Completed words trigger quick server checks for the changed block. You decide which suggestions to apply.
Off · Harper only. Turn on AI for context-sensitive wording and deeper paragraph checks.
Write or paste a chapter of about 3,000 words. With AI on, Final review checks paragraphs still awaiting full text coverage. Unchanged text reuses valid results. Story names and vocabulary below apply to this session.
Coverage pending · preparing the checker…Session usage, monthly breakdown and editable assumptions
No proofreading requests yet.
The session estimate covers proofreading POST requests from this page only. It excludes earlier visits, other tabs, account credits and free allocations, the account subscription, and non-proofreading page traffic. Worker CPU uses an assumed 5 ms per quick request and 2 ms per deeper request; browser latency is not CPU time. Cache reuses do not create a new AI charge.
| AI input and output | — |
|---|---|
| Workers requests and assumed CPU | — |
| Before planning reserve | — |
| Planning reserve · 15% | — |
| Monthly planning estimate | — |
AI is off; the monthly projection uses zero AI calls.
Production assumptions · these are editable planning inputs, not measurements from the sample text.
Starting volume: 300,000 published chapters/month is used as a writing-volume proxy. Drafts, abandoned chapters and rewrites can increase monthly cost.
AI calls include sentence checks and full paragraph coverage. Input tokens include instructions, surrounding context and accepted vocabulary. Revising text and retrying requests can increase usage. The projection does not scale this short sample to a full chapter. It excludes logs, storage, taxes and negotiated pricing; the reserve is an assumption, not a guaranteed upper limit.
Public rates verified 13 November 2026: Gemma 4, $0.10 per million input tokens and $0.30 per million output tokens. Workers Standard, $5/month with 10 million requests and 30 million CPU ms included, then $0.30/million requests and $0.02/million CPU ms. AI pricing · Workers pricing. Account allowances are applied once in the monthly projection.
One name, alias, or term per line. Names and possessives are protected from spelling complaints. Changes invalidate earlier checks. Nothing is saved to a server database.
This optimized workspace is separate from the existing test app and the live Sherpa editor. Checks run on Cloudflare; no manuscript database is connected. Full text coverage means every current block received a complete check; it does not guarantee that every error will be detected. Use sample text while access controls and story-data integration are being prepared.
Request and performance details
No response yet.
Timings start when a request is sent and exclude the deliberate sentence pause and queue wait. Quick timing ends after the synchronous display update; it does not wait for a browser animation frame. They describe this browser session, not a production benchmark. Cloudflare does not expose reliable synchronous CPU timing here. Blocks are at most 2,000 characters; quick requests never contain the entire chapter. All deeper checks share a cache and a single active request. A change elsewhere does not discard a still-valid block result.