Skip to content

Latest commit

 

History

History
149 lines (120 loc) · 8.15 KB

File metadata and controls

149 lines (120 loc) · 8.15 KB

V1.2.0-beta - GitHub release preparation

  • Prepared fraud_screen for GitHub and Streamlit Community Cloud release.
  • Cleaned temporary outputs, caches, uploaded data, and generated reports from the release package.
  • Updated README with product overview, features, workflow, limitations, privacy notes, and deployment instructions.
  • Updated Streamlit landing page copy.
  • Verified requirements.txt and .gitignore for deployment readiness.
  • Kept detection algorithms, current_run binding, target review, case review, reproducibility metadata, and progress logging unchanged.

V1.1.22 - Processing progress and activity log

  • Added processing activity log to Streamlit.
  • Added progress reporter and optional progress callbacks.
  • Added current stage, file, sheet, elapsed time, records, and flags display during screening.
  • Added long-running task waiting messages.
  • Added friendlier error display with last completed stage.
  • Did not change detection algorithms, thresholds, ingestion logic, scoring, current_run binding, target review, case review, or download paths.

V1.1.21 - Retraction case review report

  • Added optional case review note input.
  • Added case_review_summary.csv output.
  • Added case_review_report.md output.
  • Added case_review_summary to JSON reports.
  • Added Case Review section to HTML reports.
  • Added case review output paths to current_run_manifest.json and ZIP packages.
  • The case review report organizes target figure review, anomaly flags, data coverage, manual review recommendations, and reproducibility information.
  • Did not change detection algorithms, thresholds, ingestion, target matching, current_run binding, or download paths.

V1.1.18 - Streamlit page summary fix

  • Added fraud_screen/reporting/summary_loader.py with load_current_run_summary().
  • Streamlit Results section now reads from generated JSON/CSV files instead of relying on session state defaults.
  • Results header uses current_run.display_title instead of paper_id.
  • Added debug info (summary_source, anomaly_csv_rows, manual_review_csv_rows, file paths) to download info panel.
  • Did not change detection algorithms, thresholds, scoring, current_run, fingerprint, upload logic, or download paths.

V1.1.17 - Real Source Data batch smoke test

  • Added scripts/run_real_source_data_smoke_test.py for batch compatibility testing of real Source Data files.
  • Added real_source_data_compatibility_summary.csv output.
  • Added real_source_data_compatibility_notes.md output.
  • The smoke test records processed sheets, skipped sheets, extracted records, generated reports, and files needing ingestion review.
  • Did not change detection algorithms, thresholds, scoring, current_run, fingerprint, upload logic, or download paths.

V1.1.16 - Real Source Data compatibility and inventory

  • Added source_data_inventory.csv for workbook and sheet-level ingestion status.
  • Added robust handling for empty sheets, text-only sheets, unsupported layouts, and read errors.
  • Added safer Excel header-row detection for real Source Data files.
  • Added simple support for horizontal replicate tables.
  • Added data reading overview to HTML reports and inventory_summary to JSON reports.
  • Included source_data_inventory.csv in ZIP downloads and current_run_manifest.json.
  • Did not change detection algorithms, thresholds, scoring, current_run, fingerprint, or download path binding.

V1.1.15 - Low-risk report presentation

  • Added report-level classification for no-flag, only-weak-low-priority, medium-priority, and high-priority results.
  • Improved HTML report summary for cases with only weak/P3 signals.
  • Added report_level metadata to JSON and report_metadata.csv.
  • Reduced alarming presentation for low-priority background observation signals.
  • Did not change detection algorithms, thresholds, current_run, fingerprint, upload logic, or download paths.

V1.1.14 - False positive control and low-risk messaging

  • Added minimum sample-size control (MIN_TERMINAL_DIGIT_N = 50) for terminal digit anomaly detection.
  • Downgraded standalone terminal digit signals to weak/P3 evidence.
  • Prevented single weak evidence flags from inflating overall risk scores.
  • Improved low-priority flag explanations in HTML reports.
  • Added false-positive severity fields to controlled benchmark outputs.
  • Did not change current_run, session_id, fingerprint, download paths, or upload logic.

V1.1.13 - Controlled sample detection accuracy

  • Added controlled benchmark support for detector validation.
  • Improved simple Excel table ingestion.
  • Treated Index/Time/Sample-like first columns as x_label instead of y_value.
  • Added row-order fallback alignment for constant offset and constant ratio detectors.
  • Improved SEM/SD/n summary-statistic mapping from simple tables.
  • Did not change current_run, session_id, fingerprint, download paths, or report metadata.

V1.1.12 - Download consistency and sharing stability

  • Added current_run_manifest.json path updates for each completed run.
  • Ensured HTML, JSON, CSV, and ZIP downloads are bound to the same current_run metadata.
  • Added report_metadata.csv download support for metadata verification.
  • Ensured ZIP packages only include files from the current output directory.
  • Added current download information panel in Streamlit.
  • Did not change detection algorithms, upload logic, current_run generation, or report wording.

V1.1.11 - Report wording optimization

  • Replaced deterministic misconduct language with cautious numerical anomaly screening terminology.
  • Added clear disclaimer that statistical anomalies are not evidence of misconduct.
  • Added manual review guidance for P1/P2/P3 findings.
  • Added no-flag explanation: no detected anomaly does not prove data authenticity.
  • Did not change detection algorithms, current_run metadata, session binding, or file upload logic.

Changelog

V1.1.0 (2025-06-26)

  • Restructured project for GitHub and Streamlit Cloud deployment
  • Fixed requirements.txt completeness (added xlsxwriter)
  • Fixed Windows CSV encoding (utf-8-sig + gbk fallback)
  • Fixed Windows path compatibility (pathlib throughout)
  • Fixed Unicode surrogate / emoji encoding error
  • Optimized Streamlit Web Demo UI with Chinese interface
  • Added progress display during screening
  • Added 6 metric cards for result overview
  • Added risk level explanation expansion panel
  • Added P1/P2/P3 review priority labeling
  • Added 6 download buttons for all report formats
  • Enhanced built-in demo with duplicate_vector case
  • Added error handling with detailed error information
  • Added privacy notice and usage instructions
  • Added .gitignore for upload/report/cache protection
  • Added .streamlit/config.toml for deployment config
  • Added CHANGELOG.md for version tracking
  • Updated README.md with full deployment guide
  • All 70 pytest tests passing
  • Zero compileall errors

V1.0.0 (2025-06-25)

  • Initial Streamlit Web Demo
  • File upload support (PDF, XLSX, CSV, ZIP)
  • Built-in synthetic demo
  • 12 anomaly detectors
  • Risk scoring and reporting
  • JSON and HTML report output

V1.1.19 - Target Figure Review Lite

V1.1.20 - Result reproducibility and run parameter lock

  • Added run parameter metadata to current_run_manifest.json, JSON reports, HTML reports, and report_metadata.csv.
  • Added config_hash and detector_config_hash for result traceability.
  • Added deterministic_run and random_seed metadata.
  • Added summary consistency checks for JSON / anomaly CSV / manual review CSV outputs.
  • Added scripts/run_reproducibility_check.py for N-run output comparison.
  • Did not change detection algorithms, thresholds, Excel ingestion, target review matching, current_run binding, or download paths.

V1.1.19 - Target Figure Review Lite

  • Added optional target figure / panel input in Streamlit.
  • Added lightweight target figure matching against source inventory, extracted records, and anomaly details.
  • Added target_figure_review.csv output.
  • Added Target Figure Review section to HTML reports.
  • Added target review summary to JSON and target_figure_review_path to manifest.
  • Did not add deep scan or change detection algorithms, thresholds, current_run, fingerprint, upload logic, or download paths.