Kurmaca.ink
FICTION WRITING AND NARRATIVE LOGIC WORKSPACE
An editorial product that combines long-form fiction writing with chapters and scenes, character knowledge, clues, timeline, narrative logic and continuity controls in one workspace.
Before the technical detail, read the problem, decision and evidence as one narrative chain.
- STATUS
- Live · actively developed
- TYPE
- Product · Writing / Narrative systems
- ROLE
- Product and systems architecture · Editorial UX and writing experience
- PLATFORMS
- Web · PWA
What was the problem, what changed, what is different now?
Before the technology details, this shows the transformation: the original bottleneck, the intervention, and the outcome that can be verified today.
PROBLEM
Narrative state split across editors, notes and spreadsheets breaks writing focus, while scene edits on Web/PWA also need a safe path through cloud synchronization failures.
INTERVENTION
I kept the manuscript at the center and attached structured narrative state around it, using Supabase as the canonical cross-device source and IndexedDB only as a recovery journal for unacknowledged scene edits.
OUTCOME / CURRENT STATE
Kurmaca.ink is live as a Web/PWA product combining the writing editor, structured narrative logic and Supabase-backed project continuity in one workspace; its crash-recovery layer is also developed behind explicit automated verification gates.
- EVD-01 / Current state
- Live · actively developed
- EVD-03 / Delivery surfaces
- Web · PWA
ROI, revenue lift or percentages are shown only when a verifiable source exists. Otherwise the proof is the working product, distribution, system outcome and sourceable scope — never invented metrics.
Week-1 retained active writers
Share of writers who complete at least one scene-writing or editing action in a project, return the following week and complete another scene-writing or editing action in the same or another project.
Initial scene write / edit → following-week return → another scene write / edit.
Project and scene-authoring lifecycle events; the real analytics source and event names must be verified before publishing a baseline.
No verified baseline is published for this case yet; no value is invented and no numerical claim is made before the source is verified.
Context
Long-form fiction is not only a prose-writing problem; it is a systems problem where manuscript, structure, character knowledge, clues and reveal/payoff state must remain coherent together.
Problem
Narrative state split across editors, notes and spreadsheets breaks writing focus, while scene edits on Web/PWA also need a safe path through cloud synchronization failures.
My role
- Product and systems architecture
- Editorial UX and writing experience
- Authoring safety and synchronization architecture
Solution
I kept the manuscript at the center and attached structured narrative state around it, using Supabase as the canonical cross-device source and IndexedDB only as a recovery journal for unacknowledged scene edits.
What was built
- Chapter / scene structure and focused long-form editor
- Narrative-logic surfaces such as Truth Ledger, character knowledge, reader knowledge and reveal map
- Clue, timeline and continuity workflows
- Supabase-canonical project and scene synchronization
- Journal-before-cloud IndexedDB recovery for scene edits
- Installable Web/PWA workspace
System architecture
Technical stack
FRONTEND
React 19 · TypeScript · TanStack Start · TanStack Router · TanStack Query · Tailwind CSS 4 · Radix UI
BACKEND / DATA
Supabase
RECOVERY
IndexedDB · idb · Revision-checked writes
PLATFORM
PWA
VERIFICATION
Vitest · TypeScript · ESLint
Important engineering decisions
- Keep the author and manuscript central instead of making AI chat the product center
- Keep canon decisions explicitly under author control even when the system suggests changes
- Use browser storage only for recovery instead of creating a second canonical project database
- Keep the PWA service worker network-only so stale app-shell caches cannot interfere with authoring state
Code / architecture excerpts
01const authoringModel = {02 canonical: "Supabase",03 ui: "optimistic",04 recovery: "IndexedDB journal",05 conflictPolicy: "explicit author decision",06 pwa: "network-only service worker",07} satisfies AuthoringArchitecture;Data flow
Challenges and constraints
- Balancing long-form writing focus with dense editorial and narrative tools
- Reconciling optimistic authoring state with canonical cloud revisions without data loss or silent overwrite
Result / current state
Kurmaca.ink is live as a Web/PWA product combining the writing editor, structured narrative logic and Supabase-backed project continuity in one workspace; its crash-recovery layer is also developed behind explicit automated verification gates.
Engineering takeaways
- In a writing product, the critical UX measure is not feature count but how rarely the author is pulled away from the manuscript.
- Narrative AI is strongest when it improves state and contradiction visibility rather than writing prose on the author's behalf.
- A recovery layer should improve data safety without becoming a second source of truth.
OUT-LIVE / VISIT PRODUCT
Kurmaca.ink
EVIDENCE / AUTHORITY
What does this case prove?
The roles and decisions documented here are not portfolio decoration; they are concrete evidence for the work described on the services and capabilities pages.
NEXT PROJECT
PRJ-01Ordovia