The Same Rule, Every Single Time.
Deterministic prose rules you define yourself, scoped to your project, your documents, your scenes, and your characters. Same input, same output, every pass. The suggestion drift you get from other tools doesn't happen here, because the rules are rules instead of model opinions. Grammarly can't promise that.
Five Rule Categories
Five analysis dimensions that cover structure, style, meaning, continuity, and the specific fingerprints machine-written prose tends to leave behind. The Machine Tells category is the one Reddit can't tell you how to write.
Compliance scores across categories. Deterministic rules that produce identical results every time.
- Structural
- Sentence length variance, paragraph breaks, dialogue-to-prose ratio, scene pacing metrics.
- Stylistic
- Adverb density, passive voice, repeated words, filter words, purple prose detection.
- Semantic
- Cliche detection, mixed metaphors, tone shifts, anachronistic language.
- Continuity
- Character name consistency, timeline violations, setting detail contradictions.
- Machine Tells
- Detects patterns common in machine-written text: hedging language, excessive qualifiers, formulaic transitions.
Hierarchical Precision
Rules cascade through four levels. Your villain can use passive voice while your narrator follows strict style rules. Same project, different rules, zero conflict.
Scope Hierarchy
Rules cascade through four levels: Project, Document, Scene, and Character. Set a project-wide rule against passive voice, then override it for a specific character's internal monologue. The scope hierarchy means you never have to choose between consistency and flexibility.
Rule Editor
Configure every rule with precision. Set thresholds (max adverb density: 3%), choose severity levels, define exceptions. The editor shows live previews of how each rule affects your current document.
Severity Levels
Four levels: error, warning, info, hint. Configure which issues block your flow and which are gentle nudges. Your project, your thresholds.
YAML Import/Export
Define rulesets as YAML files. Share them between projects, distribute them to writing groups, version-control them in git. Your style guide, codified.
ProseGuard vs. Generic Grammar Tools
Most writing tools give you suggestions that change every run. ProseGuard gives you rules that produce the same result every time, because you wrote them.
- Rule Enforcement
- Grammarly / ProWritingAid: Suggestions that change each run.
Ishvana: Deterministic rules. Identical results every time. - Character Voice
- Grammarly / ProWritingAid: Flags dialect as errors.
Ishvana: Scoped per character. Each voice has its own rules. - Scope Control
- Grammarly / ProWritingAid: Global settings only.
Ishvana: Project > Document > Scene > Character hierarchy. - Custom Rules
- Grammarly / ProWritingAid: Limited preset categories.
Ishvana: Full YAML rule definitions, import/export between projects. - Machine-Text Detection
- Grammarly / ProWritingAid: Not available.
Ishvana: Built-in Machine Tells category catches the patterns a model leaves on the page.
Deterministic means the same input always produces the same output. No randomness. No model drift. No surprises.
Stands Alone, Travels With Everything
ProseGuard's deterministic rules stand on their own. When you do call on one of Ishvana's agents, the rules travel with them, so the advice respects the same style guide your prose does.
- Machine Tells Detection
- The Machine Tells category runs pattern analysis against the prose, looking for the specific traits machine-written text tends to leave behind: hedging language ("it's worth noting that"), excessive qualifiers, formulaic paragraph transitions, the too-smooth rhythm models default to. Useful whether you're auditing a model-assisted draft or making sure your own prose doesn't accidentally sound like ChatGPT.
- Pre-Suggestion Hints
- ProseGuard feeds your active rules to Hawken before Hawken says anything. If your rules say "no passive voice in action scenes" and you're editing an action scene, Hawken respects the constraint before it opens its mouth. Prevention baked in at the prompt level, so the mistake never reaches the page.