55-决策与变更记录 Template v1.0 | 第13章 示例样稿

目录 ／ 文档-技术白皮书 ／ 55-决策与变更记录 Template v1.0

第13章 示例样稿

I. 章节目标与范围（强制）

• 提供可直接拷贝的高质量样稿，覆盖：完整 DR（人读 + 机读）、含回滚的变体、常见错误对照与可复用片段。
• 全文严格遵循：统一符号与度量、状态机、门禁（gate）、引用口径（卷名 + 版本 + 锚点 S/P/M/I）。

II. 示例 A：统一到达时口径（完整 DR，人读正文）

1. 封面与摘要
   • DR 编号：DR-2025-0216；状态：Review；提出者：Zhang；创建：2025-09-27。
   • 摘要：跨卷 T_arr 表达存在分歧；拟统一为一般口径，保留常量外提作为例外条款。
2. 问题/动因
   现有白皮书与实现中并存 T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell ) 与 T_arr = ( ∫ ( n_eff / c_ref ) d ell )，导致验证口径与回放脚本不一致。
3. 候选方案与证据
   • 方案 A（一般口径）：T_arr = ( ∫ ( n_eff / c_ref ) d ell )。
   • 方案 B（常量外提）：T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell )，仅当段内 c_ref 近似常量。
   • 证据：cmb_set_v3、lens_v1；脚本 arrive_time_check.py@a1b2c3；CI_95% bootstrap。
4. 影响评估（摘要）
   性能：Eval gate_accuracy ≥ 0.99@7d；兼容：回放通过率 compat_rate ≥ 0.995@replay；成本下降 ~10%。
5. 决策与不做清单
   • 决策：采用一般口径；允许在 L2 区域常量外提，需标注例外。
   • 不做：不改动 API v1 消费者；不修改现有数据契约字段名。
6. 回滚设计
   • 触发器：gate_accuracy<0.98@7d 或 compat_rate<0.99@replay。
   • 步骤：切回 release-stable → 恢复快照 → 复原回归 → 观察 24h。
7. 实施与验证（摘要）
   • 环境：dev/stage/prod；flag：feature.tarr.general_form=true。
   • 门禁（硬）：gate_accuracy>=0.99@7d，compat_rate>=0.995@replay。
   • 路径与测度：gamma(ell) 为观测路径，d ell 为弧长微元；check_dim=true。
8. 引用与附注
   见《EFT.WP.Core.Equations v1.1》Ch.2 S20-1；《EFT.WP.Core.Metrology v1.0》Ch.1–3；《EFT.WP.Core.DataSpec v1.0》I30-*。

示例 A：机读清单（YAML；可直接拷贝）

decision_record:

id: "DR-2025-0216"

title: "Unify T_arr to general form with scoped exception"

status: "Review"

requester: { name: "Zhang", role: "Requester" }

created_at: "2025-09-27"

motive: "Inconsistent T_arr expressions hinder verification and replay."

options:

- name: "A - general"

assumptions: []

pros: ["handles spatial variance in c_ref and n_eff"]

cons: ["slightly higher expression complexity"]

- name: "B - factored"

assumptions: ["c_ref approx. constant within segment"]

pros: ["concise"]

cons: ["sensitive to c_ref drift"]

decision: "Adopt general T_arr; allow factored as exception for region L2."

impact:

dimensions:

performance: { metric:"gate_accuracy", target: ">=0.99@7d", baseline:0.97, after:0.995, sigma:0.003, weight:0.30 }

compatibility:{ metric:"compat_rate", target: ">=0.995@replay", baseline:0.98, after:0.996, sigma:0.004, weight:0.10 }

cost: { metric:"unit_cost", target: "<=1.0x@base", baseline:1.00, after:0.90, sigma:0.030, weight:0.15 }

scoring: { k_sigma: 1.0 }

rollback:

triggers: ["gate_accuracy<0.98@7d","compat_rate<0.99@replay"]

steps: ["switch_traffic: release-stable","restore_snapshot: params@2025-09-20","run_suite: restoration_regression","observe: 24h"]

verification:

gates:

hard: ["gate_accuracy>=0.99@7d","compat_rate>=0.995@replay"]

regression_suite:

cases: ["cmb_set_v3","lens_v1"]

references:

see:

- "EFT.WP.Core.Equations v1.1:S20-1"

- "EFT.WP.Core.Metrology v1.0:check_dim"

- "EFT.WP.Core.DataSpec v1.0:I30-2"

arrival_time:

form: "general"

path: "gamma(ell)"

measure: "d ell"

c_ref: "ref:L2"

n_eff_profile: "dataset:cmb_set_v3@v3"

check_dim: true

III. 示例 B：含回滚闭环的差异裁定（人读简版）

• 封面与摘要：DR-2025-0220；状态：Approved；裁定：保持一般口径，L2/L3 区域临时允许常量外提，期限 30 天。
• 风险与触发：incident_level>=2@24h 或 data_drift>0.03@14d 触发回滚。
• 回滚执行：
  冻结写入 → 切回 release-20250930-1.3.2 → 恢复 schema@v2.2 → 运行 restoration_smoke 与 restoration_regression。
• 复原验证（硬门禁）：gate_accuracy>=0.99@24h，compat_rate>=0.995@replay，gate_error_rate<=1e-3@24h。
• 签署与通告：内部 oncall/owner/auditor；对外 status_page。

示例 B：机读（YAML）

decision_record:

id: "DR-2025-0220"

title: "Scoped exception for T_arr factoring in L2/L3"

status: "Approved"

exceptions:

- name: "Factoring in L2/L3"

domain: ["region:L2","region:L3"]

until: "2025-10-31"

recheck_gates: ["compat_rate>=0.995@replay"]

owner: "Approver/Owner"

rollback_plan:

freeze_io: true

baseline: "release-20250930-1.3.2"

steps:

- "switch_traffic: release-20250930-1.3.2"

- "restore_snapshot: schema@v2.2"

- "run_suite: restoration_smoke"

- "run_suite: restoration_regression"

- "observe: 24h"

success_gates:

- "gate_accuracy>=0.99@24h"

- "gate_error_rate<=1e-3@24h"

- "compat_rate>=0.995@replay"

communication:

internal: ["oncall","owner","auditor"]

external: { channels: ["status_page"] }

references:

see:

- "EFT.WP.Core.Equations v1.1:S20-1"

- "EFT.WP.Core.Metrology v1.0:check_dim"

IV. 常见错误对照（可直接拷贝）

V. 可复用片段库（直接粘贴）

1) 引用与锚点

references:

see:

- "EFT.WP.Core.Terms v1.0:P10-3"

- "EFT.WP.Core.Equations v1.1:S20-1"

- "EFT.WP.Core.Metrology v1.0:check_dim"

- "EFT.WP.Core.DataSpec v1.0:I30-2"

2) 门禁（gates）

verification:

gates:

hard: ["gate_accuracy>=0.99@7d","compat_rate>=0.995@replay"]

soft: ["unit_cost<=1.0x@30d"]

3) 到达时上下文

arrival_time:

form: "general" # general|factored

path: "gamma(ell)"

measure: "d ell"

c_ref: "ref:L2"

n_eff_profile: "dataset:cmb_set_v3@v3"

check_dim: true

4) 回滚触发器

triggers:

- "gate_accuracy<0.98@7d"

- "compat_rate<0.99@replay"

- "incident_level>=2@24h"

- "data_drift>0.03@14d"

5) 审计轨字段

audit_trail:

record: ["timestamp","actor","from","to","gate","evidence_hash","notes"]

VI. 人读 × 机读对位（核对表）

VII. 发布就绪断言（最小）

• 断言：hard_gates_passed=true；audit_artifacts in ["yaml","json","pdf"]；anchors_present=true；rollback_ready=true。
• 完成：打 release-<date>-<semver> 标签，更新版本关系与通告节奏。