Skip to content

CHANGELOG.md

Latest commit

 

History

History
132 lines (80 loc) · 5.81 KB

CHANGELOG.md

File metadata and controls

132 lines (80 loc) · 5.81 KB
Raw

Changelog

All notable changes are documented in this file.

[v3.4] - 2025-12-20

Added

  • README coverage breakdown: Added a package-level coverage table for quick visibility into what’s tested.

Changed

  • Testability + coverage: Refactored entrypoints (including the demo screenshot server) to be more test-friendly and expanded unit tests across error paths.

Fixed

  • MockDataStore suite state: SetCurrentSuite now updates CurrentSuite when no error hook is configured.

[v3.3] - 2025-12-20

Fixed

  • Stale tooling references: Removed dead Makefile targets that referenced legacy JSON→SQLite tooling.

Changed

  • Documentation consistency: Updated AUTOMATED_EVALUATION_SETUP.md to match the README’s current commands and style.

[v3.2] - 2025-12-20

Fixed

  • Stats chart rendering: Ensured the Chart.js container has a stable height so stacked bars render correctly.

Changed

  • Navigation rail width: Further reduced the left sidebar width to free up content space.
  • Manual evaluation layout: Centered score selection and action controls for a more balanced layout.

[v3.1] - 2025-12-19

Added

  • Arena UI CSS regression tests: Basic tests to prevent layout regressions (sidebar width variable, toolbar flex layouts, dropdown + file input styling, and scroll button anchoring).
  • UI Tour screenshot automation: A Playwright-based capture script (npm run screenshots) to regenerate the README UI Tour screenshots deterministically.

Changed

  • More compact Arena layout: Thinner left navigation rail and tighter top bar spacing to prioritize content.
  • Toolbar layout fixes: Sticky headers/footers and title/tool rows now compact into flex rows and only wrap when needed.
  • Styled dropdowns and file inputs: <select> and input[type="file"] now match the Arena theme instead of rendering unstyled defaults.
  • Scroll buttons repositioned: “↑/↓” buttons use the left sidebar space on shell pages (and stay bottom-right on solo pages).

[v3.0] - 2025-12-19

Added

  • Automated LLM evaluation (optional): Multi-judge consensus scoring (via a Python FastAPI judge service) with job persistence, WebSocket progress, cost tracking, and an audit trail. See AUTOMATED_EVALUATION_SETUP.md and python_service/.
  • Encrypted API key storage: AES-256-GCM encrypted key storage in SQLite with UI masking (configured via ENCRYPTION_KEY).
  • Arena UI overhaul: A shared “Neon Glass Foundry” visual system + layout shell (templates/arena.css) applied across templates. See DESIGN_CONCEPT.md and DESIGN_ROLLOUT.md.
  • Coverage automation: Coverage badge generation plus update scripts (scripts/update-badge.sh, scripts/update-badge.ps1) and make update-coverage.

Changed

  • UI layout: Templates standardized around a shared top bar + left rail (Arena shell), while preserving SSR Go templates and zero build tooling.
  • Evaluation workflow: Added automated evaluation endpoints and async processing while keeping manual scoring fully supported.
  • Testability: Refactors (e.g., handler dependency injection) to enable broader unit and integration testing.

Removed

  • Legacy JSON migration tooling: Removed the v2.0 JSON→SQLite migration/remigration/dedup code paths and CLI flags.
  • Gemini CLI support: Removed Gemini CLI integration.

[v2.1] - 2025-03-16

Added

  • Dynamic Profile Grouping:
    Implemented dynamic grouping of prompts by profile with color-coded borders and enhanced visual separation.
    (See middleware/utils.go and templates/results.html for implementation details.)

  • Enhanced Score Color Management:
    Centralized score colors and added utility functions in templates/score-utils.js so that pages and charts now share a unified color scheme.

  • Results Table Enhancements:
    Introduced row highlighting, sticky headers, tooltips, and a progress bar in the results table to boost user experience.

  • Keyboard Navigation:
    Enabled keyboard navigation in the evaluation grid for rapid score selection.

  • Smart Mock Score Generation:
    Improved random mock score generation using tiered, weighted distributions for realistic prototype testing.

  • WebSocket Recovery:
    Added auto-reconnection and connection status monitoring on the results page for reliable real-time updates.

Fixed

  • Prompt Move Restrictions:
    Limited prompt moves to preserve profile group contiguity (feat: restrict prompt moves).

  • Profile Border Application:
    Corrected the application of border classes in both header and data cells, ensuring proper vertical borders, handling text overflow, and maintaining a consistent 50px cell size.

  • Model Deletion Logic:
    Fixed deletion in the WriteResults function to correctly remove models from the database.

  • UI Styling and Consistency:
    Addressed issues such as cell size uniformity, column separator accuracy, and constrained header widths in templates/style.css.

  • Case Sensitivity in Profiles:
    Resolved problems with profile ID mismatches and case sensitivity in grouping.

Refactored

  • Unified Code Cleanup:
    Removed duplicate code, extracted hardcoded values, and consolidated inline CSS into centralized styles in templates/style.css.

  • Profile Group Utility:
    Moved profile grouping logic to middleware/utils.go to enhance maintainability.

  • Score Visualization Synchronization:
    Standardized score color schemes across results pages and charts through centralized utilities in templates/score-utils.js.

Removed

  • Legacy Data Files:
    Deleted outdated JSON files (e.g., data/current_suite.txt, data/profiles-default.json, etc.) and obsolete SQLite WAL/shm files to streamline data management.

For full details, please refer to the git commit history.