Mystery Writer
Status: In ProgressThis tool to logically transition the story state with the narrative.
Why I Built It
This tool was inspired by the HuggingFace dataset
dmitriihook/qwq-32b-planning-mystery-7-24k, which is made
up of state and verb objects that transition the state. My intent was
to create a tool that could capture narrative elements and show the
mystery writer the gaps and loose ends within their story. The current
version is 90% manual, but I hope to use this dataset in a novel way.
Technology
- Angular
- Python / FastAPI
- Vercel
How to Use It
- Sign in, or work as a guest.
- Load or import a document.
- Works best with
.txtor.rtf;.pdfand.docxare supported with some bugs. - If you sign in, you can see documents you've worked on.
- A guest can work on a document, export it, and re-import it later.
- Works best with
- Once loaded, click twice to specify a range, or click and drag over text.
- In the right panel, select the kind of state or action you're labeling.
- Open the filter dropdown and hover over terms to see their definitions.
- You can add a clarifying note to whatever you labeled.
- In the Narrative Engine, add characters and artifacts.
- When you seed a state instance, you associate something you labeled with its state type and give it an owner — a person, group, or thing. Example: choose a dialect label, select "dialect" as the state type, and assign it to a person or group.
- Once you have a state, you can apply a transition. This is where the tool ties up loose ends, by invalidating/consuming the seed states. Any seed state remaining at the end of the narrative is a loose end.
- PLAN import is where the
dmitriihookdataset comes in.- Each plan represents a subplot in the narrative.
- Select an import and a dropdown appears; each option shows a step-by-step meta resolution.
- Most are broken because they're missing a state object.
- Match your seed states to a PLAN to figure out what's missing from the narrative.
- The Import CSV button is for later expansion.
- Mapping profiles (needs work)
- You create patterns you can find in your seed states.
- Finding: Dangling Elements (needs work)
- This tool auto-captures any state object that was never resolved.
Technical Details
Architecture overview and notable implementation details.
Known Limitations
- Matching PLANs to seed states is manual.
- State generation actions produce too many duplicates.