WorthIR 1.3.1 简体中文版

获取与运行

git clone --branch v1.3.1-zh-cn --depth 1 https://github.com/HongTianShi/worthir-artifact.git
cd worthir-artifact
./worthir demo-custom

Windows 使用 .\worthir.cmd demo-custom。

本次修正

• 区分框架使用、论文结果复算和原始检索路线重建。
• 更正源码启动器说明：创建本地 .venv 不访问 PyPI。
• 明确 FiQA-Compression260 是完整原始路线重建参考实现；其他任务保留统一工作流和适配器接口。
• 防止简体中文源码发布进入 PyPI 上传任务。

本补丁版本没有改变 WorthIR 的评价定义、任务契约或论文数值结果。

不可变源码版本

• 英文：v1.3.1
• 简体中文：v1.3.1-zh-cn

WorthIR 1.3.1

Install

python -m pip install worthir-eval==1.3.1
worthir demo-custom

If worthir is not on PATH, run python -m worthir demo-custom.

Changes

• Distinguishes framework use, paper-result recomputation, and original retrieval-route reconstruction.
• Corrects the source-launcher documentation: creating the local .venv does not contact PyPI.
• Identifies FiQA-Compression260 as the complete original-route reconstruction reference implementation; other tasks retain the shared workflow and adapter interface.
• Prevents Simplified Chinese source releases from entering the PyPI publishing job.

This patch release does not change the WorthIR evaluation definition, task contracts, or reported paper results.

Immutable source versions

• English: v1.3.1
• Simplified Chinese: v1.3.1-zh-cn

WorthIR 1.3.0 简体中文版

适用范围

此标签固定 WorthIR 1.3.0 的简体中文源码与命令行界面，对应提交 c0b8fc2。英文软件包使用同一套任务契约、评分器和组织者分析接口。

获取与运行

git clone --branch v1.3.0-zh-cn --depth 1 https://github.com/HongTianShi/worthir-artifact.git
cd worthir-artifact
./worthir demo-custom

Windows 将最后一行替换为：

.\worthir.cmd demo-custom

源码启动器会在本地创建 .venv 并指向当前源码，不访问 PyPI。若只需要英文 wheel，可运行 python -m pip install worthir-eval==1.3.0。

已验证环境

• Ubuntu、Windows 与 macOS；
• Python 3.10 与 3.13；
• 框架诊断、自定义任务、任务契约检查和 Python API；
• 逐查询分析、lambda 敏感性、硬预算与 Pareto 图；
• 论文发布 ledger 的表图复算。

使用边界

• 论文结果复算使用仓库提供的 query--route ledger，不等同于重新运行全部检索系统。
• FiQA-Compression260 是完整的原始路线重建参考实现；其他任务提供统一工作流和适配器接口，需要用户另行取得相应资源。
• 原始语料、索引和模型受各自许可约束，不随本仓库分发。
• PyPI wheel 的终端提示为英文；本标签提供简体中文命令行和文档。

WorthIR 1.3.0

Install

python -m pip install worthir-eval==1.3.0
worthir demo-custom

If the command is not on PATH:

python -m worthir demo-custom

Organizer analysis

• worthir analyze TASK --organizer-output organizer_private/per_query_scores.csv writes selected effectiveness, cost, utility, oracle route, regret, and opportunity stratum.
• worthir sensitivity TASK choices.csv evaluates frozen actions across a declared lambda grid.
• worthir budget TASK choices.csv reports feasibility and effectiveness under hard per-query cost ceilings.
• worthir plot TASK choices.csv writes an effectiveness-cost Pareto SVG without additional plotting packages.
• Evaluator-only outputs cannot be written under participant/.

Start and recovery

The root README now begins with PyPI and pinned-source installation, expected success output, generated files, a comparison example, and the two-column choices.csv contract. docs/TROUBLESHOOTING.md covers environment, PATH, network, model-download, disk, and task-validation failures.

Paper-result checks run on Linux, Windows, and macOS. The framework doctor exercises all organizer commands.

Immutable language versions

• English: tag v1.3.0, commit 0c0d59c
• Simplified Chinese: tag v1.3.0-zh-cn, commit c0b8fc2

WorthIR 1.2.1

WorthIR 1.2.1

• Rebuilds the eight FiQA-Compression260 routes from the official BEIR corpus and registered public models.
• Adds a CPU-only FiQA installation path using the official PyTorch CPU wheels.
• Verifies the official FiQA archive before extraction.
• Adds scheduled and manual 20-query FiQA route smoke tests.
• Publishes complete author and project metadata on PyPI.

English source: commit 6dda1c7
Simplified Chinese source: commit c2e03b8

WorthIR 1.2.0

WorthIR 1.2.0 binds the artifact to the 2026-08-16 paper submission.

• Recomputes Figure 1–7 and Table 1–6 under their manuscript numbering.
• Writes a persistent paper-result index with reproduction levels and appendix outputs.
• Adds a locked paper environment and executable raw-route replay stages.
• Adds Python API, contributor, and new-task entry points.

Downloads:

• worthir-artifact-1.2.0-en.zip: English source archive
• worthir-artifact-1.2.0-zh-cn.zip: Simplified Chinese source archive
• worthir_eval-1.2.0-py3-none-any.whl: installable framework

WorthIR v1.1.1

WorthIR v1.1.1 aligns the tagged source with the downloadable source archives and installed wheel.

English source and release tag: 8817ead

Simplified Chinese counterpart: c805bfe

Changes:

• preserves LF line endings and Unix execution permissions for the worthir launcher;
• tests the built wheel from outside the repository on CI;
• uses worthir in installed-mode guidance from demo-custom, build-custom, build-trec, and init;
• keeps generated task directories free of packaging cache files.

The v1.1.0 release remains available unchanged.

WorthIR 1.1.0

This version exposes decision-time costs to external routers without crossing the evaluator information boundary.

• Fixed and query-dependent public costs declare when they become available.
• Task validation reconciles every public cost with the evaluator ledger.
• The example router reads legal state, routes, lambda, and public costs.
• A normal wheel installation includes the CLI, templates, contracts, and runnable examples.
• Figures 1--7 can be rebuilt from the released paper inputs.

The earlier v1.0.0-ipmc2026 release remains unchanged as the artifact originally submitted with the accepted paper.

WorthIR artifact for IP&MC 2026

Frozen artifact for the accepted IP&MC 2026 paper. Run python setup_environment.py for the reusable framework, ./worthir demo-custom for the generic task and external-router example, or python paper_results/run.py for persistent paper-result reproduction.