation, or a different replacement cadence.
     • Subclass spell tables that explicitly say the spells are 始终准备 / always prepared are extra prepared spells and do not count against the class table's prepared-spell limit. Verify the local base-class wording before applying this: PHB2024 local text for 圣武士, 牧师, and 德鲁伊 says that spells another class feature makes always prepared do not count against the number prepared through that class's preparation feature and count as that class's spells. Example: PHB2024 奉献之誓 oath spells say the oath's magic makes you always have the listed spells prepared at the listed Paladin levels, so they are scored as free always-prepared Paladin spells. Do not apply this to features that merely add spells to a preparation menu, spell list, known-spell list, or spellbook, or to text that explicitly requires normal preparation/replacement.
     • When scoring subclass bonus spells, check each listed spell against the base class's existing spell list and preparation model. A bonus spell already on the base class list is still useful if it is always prepared, because it saves a prepared/known slot and may make a situational spell painless to carry; however, it is not new access and should not be credited like a cross-list spell. Separate "new spell access", "free always-prepared copy of an existing class spell", and "already easy to prepare anyway" in the review reason, especially for high-tier half-caster spells whose slot count is already narrow.
     • Apply the same audit to every class rather than assuming a universal model. Bard-style 魔法奥秘, Sorcerer / Warlock known-or-prepared pressure, Wizard spellbook preparation, Cleric / Druid daily preparation, Ranger / Paladin slow replacement, Artificer / third-party preparation, and UA / homebrew spell systems can all produce different practical strength from the same written spell-list expansion.
   • If a caster can change prepared/known spells only when leveling up, treat each spell choice as a long-term build commitment, not a daily toolbox. Identify which spells are actually worth locking at each tier, then explain what is crowded out. Examples of high-pressure locked choices include defense (护盾术, 吸收元素 if locally available), encounter control (催眠图纹, 恐惧术, 缓慢术, 放逐术), anti-caster tools (法术反制, 解除魔法, 沉默术), mobility/escape (迷踪步, 任意门), healing/recovery (治愈真言, 次等复原术, 回生术), summon/companion spells (野兽召唤术, 妖精召唤术, 亡灵召唤术, 龙类召唤术), and battlefield division (蛛网术, 力墙术, 力场墙, 墙-type spells where locally present). A class with level-up-only swaps can still be strong, but its rank must reflect lower day-to-day adaptability and the real opportunity cost of carrying situational spells.
   • Bard-specific example: 2024 吟游诗人 can replace only one prepared spell when gaining a Bard level. 魔法奥秘 expands future preparations and replacement choices to the Bard/Cleric/Druid/Wizard lists, but it does not create daily full-list preparation. At level 10, optimized Bards often need to replace several low-level transitional Bard spells over multiple later levels because cross-list staples such as 法术反制, 解除魔法, 回生术, 野兽召唤术 / 妖精召唤术, 力墙术 / 力场墙, 护盾术, 迷踪步, or another table-specific answer are competing for limited prepared slots and replacement opportunities. At Tier 4, 祈愿术 is a viable and often premium Magical Secrets choice once the Bard has 9th-level spell access, but choosing it further increases spell-choice pressure rather than granting every other high-level Wizard/Cleric/Druid answer at the same time. When ranking Bard or Bard subclasses, credit 魔法奥秘 as a major power spike, but do not score every possible cross-list role as simultaneously online immediately at level 10 or level 17.
   • Community calibration for 2024 魔法奥秘: when evaluating a Bard that starts play above level 10, reconstruct spell choices level by level instead of assuming the character appears with every possible cross-list slot already optimized. Community discussions commonly count one new Magical Secrets-eligible prepared spell when the prepared-spell number increases and one replacement on Bard level-up; this matters for how quickly a Bard can purge low-level transitional spells. Also verify cantrips locally: 魔法奥秘 keys off prepared spells of level 1+ in the Bard table, so do not assume it grants cross-list cantrips unless a subclass/feature says so.
   • Spell scroll interaction matters for caster rankings. Under the 2024 法术卷轴 rule and local 贤者谏言2025, the spell on a scroll must be on a spell list used by the character. For Bards, 魔法奥秘 does not let them read every Wizard/Cleric/Druid scroll. A cross-list spell counts as a Bard spell for scroll use only if the Bard has actually prepared that spell through 魔法奥秘. Example: a Bard can use a 灵体卫士 scroll only if 灵体卫士 is currently prepared through 魔法奥秘; otherwise the Bard cannot understand that scroll. When ranking Bard scroll access, treat scrolls as improving depth for already-locked Magical Secrets choices, not as solving every missing role.
   • EGW 秘迹学法术 Dunamancy Spells are not a normal Wizard / Bard / Cleric / Druid spell-list expansion. Local EGW text says these spells can be used directly by the Wizard subclasses in that chapter (时间魔法 and 重力魔法) and should not simply be added to other casting classes' full spell lists; a DM may award other casters a small number of such spells as campaign rewards. Therefore do not credit ordinary 魔法奥秘 with access to 秘迹学法术 such as 灵敏之赐, 时流刹转, 暗黑星辰, 饕餮虚空, or similar unless another verified local rule explicitly adds them to a usable spell list or the table assumption says the DM awarded them. Conversely, when ranking 时间魔法（EGW） and 重力魔法（EGW）, count their relevant 秘迹学法术 as part of the subclass ecosystem and compare those spells against official Wizard subclass ceilings.
   • When scoring a level-up-only caster, do not give full credit in every spell-related aspect merely because the spell list contains options. Credit only the functions that a reasonable optimized character can afford to lock simultaneously at that tier. If choosing 法术反制 means delaying 催眠图纹, a summon, healing, or mobility, discuss that tradeoff in the relevant aspect and overall reason.
   • Check species, origin feats, general feats, setting feats, and common magic items that realistically interact with the class/subclass at each tier. Verify feat spell-level limits, item activation, attunement, recharge, and opportunity cost from JSON/local text. Do not infer access to higher-level rider spells from a feat or item that grants only a lower-level spell; when a feat-rich route uses a low-level concentration rider, score it as that low-level rider package rather than as a shortcut to later spells.
   • Wizard feat-entry calibration: for ordinary Wizard subclasses, do not assume 妖精触碰 for 猎人印记 / 脆弱诅咒 as a default DPR plan. Wizards are usually feat- and ASI-sensitive because Intelligence, concentration protection, initiative, defensive feats, ritual/spellbook economy, and general survivability compete for the same choices; spending a general feat on a 1st-level concentration damage rider is normally inferior to stronger Wizard concentration spells. A setting-specific exception can be considered when the source itself provides a low-opportunity-cost route. Current recorded example: 神秘学者（歪曲之月） may reasonably consider the same book's 教团新进者 Cult Initiate origin feat for 脆弱诅咒, because it does not consume a general feat and fits that source ecosystem. Even then, score it as a concentration rider that competes with Wizard control, not as free damage.
   • Community calibration for 魔法学徒: 魔法学徒（法师） is widely valued because 护盾术, 寻获魔宠, 剑刃防护, and similar options can patch defenses or utility early, and the 2024 feat lets the 1st-level spell be cast once without a slot and also with any spell slots the character has. Do not treat this as free universal defense. Score the actual opportunity cost against 警觉, 音乐家, 幸运, 强韧, weapon/damage feats, ASI pressure, reaction economy, component/free-hand requirements, and encounter mix. 护盾术 is much less valuable against saving-throw-heavy encounters or when a later feature such as 防御决斗术 covers the same reaction-defense niche; 寻获魔宠 may outperform it in scouting/help-heavy campaigns.
   • Community calibration for 魔法学徒（德鲁伊） / 橡棍术: Charisma/Intelligence/Wisdom-based weapon plans can be real, especially for Paladins, Valor Bards, Eldritch Knights, Bladesingers, or homebrew gishes, but check bonus-action startup, weapon mastery limits, heavy-armor Strength requirements, shield/component/free-hand rules, and general feat stat bonuses. A build becoming SAD is not automatically higher damage or better defense if it delays Great Weapon Master, Polearm Master, War Caster, Defensive Duelist, an ASI, or class-specific action economy.
   • Explain whether a feat/item/species interaction is actually worth the opportunity cost. Compare against common competing choices such as Great Weapon Master, Polearm Master, Magic Initiate, resilient defenses, ASI pressure, and concentration economy.
   • Evaluate action economy explicitly: action, bonus action, reaction, concentration, free/no-action timing, precombat setup, companion command, and whether the feature competes with the class’s normal damage/control/defense loop. A “ribbon” that is free and does not compete with key actions/resources may matter more than its text size suggests; a powerful feature that consumes the same action as the class’s best turn may matter less.
   • Distinguish raw ceiling from practical reliability for multi-step engines. If a feature package needs multiple turns, repeated bonus actions, maintained states, concentration, form changes, aura uptime, stored effects, or other setup before its real payoff, evaluate the earliest realistic payoff round, action conflicts, whether the setup can be interrupted before payoff, whether interruption resets or only delays the line, and whether the reviewed version itself contains protection against that interruption. Interruptible setup lowers expected value and increases counterplay compared with immediate, passive, or protected engines, even when the maximum payoff is unchanged. Do not import protection from another version, later draft, note, or similarly named feature unless it is present in the verified source being reviewed.
   • Short scaling shorthand is not a clarity flaw by itself. Terms such as PB / proficiency bonus, ability modifier, class level, spell level, and highest prepared spell level are all understandable to normal D&D readers when used in their ordinary meaning. The review should only flag them when the chosen scaling changes the actual power curve, multiclass behavior, or table expectation compared with a nearby alternative. For example, PB can be cleaner than "highest-level class spell you can prepare," but it may also unlock higher spell-level comparisons earlier or interact differently with multiclassing; score that mechanical curve honestly rather than calling the abbreviation unclear.
   • Temporary hit point features need amount/scope/timing separation. Do not treat one-time high temporary hit points, small recurring temporary hit points, self-only upkeep, and party-wide recurring engines as the same defensive package. Check the amount, refresh window, target scope, action/resource cost, whether the old temporary hit points are replaced rather than stacked, whether the state can be interrupted, and whether the feature competes with common sources such as 英雄气概 Heroism, 虚假生命 False Life, Twilight-like party THP, Warlock 邪魔活力, or normal healing. Small ability-modifier THP each turn can be good emotional/attrition design without becoming a high-rank defense engine; large party-wide THP or no-action refreshes can change ranks.
   • Check whether the feature's benefit is already available from the class chassis, common spells, 武器精通, feats, species traits, magic items, or widely used party tactics before calling it a major strength. Apply this to every scoring aspect, not only advantage or DPR: damage riders, accuracy, defense, saves, resistance, healing, temporary hit points, mobility, forced movement, prone/restraining, debuff, anti-caster tools, battlefield division, summons/companions, skills, crafting, information gathering, ritual utility, and exploration.
   • For each overlap, ask what is actually left: lower action cost, no-save reliability, no-concentration, no spell slot, no resource, earlier level, longer duration, bigger range, team-wide effect, off-turn/reaction compatibility, better scaling, better target type, fewer counters, or lower opportunity cost. Credit those differences directly; if the feature merely duplicates a common baseline, lower uniqueness/design credit and avoid over-scoring the relevant function.
   • Audit inner feature stacking before finalizing any subclass score. Do not evaluate subclass features only one by one: check whether earlier subclass features create a trigger, stored effect, no-slot release, extra target, extra hit, extra reaction, concentration owner, resource refund, or action window that later subclass features can use in the same turn or adventuring day. Also check how those subclass features stack with the base class chassis, including Extra Attack, class spellcasting, spell preparation, class resources, Weapon Mastery, Fighting Style, Channel Divinity, Rage/Focus-like states, Cunning Action, Bardic Inspiration, Pact Magic / Mystic Arcanum, and Artificer item systems. If a feature is paid in an earlier turn or at storage time and released later without spending a spell slot, it can often coexist with another spell cast with a spell slot in the release turn under the 2024 one-slot-spell-per-turn rule; if both effects spend a spell slot in the same turn, the limit usually matters. Write this explicitly in 详细评价 and 综合评分 when it changes the Tier 3-4 ceiling or creates a late-tier spike. Example pattern: a level-10 stored touch spell that spends the slot at storage time and releases without a slot can stack with a level-14 "cast a spell as part of the Attack action" feature if the latter is the only spell-slot expenditure in that release turn.
   • Be precise with attack wording. 武器攻击 is broader than 近战武器攻击 or 远程武器攻击; if a feature says only 武器攻击, evaluate it as applying to both melee weapon attacks and ranged weapon attacks unless another clause, weapon property, range, target, or trigger narrows it. If the text later requires at least one 近战武器攻击, a melee hit, a creature within reach, a thrown/ranged attack, or a specific weapon mode, apply that narrower trigger exactly. This affects DPR, target access, safety, Counterspell/close-range risk, Weapon Mastery triggers, dual-wield loops, thrown-weapon loops, and whether a feature can be used from range without entering melee.
   • Advantage is a common example, not a special-only case: 烦扰 / Vex, 失衡 / Topple, prone, invisibility, help actions, and spells such as 闪耀斩 can make a cheap advantage feature less unique at later tiers. Still credit the feature if it is cheaper, no-save, no-concentration, no-slot, works before the first hit, applies to more attacks, works at range, or stacks with reaction/off-turn attacks; but do not score it as if no other advantage engine exists.
   • Community calibration for weapon mastery advantage: 烦扰 is easy advantage but usually needs a hit first and is strongest when it can chain on the same target; 失衡 can create prone advantage for melee allies but may penalize ranged allies and usually involves a save. When comparing a subclass advantage engine against weapon mastery, write whether the feature works before the first attack, works for ranged attacks, bypasses saves, helps the whole team, supports off-turn attacks, or avoids weapon restrictions.
   • Evaluate replaceability. If a class identity is built around an effect that can be copied by common feats, spells, or magic items, lower the uniqueness/design value unless later class features meaningfully deepen or protect that niche. Example: treat Hunter's Mark-style damage as more replaceable when 妖精触碰, 教团新进者, 魔咒武器, or 脆弱诅咒 can cover a similar combat role.
   • Community calibration for 猎人印记 vs `

Content truncated for page performance. Open the source repository for the full SKILL.md file.