Six dimensions, scored 0–100
Every review scores grammar, clarity, style, citation integrity, source transparency, and originality readiness from 0 to 100, then reports an overall score, a low, moderate, or high review-risk level, and a short evidence-based summary. Citation integrity means visible citation consistency and attribution, and originality readiness means specificity and traceable claims — neither is a plagiarism or AI-detection score.
Findings that quote your text
The report lists up to 12 findings, each with a category, a low, medium, or high severity, an exact excerpt, an explanation, and a specific recommendation, plus up to 5 strengths and 6 prioritized next steps. Excerpts are verified verbatim against your submitted text — a finding whose excerpt cannot be located is discarded. Filter findings by category and copy the whole report as Markdown.
Built for seven document types
Choose general, essay, research paper, blog post, business report, policy, or email so the review matches the document's conventions. An optional reviewer instruction of up to 500 characters adds assignment, policy, audience, or editorial requirements as review context.
Source checks from Literature Review Matrix packages
Attach a reference-only source package selected in the Literature Review Matrix and the report adds claim-level source coverage: up to 20 checks comparing exact draft excerpts against your selected evidence, with six statuses from supported to disclosure missing. Supported, partial, and paraphrase-drift checks must cite at least one supplied evidence ID, and evidence links lead back to the source package.
Excerpts are confirmed before they are sent
Nothing from a source package leaves your device until you confirm it. The preview dialog lists every selected excerpt and states exactly what the request sends: the opaque package ID, version, content SHA-256, and per item the opaque evidence alias, document title, matrix field, page number, optional claim, and exact quote — up to 50 excerpts within a 20,000-character evidence limit. The saved report may cite evidence IDs and paraphrase comparisons, but a report that copies a selected source quote is rejected server-side.
What this review will never claim
The review does not compare your document with plagiarism or publication databases, does not determine whether a person or an AI system wrote it, and does not verify that cited sources are accurate. These limits are printed on every report together with a disclaimer: the report is review evidence to use alongside sources, drafts, policy, and human judgment — not proof of misconduct.
Billing you can predict
The cost is 5 credits per 5,000 characters, rounded up — never more than 15 credits at the 12,000-character maximum — and the estimate is shown before you run. Credits are reserved at submission, finalized only on success, and released when a review fails, so failed reviews bill 0. Unlimited plans are unmetered. An idempotency key on each submission prevents duplicate charges for retried requests.
History you control
Reviews run asynchronously with a five-minute processing limit, and a review still in progress can be reopened from history. Each record keeps the input text, the report, and a credit receipt. You can reload a record to restore its text and settings, re-run it, or delete it once processing has finished.