v0.1.1

Release Notes v0.1.1

Summary

This release expands the research MVP into a more configurable and inspectable system.

Major themes in this release:

• richer sampling and preference-mode support
• stronger per-session YAML configuration
• better runtime trace reporting and portable HTML artifacts
• improved docs-site linking and roadmap detail
• real seed-policy behavior in the orchestration layer

Highlights

Sampling and preference updates

• added axis_sweep and incumbent_mix samplers
• added winner_only and approve_reject feedback modes
• documented the current sampler and preference-model behavior more clearly

Session and orchestration updates

• first round always includes the unmodified prompt baseline
• later rounds carry forward the previous winner as the incumbent
• default candidate count is now 5
• implemented seed policies:
  • fixed-per-round
  • fixed-per-candidate
  • fixed-per-candidate-role

Reporting and examples

• improved HTML session trace reporting
• ensured initial prompt visibility in generated HTML reports
• regenerated the real end-to-end sample bundle
• added a configuration-matrix sample generator

Documentation and publishing

• fixed broken Markdown links that pointed at machine-local paths
• regenerated the GitHub Pages site with corrected document and code links
• expanded the roadmap docs with:
  • why each item matters
  • implementation notes
  • success signals

Verification

Validated before release with:

• python -m pytest -q
• npm run test:e2e:chrome
• python scripts/build_pages_site.py

Known limitations

• multi-seed averaging is still specified in docs but not yet implemented
• mode-specific frontend controls are still incomplete for some preference modes
• the real-backend Playwright smoke remains opt-in

Release 1

Release 1

Highlights:

• real GPU-backed Diffusers runtime with local diagnostics
• prompt-first session setup with per-session YAML configuration
• resumable sessions, replay, trace reports, and GitHub Pages documentation
• one-click feedback submission plus next-round generation
• improved first-round diversity with one unmodified baseline and broader exploratory candidates
• image-by-image backend progress updates during generation
• human-readable session, replay, and dashboard labels
• browser and backend test coverage for core flows

Local verification on this release state:

• backend running with diffusers on CUDA
• Playwright browser suite passing except the opt-in real-backend smoke