This folder contains the specification set for the StableSteering research platform.
Published HTML version:
The current repository also includes a runnable prototype. The most implementation-specific documents are:
- quick_start.md
- configuration_manual.md
- student_tutorial.md
- user_guide.md
- developer_guide.md
- faq.md
- system_improvement_roadmap.md
- research_improvement_roadmap.md
Current implementation highlights reflected in these guides:
- GPU-only real Diffusers runtime by default
- prompt-first setup with editable per-session YAML configuration
- async round-generation and feedback jobs with visible progress
- SQLite-backed local persistence
- backend-saved per-session HTML trace reports
- real GPU-backed end-to-end example bundle under
output/examples/real_e2e_example_run/ - Gemini-generated conceptual illustration assets plus precise SVG diagrams in the published HTML docs
-
motivation.md Explains the research problem, value, goals, non-goals, and success criteria.
-
theoretical_background.md Explains the theory behind diffusion conditioning, embedding steering, and preference learning.
-
system_specification.md Defines the functional system behavior, architecture, contracts, APIs, and constraints.
-
system_test_specification.md Defines the verification contract, replay guarantees, and acceptance criteria.
-
pre_implementation_blueprint.md Converts the specification set into an implementation-ready plan with defaults, scope, and milestones.
-
student_tutorial.md Provides a student-oriented walkthrough from motivation and theory to the implementation structure and learning exercises.
-
assets/illustrations/steering_loop.png Gemini-generated conceptual illustration of the steering loop.
-
assets/illustrations/system_architecture.png Gemini-generated conceptual illustration of the main system layers.
-
assets/illustrations/trace_report.png Gemini-generated conceptual illustration of the saved HTML session trace report.
-
assets/illustrations/runtime_flow.svg Precise SVG diagram of the main runtime architecture from frontend to generation, storage, and trace reporting.
-
assets/illustrations/session_lifecycle.svg Precise SVG diagram of the prompt-first session lifecycle, including async jobs and incumbent carry-forward.
-
assets/illustrations/feedback_modes.svg Precise SVG diagram of the supported feedback modes and the signal each one sends to the backend.
-
assets/illustrations/config_to_generation.svg Precise SVG diagram of how per-session YAML becomes a validated config snapshot and affects runtime behavior.
-
quick_start.md Gives the shortest path to installing, running, and trying the prototype.
-
configuration_manual.md Explains the per-session YAML configuration flow, parameter meanings, validation rules, and practical editing patterns.
-
user_guide.md Explains the current app from the perspective of a user running sessions.
-
developer_guide.md Explains the code layout, extension points, and local development workflow.
-
faq.md Answers common questions about the prototype and its current limitations.
-
system_improvement_roadmap.md Tracks engineering, UX, observability, performance, and release priorities for the system itself.
-
research_improvement_roadmap.md Tracks study design, baselines, measurement, and analysis priorities for the research program.
-
document_audit.md Records the top 30 improvements applied to each main document.
-
system_spec_legacy_combined.md Preserves the original combined system specification as an archival source.