delay-dispute-leap

delay-dispute-leap is a research-oriented toolkit for reproducible experiments on construction delay-dispute outcome prediction and responsibility diagnosis. It packages the parts of a larger PhD research workflow that are safe to publish: evaluation utilities, audit helpers, figure generation, lightweight configs, and a small smoke-test path.

Scope

• Pre-decision text processing and audit helpers for delay-dispute analysis
• Candidate benchmark and evaluation scripts for outcome prediction experiments
• Reproducible reporting assets and example tables/figures
• Public-safe sample artifacts only; no full raw corpus or private case archive

Repository Layout

• src/: research scripts and utilities
• config/: public-safe configuration templates
• tests/: preserved test entrypoints from the local research workflow
• docs/assets/: sample figures and tables from non-sensitive experiment outputs
• examples/: toy inputs and sample prediction rows
• smoke_test.py: lightweight offline sanity check

Quick Start

python -m pip install -r requirements.txt
python smoke_test.py

The smoke test does not call external APIs. API-backed scripts require DASHSCOPE_API_KEY to be set explicitly in your environment.

Included Public Artifacts

• docs/assets/fig3_outcome_prediction_comparison.png
• docs/assets/fig4_confusion_matrix.png
• docs/assets/table_main_results.csv
• docs/assets/table_responsibility_results.csv
• examples/toy_case_record.json
• examples/sample_predictions.csv

Notes

• This repo intentionally excludes the full raw-text corpus, large result caches, and unpublished manuscript materials.
• Any script that uses a live LLM endpoint now requires an explicit environment variable. No debug key is bundled.
• For the multi-agent responsibility-attribution layer built on top of this workflow, see the sister repo madra-delay-attribution.

Citation

If this repository helps your work, cite the software record in CITATION.cff.