Large projects and low-end hardware — what performance should WorldScript promise? #646

Why this exists

WorldScript needs honest performance expectations for large projects and constrained hardware. This Discussion explores user-facing budgets and trade-offs rather than promising unsupported scale numbers.

What should feel reliable?

Please discuss expectations for editor input latency, project open/save/autosave, scene navigation and search, memory use, 30-minute or longer writing sessions, background/resume behavior, lazy loading, and coexistence with local AI.

Large manuscript size classes can be useful prompts for benchmarks, not support guarantees. Consider projects with thousands of scenes or entities, large research or image sets, and long-running sessions without turning one anecdotal fixture into a universal promise. What should authors be told about degraded modes, progress indicators, cancellation, and recovery?

Compare the PWA, current transitional Tauri packaging, and future Qt product where useful. Eco Mode and other performance modes should make meaningful trade-offs visible. Mobile/tablet and low-memory behavior may require different budgets from a capable desktop, but they should not silently lose acknowledged work.

Context

This Discussion is informed by #332's cross-renderer performance baseline, #478's PWA-first product direction, and #556's local model lifecycle and Eco-Mode context. Those Issues remain authoritative; no duplicate Issue comments are added here.

Role perspectives

Role-perspective note: The viewpoints below are maintainer-curated, AI-assisted design lenses. They are not separate community members, votes, user testimony, or evidence of consensus.

• 📚 Epic/series author: large manuscripts, many scenes, deep continuity, and long sessions without friction.
• ⚡ Low-end laptop: bounded memory, responsive input, predictable thermals, and graceful degradation.
• 📱 Mobile/tablet: touch, constrained memory, resume behavior, and useful offline editing.
• 🤖 Local-AI user: model loading, inference memory, indexing, and editor responsiveness must coexist visibly.
• 🧪 Benchmark/reliability: reproducible workloads, p95 latency, failure recovery, and representative size classes.
• 🧭 Product/budget setting: publish honest budgets and trade-offs without turning every fixture into a support guarantee.

Questions

• Which latency and memory budgets matter most during a real writing session?
• What project size classes should benchmark prompts represent?
• How many scenes, entities, research items, or images create a noticeable threshold for you?
• Which operations should remain responsive while local AI is loaded or running?
• What should Eco Mode disable, defer, or simplify?
• Which PWA, Tauri, and future Qt differences are acceptable?
• What performance and recovery evidence would make a large-project claim trustworthy?

Relationship to implementation

This Discussion is performance and product-budget exploration. It does not create support guarantees, change #332/#478/#556, or alter their priorities, roadmap admission, acceptance criteria, or execution sequencing.