phaser4-game-qa-performance

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Phaser 4 测试与性能

Phaser 4 Testing and Performance

全局控制接入

Global Control Access

控制面边界：可提议、可审查、可在 Work Item 任务授权或显式批准范围内修改，且必须回到

$phaser4-game-workflow-control

风险门。

本领域可提议验证计划并审查候选，只能在已建立且任务授权有效的 Work Item、A 等级和路径内写测试或证据；结果回到

phaser4-game-workflow-control

审计和状态迁移。测试通过不能覆盖授权、范围、路径、基线或 A4-A6 操作批准失败。

证据优先于主观结论。项目未定义量化指标时只报告实测与决策缺口，不自造通用阈值。

读取 Work Item、质量门、TDD、qa-plan 和当前候选。F0 校验授权与流程，F1 校验规格一致，F2 形成独立 QA 领域结论，F3 绑定实际命令、环境、数据源、文件、判定与未覆盖项；缺少独立审查时标记未验证。
启动服务前检查同一项目、服务类型、模式、端口和健康状态；存在健康实例时复用，不终止归属不明的进程。
视觉任务核对 V0 分流及 V1/V2 条件门。Work Item 指定效果图为还原目标时，QA 按视觉还原核对冻结视觉目标、对比条件、逐状态/逐区域忠实度矩阵和已批准例外；容差内且不改变视觉事实的适配可
```
AUTO
```
，任何可见偏差或实质取舍只请求一次精确确认。专业修复不自动授权改变冻结视觉事实。
适用 V1 时检查视觉目标、信息层级、草图/灰盒、交互、布局、失败恢复与预算。忠实还原只有在没有可见偏差或实质取舍、且工程适配处于项目预定义容差内时记录
```
AUTO
```
决策依据并进入 V2；否则请求一次精确选择，记录
```
USER_DECISION
```
与已批准例外。原子资源绑定当前
```
AUTO
```
或
```
USER_DECISION
```
记录。
普通资产/布局允许还原
```
not-applicable
```
。效果图在
```
v3-ready
```
核对目标、合同回对、带 bounds 的 coverage 和条件编号证据；V3 还必须逐
```
annotation_number/region_id
```
记录显式
```
production_method
```
、
```
delivery_kind
```
、
```
image_generation_required
```
、
```
generation_record_required
```
、
```
substitution_policy
```
与
```
expected_assets
```
。V3 实施包的
```
visualProductionUnits
```
必须与 coverage 一一映射，输出路径、所有权和格式不得冲突。
V4 必须提交
```
production_contract_audit
```
，逐区域核对预期方法与实际方法、交付类型、输出文件、生成/提示词记录和运行时消费；F2 只接受
```
validationMode=MACHINE
```
的当前身份机器验证事实。只有 V5
```
complete/verified
```
才要求运行测量、F3 runtime replay、freshness-bound fidelity cases、实际消费和无未批准替换；任一身份变化拒绝旧证据，V2 唯一人工确认通过后不再重复视觉复核。
同条件截图记录视口、设备像素比、状态、轨迹、语言、随机种子、ROI、实施前定义的项目容差、动态时间采样/稳定帧与遮罩理由。冻结目标视口/状态以完整 viewport 为主证据，ROI、并排、叠加或像素差只作补充；其他视口按布局合同验证视觉意图和关系不变量，生成式、动画和 VFX 不得只靠像素差。未解释或超容差差异、缺参考/候选同条件证据、缺已批准例外或仅凭主观结论一律判定未通过。
UI 读取
```
phaser4-game-ui-layout
```
合同，并按响应式视觉验证在基准、最窄、最宽视口、断点邻值、同宽窄高度、竖/横屏、默认/大字号、默认/最长文案和零/非零安全区采集稳定帧。每个视口记录 viewportRect、canvasRect、逻辑尺寸、四边空隙、背景覆盖、安全区、关键 UI 边界、CSS/物理缩放和 resize 前后变化，验证动态文本、成员显隐、关键动作状态、滚动所有权和 resize/方向重排；设备 DPR 由运行时动态读取并封顶为 1.5，报告记录实际有效值且验证其位于 (0,1.5]。正式报告必须绑定候选 SHA、scene/state、布局合同版本和视觉基线版本；效果图还原还绑定目标 SHA，fidelity case 引用该报告。完整 viewport 缺失只能标记“未验证”；普通测试断言关系不变量，Golden 只在冻结视口、有效 DPR、语言、状态和稳定帧验证精确视觉。
V5 检查结构化集成、玩法所有权、低保真零引用、性能峰值和功能契约，并提交动态证据供 F3。修订候选重跑受影响 F0-F3；仅 A4-A6 重跑 F4。模块边界变化仅有实质取舍时进入 grilling。
共享基础完成后验证代表性 Vite/Phaser 启动、Boot/首场景、资源加载和适用插件；关键玩法流完成后在主目标平台烟测核心循环与失败恢复；只有最终集成/G3/release 执行完整渠道/平台矩阵。局部任务证据不得宣称全平台通过，不得自动发起真机验收。

没有可复现证据时标记“未验证”。V1/V2 条件门必须绑定当前候选；沉默、继续工作或旧决定不能替代所需的

USER_DECISION

，用户选择记录也不能跳过 V2a/V2b 的确定性机器 F2。FIT 只证明等比不溢出，不证明满屏或响应式重排；禁止用构建成功、元素存在或无控制台错误假通过。

可复用自动化脚本位于

scripts/responsive-visual-validation.mjs

，纯计算测试位于

scripts/responsive-visual-validation.test.mjs

。脚本通过 Node ESM 动态导入 Playwright；运行时从设备读取 DPR 并按

dynamic-capped-1.5

封顶，在同一页面通过

setViewportSize

调整视口并保留 resize 语义；有效声明必须为

(0,1.5]

，原始

deviceScaleFactor

允许大于 1.5 但统一解析后记录为 1.5，零、负数、非有限数和字符串在验证前失败。报告输出完整页面截图、运行时实测矩阵和结构化 JSON。

Control Plane Boundaries: Proposals, reviews, and modifications are allowed within the scope of Work Item task authorization or explicit approval, and must return to the

$phaser4-game-workflow-control

risk gate.

This domain can propose verification plans and review candidates, and can only write tests or evidence within established, validly authorized Work Items, Level A and paths; results are returned to

phaser4-game-workflow-control

for audit and state migration. Test passing cannot cover failures in authorization, scope, path, baseline or A4-A6 operation approval.

Evidence takes precedence over subjective conclusions. When the project has not defined quantitative indicators, only report actual measurements and decision gaps, and do not create universal thresholds on your own.

Read Work Item, Quality Gate, TDD, qa-plan and current candidate. F0 verifies authorization and process, F1 verifies specification consistency, F2 forms independent QA domain conclusions, F3 binds actual commands, environment, data sources, files, judgments and uncovered items; mark as unverified if independent review is missing.
Check the same project, service type, mode, port and health status before starting the service; reuse existing healthy instances and do not terminate processes with unknown ownership.
Visual tasks check V0 diversion and V1/V2 condition gates. When the Work Item specifies a render as the reconstruction target, QA checks the frozen visual target, comparison conditions, per-state/per-region fidelity matrix and approved exceptions according to Visual Reconstruction; adaptations within tolerance that do not change visual facts can be marked as
```
AUTO
```
, any visible deviation or substantial trade-off only requests precise confirmation once. Professional fixes do not automatically authorize changes to frozen visual facts.
When V1 applies, check visual targets, information hierarchy, sketches/gray boxes, interactions, layouts, failure recovery and budget. Faithful reconstruction is recorded with
```
AUTO
```
decision basis and proceeds to V2 only when there are no visible deviations or substantial trade-offs, and engineering adaptations are within the project's predefined tolerance; otherwise, request precise selection once, record
```
USER_DECISION
```
and approved exceptions. Atomic resources are bound to the current
```
AUTO
```
or
```
USER_DECISION
```
record.
Common assets/layouts allow
```
not-applicable
```
for reconstruction. Renders in
```
v3-ready
```
check targets, contract alignment, coverage with bounds and condition number evidence; V3 must also record explicit
```
production_method
```
,
```
delivery_kind
```
,
```
image_generation_required
```
,
```
generation_record_required
```
,
```
substitution_policy
```
and
```
expected_assets
```
per
```
annotation_number/region_id
```
. The
```
visualProductionUnits
```
of V3 implementation packages must map one-to-one with coverage, and output paths, ownership and formats must not conflict.
V4 must submit
```
production_contract_audit
```
, checking expected vs actual methods, delivery types, output files, generation/prompt records and runtime consumption per region; F2 only accepts machine-verified facts with
```
validationMode=MACHINE
```
for the current identity. Only V5
```
complete/verified
```
requires runtime measurement, F3 runtime replay, freshness-bound fidelity cases, actual consumption and no unapproved substitutions; any identity change rejects old evidence, and no repeated visual review is required after V2's only manual confirmation passes.
Screenshots under the same conditions record viewport, device pixel ratio, state, trajectory, language, random seed, ROI, project tolerance defined before implementation, dynamic time sampling/stable frames and masking reasons. Frozen target viewport/state takes the complete viewport as primary evidence; ROI, side-by-side, overlay or pixel difference are only supplementary; other viewports verify visual intent and relationship invariants according to layout contracts, and generative content, animations and VFX cannot rely solely on pixel difference. Unexplained or out-of-tolerance differences, lack of reference/candidate evidence under the same conditions, lack of approved exceptions or conclusions based solely on subjectivity are all judged as failed.
UI reads the
```
phaser4-game-ui-layout
```
contract, and collects stable frames at baseline, narrowest, widest viewports, breakpoint adjacent values, same width different heights, portrait/landscape, default/large font sizes, default/longest copy and zero/non-zero safe areas according to Responsive Visual Validation. Each viewport records viewportRect, canvasRect, logical dimensions, gaps on four sides, background coverage, safe area, key UI boundaries, CSS/physical scaling and changes before/after resize; verify dynamic text, member show/hide, key action states, scroll ownership and resize/orientation rearrangement; device DPR is dynamically read at runtime and capped at 1.5, report records the actual valid value and verifies it is within (0,1.5]. Formal reports must be bound to candidate SHA, scene/state, layout contract version and visual baseline version; render reconstruction is also bound to target SHA, and fidelity cases reference this report. Missing complete viewport can only be marked as "unverified"; ordinary tests assert relationship invariants, and Golden only verifies precise visuals in frozen viewport, valid DPR, language, state and stable frames.
V5 checks structured integration, gameplay ownership, low-fidelity zero references, performance peaks and functional contracts, and submits dynamic evidence for F3. Revised candidates re-run affected F0-F3; only A4-A6 re-run F4. Module boundary changes enter grilling only when there are substantial trade-offs.
After shared foundation is completed, verify representative Vite/Phaser startup, Boot/first scene, resource loading and applicable plugins; after key gameplay flow is completed, smoke test core loop and failure recovery on main target platforms; only final integration/G3/release executes full channel/platform matrix. Partial task evidence cannot claim full platform pass, and cannot automatically initiate real-device acceptance.

Mark as "unverified" when there is no reproducible evidence. V1/V2 condition gates must be bound to the current candidate; silence, continued work or old decisions cannot replace the required

USER_DECISION

, and user selection records cannot skip the deterministic machine F2 of V2a/V2b. FIT only proves proportional no overflow, not full screen or responsive rearrangement; it is prohibited to falsely pass with build success, element existence or no console errors.

Reusable automation scripts are located at

scripts/responsive-visual-validation.mjs

, and pure calculation tests are at

scripts/responsive-visual-validation.test.mjs

. The script dynamically imports Playwright via Node ESM; at runtime, it reads DPR from the device and caps it at

dynamic-capped-1.5

, adjusts the viewport via

setViewportSize

on the same page while retaining resize semantics; valid declarations must be

(0,1.5]

, original

deviceScaleFactor

can be greater than 1.5 but is uniformly parsed and recorded as 1.5, and zero, negative numbers, non-finite numbers and strings fail before verification. The report outputs full page screenshots, runtime measurement matrix and structured JSON.