Research documents
The files behind the study, laid out as they sit in the repository. Each one opens on GitHub in a new tab, always at its latest version.
docs/capstone/The study in plain words, and its running record.
- group-guide.mdStart here (opens on GitHub in a new tab)
What is StoryBuddy, what did the study find, and what are its limits? Plain words, no code.
- research-changelog.md (opens on GitHub in a new tab)
What changed since the plan was registered, and what must the paper say about it?
- research_direction_and_goals.md (opens on GitHub in a new tab)
What are the objectives, and what does the study not claim?
- methodology.md (opens on GitHub in a new tab)
How was the system built, and how is each objective evaluated?
- research_runbook.md (opens on GitHub in a new tab)
How were the judge's dataset, training and held-out evaluation run, and what were the numbers?
- pipeline-stage-audit-2026-10.md (opens on GitHub in a new tab)
For each defect in a finished book, which pipeline stage caused it?
- model-probes-2026-10.md (opens on GitHub in a new tab)
What did the small October model screens find, did anything change, and which pipeline limits must the write-up state?
- group-guide.mdStart here (opens on GitHub in a new tab)
docs/product/What was committed to in advance.
- PREREGISTRATION_OBJ4.md (opens on GitHub in a new tab)
What was fixed before any result was seen: hypotheses, splits, metrics, and every dated amendment since.
- ADRs.md (opens on GitHub in a new tab)
Which architecture decisions were made, and which are still in force?
- PREREGISTRATION_OBJ4.md (opens on GitHub in a new tab)
docs/specs/Rules the work follows.
- labelling-rulebook.md (opens on GitHub in a new tab)
How does a rater decide Same or Different for a pair of images?
- judge-finetune.md (opens on GitHub in a new tab)
How is the judge's dataset made, how is a training run started, and how is it evaluated?
- labelling-rulebook.md (opens on GitHub in a new tab)
backend/finetune/The training code and its settings.
- train_qlora.yaml (opens on GitHub in a new tab)
Which settings was the judge trained with: epochs, learning rate, adapter size, checkpoints?
- train_qlora.yaml (opens on GitHub in a new tab)
docs/diagrams/drawio/The pipeline, drawn.
- langgraph_pipeline_after_2026-10.drawio.svg (opens on GitHub in a new tab)
What does the pipeline look like as it runs today?
- langgraph_pipeline_before_2026-07.drawio.svg (opens on GitHub in a new tab)
What did the pipeline look like in July 2026, for comparison?
- langgraph_pipeline_after_2026-10.drawio.svg (opens on GitHub in a new tab)