edit | blame | history | raw

实验存储体系

创建人员:Codex
文件职责:记录 project A 实验数据、结果包、图片、表 schema 和读取追踪方式。
管理规范/模板:../../common/exp-doc/实验存储体系创建指南.md
引用文件:实验总纲.md;实验设计.md;实验执行日志.md;实验审计报告.md;实验问题记录.md。

1. 数据存储目标和范围

本存储体系用于 project A 的轻量实验环境验证。

目标是保证实验数据、结果包、图片和审计证据能从实验总纲追到结果包,再追到原始输入和中间数据。

2. 目录结构

project A/
  exp-doc/
  exp-data/
    raw/
    result/
    img/
    tmp/

3. 数据流

raw source
-> feature/intermediate
-> result package
-> summary/readout/manifest
-> audit report

exp-data/tmp/ 只放临时脚本和临时缓存,不作为正式结论入口。

4. 结果包组织规则

结果包路径:

exp-data/result/<run_id>/

结果包至少包含:

文件 作用
summary.json 关键状态和计数
readout.md 人读版结果和边界
output_manifest.csv 结果包资产清单和 hash
input_manifest.csv 输入快照索引
intermediate/feature_panel.csv 中间特征面

5. 图片和图表存储规则

图片路径:

exp-data/img/<experiment_id>/<run_id>/

图片文件名应包含对象 ID 和实验图形类型。

6. 表清单

table_name storage_backend storage_path_or_table table_role row_grain primary_key unique_key producer consumer update_mode official_flag retention_policy
signal_cross_input.csv csv exp-data/raw//signal_cross_input.csv raw object + event_date experiment_id + object_id + event_date experiment_id + object_id + event_date runner feature step overwrite_snapshot keep
volume_spike_input.csv csv exp-data/raw//volume_spike_input.csv raw object + event_date experiment_id + object_id + event_date experiment_id + object_id + event_date runner feature step overwrite_snapshot keep
feature_panel.csv csv exp-data/result//intermediate/feature_panel.csv intermediate object + event_date experiment_id + run_id + object_id + event_date experiment_id + run_id + object_id + event_date runner result step / auditor overwrite_snapshot keep
signal_cross_result.csv csv exp-data/result//signal_cross_result.csv output signal event experiment_id + run_id + object_id + event_date experiment_id + run_id + object_id + event_date runner auditor overwrite_snapshot keep
volume_spike_result.csv csv exp-data/result//volume_spike_result.csv output signal event experiment_id + run_id + object_id + event_date experiment_id + run_id + object_id + event_date runner auditor overwrite_snapshot keep
output_manifest.csv csv exp-data/result//output_manifest.csv manifest artifact run_id + artifact run_id + artifact runner auditor / AI overwrite_snapshot keep
summary.json json exp-data/result//summary.json summary run run_id run_id runner human / auditor overwrite_snapshot keep
readout.md markdown exp-data/result//readout.md readout run run_id run_id runner human / auditor overwrite_snapshot keep

7. 字段 schema

7.1 raw input

field_name data_type required nullable enum_values meaning source time_semantics example notes
experiment_id string 实验事项 ID runner generated-time EXP-20260601-PROJA-SIGNAL-CROSS-001 追踪字段
object_id string 对象 ID runner generated-time OBJECT001 合成对象
event_date string 事件日期 runner event-time 2026-01-01 ISO 日期
value number 观测值 runner event-time 10 通用时序值
volume number 观测量 runner event-time 100 仅 volume_spike 输入使用

7.2 feature_panel.csv

field_name data_type required nullable enum_values meaning source time_semantics example notes
experiment_id string 实验事项 ID runner generated-time EXP-20260601-PROJA-SIGNAL-CROSS-001 追踪字段
object_id string 对象 ID raw input event-time OBJECT001 追踪字段
event_date string 日期 raw input event-time 2026-01-04 追踪字段
value number 观测值 raw input event-time 14
trailing_ma3 number 前序 value 均值 runner as-of previous rows 11.0000 signal_cross 使用
volume number 观测量 raw input event-time 260 volume_spike 使用
volume_ma3 number 前序 volume 均值 runner as-of previous rows 101.0000 volume_spike 使用
signal_flag integer 0/1 是否命中 runner generated-time 1

7.3 result tables

field_name data_type required nullable enum_values meaning source time_semantics example notes
experiment_id string 实验事项 ID runner generated-time EXP-20260601-PROJA-SIGNAL-CROSS-001 追踪字段
design_id string 设计 ID runner generated-time DESIGN-20260601-PROJA-SIGNAL-CROSS-001 追踪字段
run_id string run ID runner generated-time RUN-20260601-PROJA-SIGNAL-CROSS-001 追踪字段
object_id string 对象 ID raw input event-time OBJECT001
event_date string 命中日期 feature_panel event-time 2026-01-04
signal_reason string 命中原因 runner generated-time value_gt_trailing_ma3_by_8pct

7.4 output_manifest.csv

field_name data_type required nullable enum_values meaning source time_semantics example notes
artifact string 资产名 runner generated-time summary
path string 项目相对路径 runner generated-time exp-data/result/... 不写本机绝对路径
artifact_role string raw/intermediate/output/summary/readout/image/manifest 资产角色 runner generated-time output
row_count_or_count integer 行数或数量 runner generated-time 8
sha256 string 文件 hash runner generated-time abc...

8. 读取方式

人读路径:

目录导读
-> 实验总纲
-> 实验设计
-> 实验执行日志
-> 结果包 summary/readout
-> 图片
-> 实验审计报告

AI 读路径:

实验总纲
-> 实验设计
-> output_manifest.csv
-> feature_panel.csv / result table
-> summary.json
-> 实验审计报告