Account and study viewers can open TrialRegistry(path, read_only=True) to read an
existing registration without creating directories, tables, or triggers. Missing
databases fail explicitly. Connections use SQLite mode=ro and query_only, and
registry mutation methods reject calls in this mode. This does not freeze a live
writer or turn an unverified database into authenticated evidence; consumers must
still verify definitions and referenced artifacts.
Registry connection contexts commit or roll back on exit and then close their
SQLite handle, including when the consumer raises. Direct connections remain
available but must be explicitly closed by their caller.
研究可信度升级：接口、使用示例、验收及限制见 11–20 使用说明。
Cross-project experiment scanner and SQLite index for quant research outputs.
quant-lab同时保留历史standard/v1读取能力，并提供严格、不可变的
standard/v2Parquet产物契约。统一reader仅在v2目录完全不存在时回退v1；
检测到损坏或不完整的v2会直接失败。
索引中的项目与运行ID来自已验证的标准清单，允许本地输出目录使用不同名称。
只有声明quant.decision/v1的decision.json才按模拟决策卡读取状态，
并核对运行身份；择时等应用的同名研究文件不会被误认成模拟调仓授权。
已有索引中按目录名保存的标准运行应重新扫描生成新索引；索引升级不改写原研究产物。
滚动因子研究SDK需显式安装.[rolling-research]，使用已固定的QDK共享快照和Factors持久缓存。完整入口见滚动研究说明。开发锁包含该extra，以便CI实际运行对应集成测试。
cd quant-lab
python -m venv .venv
.venv\Scripts\activate
python -m pip install --requirement requirements.lock
python -m pip install --no-deps --no-build-isolation --editable .
python -m pip checkquant-lab init
quant-lab scan --workspace configs/default.yaml
quant-lab validate --run-dir ../a-share-multifactor/outputs/run_001
quant-lab list
quant-lab compare --project a-share-multifactor run_a run_bconfigs/default.yaml lists project output roots. Edit paths for your machine.
| Type | Markers |
|---|---|
equity_backtest |
capital_curves.csv |
multifactor_compare |
synthesis_comparison_summary.csv |
sklearn_ml |
feature_importance.csv |
spread_backtest |
performance/summary.csv |
standard_v2_research |
validated standard/v2 research profile |
standard_v2_backtest-ledger |
validated standard/v2 ledger profile |
from quant_lab import load_and_validate_standard_run, write_standard_run_v2
manifest = load_and_validate_standard_run("outputs/run_001")write_standard_run_v2要求调用方显式提供代码版本、内部依赖版本、随机种子、
数据快照、标的主数据版本、执行模型版本、基础币种和完整血缘DAG。它先在临时
目录完成schema及hash校验，再原子发布到standard/v2，已有目录永不覆盖。
本仓库属于contract层，公开声明standard/v2@2.0.0和
puresaber.run-manifest@2.0.0，声明位于pyproject.toml的
[tool.quant-workspace]，由全栈清单校验。任何契约字段或schema变更必须先更新
黄金样例、兼容测试和迁移说明；standard/v1历史产物只读且不可改写。
requirements.lock同时覆盖运行时、开发和editable构建依赖，CI先按锁文件安装，再以
--no-deps --no-build-isolation安装本仓库。更新锁文件时应在干净分支重新解析并运行完整测试、覆盖率、
pip check和跨仓清单校验，锁文件与代码一并提交。若发布验证失败，回滚到上一个
默认分支提交及其锁文件；不得移动旧tag，也不得修改已生成的研究产物。
锁文件使用uv pip compile pyproject.toml --extra dev --extra rolling-research --python-version 3.10 --constraint=requirements-constraints.txt --output-file requirements.lock
生成，覆盖Python3.10–3.12及editable构建依赖，禁止在锁之外临时解析构建后端。
Equity research recipes may retain risk_model and PIT industry caps when
switching allocation between cost_aware and equal. The consuming allocator
must jointly enforce absolute factor, budget, position and turnover constraints;
this contract change does not waive the execution adapter's tracking-error gate.
Other allocation modes remain rejected for those risk extensions.
quant_lab.counterfactuals.intervention_plan在已有六维干预之外支持：
cash_buffer：只替换strategy.cash_buffer，值必须为有限的[0,1)比例；持仓数量、单只上限、风险模型和风险门禁不变。现金约束可能被其他上限覆盖，配置变化不保证实际投资额变化。trend_filter：只允许strategy.family在etf_trend和rank之间切换，保留排序因子、趋势窗口、调仓频率和其他约束。不能借此切换到buy_hold或改变持仓规则。
新计划沿用quant.paired-interventions/v1结构和单项干预核验；已有计划与哈希不变。消费者必须使用支持这些维度的Lab提交，旧消费者会明确拒绝未知维度。低、高现金缓冲情景应分别预登记，不按回放收益挑选；同一批历史日期不得累加为独立证据。单项效应仍为模型内终值收益差，残差不视为已识别的因果贡献。
Economic comparison requires the explicit measurement contract. Run the independent financial benchmark suite with pinned real repository adapters; synthetic fixture agreement does not certify market validity.
Run quant-lab findings --registry path/to/experiments.db for a compact inventory,
or add --json for recorded scope, every attempt and verified result metrics.
The command opens the existing registry read-only and preserves failed/retried
attempts. Missing or altered result files are marked unverified. Execution success
is not a hypothesis verdict; human research decisions remain in Studio.