How the workbench works
Inspectable rules, explicit controls, and room for the writer’s judgment.
Data and structure
The catalog contains 78 character briefs, 69 place briefs, 64 society briefs, 58 object briefs, 51 narrative briefs, and 45 nature and world briefs. Each has a bilingual title, central question, concrete fictional situation, and two practical actions to test. Eighteen method families supply scene materials and teaching approaches. The actions are not necessarily mutually exclusive: test the actual constraints first.
Specific settings and worked revisions
Each collection has two specific dimensions: responsibility and relationship; layout and access; authority and participation; material condition and use; trigger and turn; or fictional rule and scale. Automatic choices explore conditions across drafts, while an explicit choice retains that constraint. Nineteen bilingual worked revisions cover all eighteen methods, showing a first draft, diagnosis, revised scene, and reasons. Every tool also has its own bilingual dialogue seeds. Other pages clearly label their rule-assembled first drafts, diagnoses, and readable revisions.
Exploring the library
Each tool offers nine incidents and nine clues for its writing method, with automatic exploration available. Each of the twelve collection-specific dimensions has six choices. Exchanges can use a direct response, inquiry, alternative, or boundary. Character speech is separate from writing steps. These materials provide comparable conditions and working scenes; a combination is not automatically an independently written short story.
Story length and feeling
Brief sketch retains the incident, exchange, action, and bounded outcome. Developed scene adds setting and reactions; Extended story adds memory, emotional attachment, and a future consequence. Changing length preserves the name and core incident under the same settings. Each brief has a paired action and outcome. Incident and clue combinations are filtered for compatibility; when a requested clue is unsuitable, the result explicitly names its replacement. These remain rule-composed fictional drafts for a writer to assess.
Branch reading and practice
Choose A, B, or automatic exploration. A/B comparison keeps the name, seed, and conditions fixed while changing the direction. Thirty-six deep-reading briefs have separately written branch actions, bounded outcomes, responses, and specific scene passages. The other 329 compose their task choices with shared rules and retain reused prose. Every handbook includes a shared opening, diverging paths, quoted analysis, and exercises. Eighteen original technique examples and the nineteen existing worked revisions remain available. Reading notes travel with saved generated drafts; editing the prose removes its earlier generated analysis.
Generation
The engine normalizes settings and derives reproducible randomness from a tool identifier, seed, and fixed-order configuration. It selects naming material, an incident, and a scene variation, then applies stakes, cost, viewpoint, and connection. It does not understand a complete manuscript or conduct background research.
Constraints and preferences
A required motif appears literally in names. Exclusions are checked against names. Frame and tone affect vocabulary and scenes; structure controls name form, and the subtitle option creates an expanded form. Sound controls appear only for character tools. Overly restrictive constraints can yield no result rather than being silently ignored.
Reproduction and revision
The same version, task, locale, settings, and seed reproduce the same candidates. Changing result count preserves the earlier candidates. Rewriting can lock a name, premise, cost, scene, or connection. Export the full text of mixed drafts because a seed alone cannot recreate locked combinations.
Reading and continued exploration
With JavaScript off, articles, guides, worked revisions, and pre-generated examples remain readable. Interactive generation, comparison, and local notebooks require JavaScript. Next batch continues the same seeded sequence while avoiding earlier names. Available names depend on structure and constraints; the pool is finite. Fresh ideas changes the seed; setup links retain the batch and library choices. Edit a scene in the workbench and save it as another version without overwriting earlier saved drafts. A new engine version changes candidates, while old exported text can still be retained and imported.
Languages
English and Traditional Chinese are parallel creative versions, not certified word-for-word translations. Language switching opens the equivalent task page; it does not automatically translate saved project text. Names are not claimed to authenticate a particular region, real culture, or historical language.
Limits and verification
Automated checks cover counts, links, language annotations, seeds, exclusions, and data validation. They do not prove literary quality, scientific accuracy, or rights availability. The material is not professional medical, legal, or financial guidance. Verify factual claims separately.