Watch the workbench evolve
A demo area for showing the manuscript setup process and how evaluation outputs change as the story develops.
Open learning hubLocal-first story intelligence
Measure the story without giving the story away.
ObservaStory turns an Obsidian writing vault into an offline evaluation workbench: scenes stay as Markdown, story objects stay author-owned, and local AI writes structured numeric feedback back into frontmatter.
Offline capableDesigned around local Ollama evaluation.
Writer ownedScenes, claims, and definitions stay in Markdown.
Numeric firstStructured scores beat chatty opinion.
InspectableResults are stored where the writer can see them.
Feature map
Dedicated tools for the questions fiction writers actually ask.
The current workbench treats scenes as the unit of measurement, then connects each scene to characters, arcs, plot threads, chronology, claims, and project-specific metrics.
Foundation
Local Offline Workbench
Run evaluation against an Obsidian vault with local scripts and local model calls.
Benefit: privacy, portability, and fewer SaaS lock-in worries.
StructureAuthor-Owned Story Model
Keep characters, plot threads, arcs, and story engines as editable notes.
Benefit: the writer defines the story, not the tool.
ScoringNumeric Evaluation
Store normalized measurements for relevance, tension, resolution, and custom axes.
Benefit: feedback becomes comparable across a manuscript.
KnowledgeReader and Character Awareness
Separate what the reader knows from what a character plausibly knows.
Benefit: reveals, secrets, and flashbacks can be measured cleanly.
OrderStoryboard and Chronology
Track presentation order separately from story-world time.
Benefit: nonlinear fiction stops collapsing into one timeline.
CanonTruth Ledger
Collect author-declared claims and distinguish canon from model inference.
Benefit: the factual backbone remains writer controlled.
ReviewReports and Goal Review
Use stored frontmatter results to inspect manuscript health without re-calling AI.
Benefit: reporting can evolve from the saved evaluation data.
ScaleLocal Scheduler
Queue scene evaluations, run batches, and stop gracefully after the current job.
Benefit: large manuscript passes become practical.
Learn
Training, tutorials, and the evolving demo belong here.
ObservaStory.dev can grow into the public learning hub: setup notes, usage training, tutorial posts, and videos that show how story measurements change while a manuscript is configured and written.
Build a vault from zero
Step-by-step posts can cover scene metadata, story definitions, metric setup, scheduler use, and report reading.
Browse tutorialsExplain the philosophy
Short essays can explain numeric-first feedback, local AI, author-owned canon, and why Markdown matters.
Read notesWorkflow
Write normally. Measure deliberately. Review from stored results.
1. Author the scene
Write Markdown and attach the story objects that matter: characters, arcs, plot threads, chronology, and engines.
2. Run local evaluation
The evaluator reads scene text, definitions, and configured metrics, then writes structured output under frontmatter.
3. Review and revise
Reports, tutorials, and demos can use the saved values to show where the manuscript is strong, thin, or changing.
Contact
Interested in the workbench?
Send a note about the project, the training material, or a possible collaboration. The contact form sends through AWS without publishing a private email address.