基准测试台的构建与校准
标定 AxialProof 全部基准共用的测量仪器（节点、时钟、HTTP 客户端、采样调度、结局分类、归档链路），量化仪器自身的噪声地板 σ_rig、注入线性度与最小可判别差异 MDD，使每一次跨平台差异的发布都能出示"该差异大于我们仪器噪声"的证据，且第三方能凭公开的镜像 digest、配置与参照端点重建等价仪器并复算。
AXP-RIG-v1 / 7 procedures / 71 recorded fieldsMEASUREMENT SPECIFICATIONS
How to measure a trading venue, written so someone else can check us.
Most venue comparisons publish numbers without publishing how the numbers were produced, which makes them impossible to check and impossible to reproduce. These are the specifications we measure against: for each metric, the quantity being measured, the parameters held fixed, the sampling design, the statistical treatment, and the check a third party can run to establish that our instrument — not the venue — was the source of a difference.
- SPECIFICATIONS
- 10
- PROCEDURES
- 76
- RECORDED FIELDS
- 1172
- DATASETS
- NONE YET
DATASET / NOT COLLECTED
Every specification below is complete and none of them has been run yet. No figure anywhere in this section was produced by measuring a venue. We publish the method first so that the results, when they exist, arrive against a standard that was fixed in advance and can be audited against this page — rather than one written afterwards to fit whatever came out.
SPECIFICATION REGISTER
Each metric carries a version, because a metric that changes silently is not a metric.
Datasets are versioned against these identifiers. If a specification is revised the identifier increments and earlier data stays bound to the version it was collected under.
SPEC ONLY
端到端有效成本（ERTC，全路径 bps）
定义并标定一个可被第三方独立重做的标准量 ERTC（End-to-end Round-trip Cost）：以欧元为计价基准，测量资金从法币/稳定币进入某中心化平台、完成成交、到资产离开平台为止的全路径价值损耗（单位 bps），并把它精确分解为入金通道费、明码撮合费、平台报价对独立参考价的偏离、点差、滑点、提现与链上费、碎片损失与资金滞留机会成本，从而使"0% 手续费""零点差"这类宣称可以被一个标准化的数字检验。
AXP-COST-v1 / 8 procedures / 91 recorded fields执行质量基准（Implementation Shortfall 三层分解）
回答一个具体问题：用户在提交订单那一刻看到的价格，和最终真正成交的价格，差了多少 bps，以及这个差额里有多少应该归给平台、有多少只是这段时间市场自己在动。方法是把偏离拆成三层——前端预览价相对参考中价的偏离、按平台自己当时的订单簿模拟吃单得到的理论 VWAP 偏离、以及实际成交 VWAP 偏离——再用同窗口参考指数的漂移做校正，得到可归因于场所的执行成本。所有实测值留空，由执行团队填写。
AXP-EXEC-v1 / 7 procedures / 90 recorded fields流动性韧性
规定一套任意第三方可独立复现的测量程序，把中心化加密交易平台的「流动性韧性」从形容词变成带单位、带不确定度的数：冲击成本曲线 C(x)（把中间价推动 x bps 所需名义金额）、深度补充时间常数 τ 与非参数恢复时间 RT50/RT90、深度的时间稳定性与日内模式、以及幽灵深度率 GDR（展示深度与实际可成交深度之差）；并规定一套扰动预算受控的主动探针设计，使测量本身不显著影响被测市场。
AXP-LQR-v1 / 7 procedures / 125 recorded fields价格发现与锚定
在一个不含被测方、成分与权重全部预注册的多源共识指数下，量化平台展示价格的持久性偏离与动态跟随关系（领先滞后、信息份额、跳变响应、微观结构指纹），并给出区分「正常做市差异」与「价格不跟随外部市场」的可复算判据与不确定度。
AXP-PRICE-v1 / 9 procedures / 112 recorded fields可用性与延迟基准
用地理与 ASN 分层的多探针、开环固定间隔主动探测，测量中心化加密交易平台各公开服务面的可用状态与响应延迟分布，产出按功能面拆分、带显式计数口径与 95% 区间的可用性估计和分位数延迟，并同时给出该观测窗口在统计上能排除与不能排除什么。
AXP-AVAIL-v1 / 9 procedures / 154 recorded fields披露完整性评分体系
用一套冻结版本的 58 条目量表，把"某中心化加密平台在其自有公开面上就法律主体、牌照及其法律含义、费率、托管、储备证明、安全控制、条款风险这七个域说了什么、说得多具体、能否被定位"转成两名评分员独立打分即可复现的数值向量，并同时量化该数值本身的评分员间一致性与不确定度。
AXP-DISC-v1 / 9 procedures / 122 recorded fields欧盟/MiCA 监管声称—公开登记事实映射
定义一套可被第三方原样重放的检索与编码流程：把中心化加密平台的监管相关声称逐条原子化，映射到欧盟及相关辖区的公开登记中可核验的记录，并把"声称 / 记录 / 差距"三层强制分离、分类、量化、留痕——全程只呈现事实与差距，不做任何法律定性。
AXP-MICA-v1 / 6 procedures / 118 recorded fields储备证明可验证性
定义一个带版本号、可被第三方独立重做的测量规范，量化中心化加密平台的储备证明（PoR）本身"能被独立验证到什么程度"：是否公布 Merkle root 并锚定、用户能否真正自查 inclusion（含验证器拒伪能力）、负债端是否受数学约束及覆盖哪些余额状态、储备地址是否公开且控制权与链上余额可独立重算、快照制度是否精确到可复现、第三方执行方属何种保证等级、以及该设计是否允许第三方检测快照日装点。不评价储备是否充足。"平台完全没有 PoR"是量表的 D0 分支，本身即一等结论，且要求比有 PoR 时更严格的取证纪律。
AXP-POR-v1 / 7 procedures / 168 recorded fields开放数据集与复现规范
测量并保证 AxialProof 全部指标数据集在「字段契约、缺失编码、时间与溯源、脱敏、完整性、版本与许可」六个维度上达到可验证水平，使第三方仅凭公开发布包（无需联系我们、无需任何账户或资金）即可重算出文章中出现的每一个数字，并把「这套数据到底有多可复现」本身量化成可发布的分数。
AXP-REPRO-v1 / 7 procedures / 121 recorded fieldsUSE
Using these specifications
They are written to be executed by someone other than us, which is the only test of whether a method is a method.
NOTES
- Start with AXP-RIG-v1. It calibrates the measuring apparatus itself and establishes the noise floor. Without it, a reported difference between two venues cannot be distinguished from a difference between two network paths.
- Each procedure names what its output supports and what it does not. Quoting a result past that boundary is the most common way these measurements get misused, including by the people who took them.
- Where a protocol cannot establish something, the specification says so rather than substituting a weaker proxy and reporting it as the real thing.
- Corrections and disputes about a specification are welcome against the metric identifier. A specification error is more consequential than a data error, because every dataset collected under it inherits the fault.