Paste a section of a project specification. The scanner identifies every MEP-relevant clause, owner-specific override, and cross-reference contradiction — with a quoted excerpt from your text for every finding. Claims that can't be grounded go to the refuse list instead of the output.
Every finding passes two independent AI checks before reaching you. The second pass is specifically adversarial — its job is to catch citations the first pass fabricated.
v0 preview — paste-text input only. Full 800–1,400 page PDF processing is v1. Right now: paste up to 30,000 characters of spec text (roughly 20–25 pages). Optionally paste drawing notes for cross-reference detection. 5 scans per IP per 24 hours. No account required.
Cite-or-refuse scope. Every finding includes a verbatim quoted excerpt from your pasted text. The tool reads your spec section, not its training data. Claims the tool can't ground in your text go to the refuse list — they don't appear as findings. No dollar estimates. No page numbers unless your text contains them.
Paste the spec section to scan — divisions 01, 03, 07, 09, 22, 23, 25, 26, 27, 28 all supported
0 / 30,000 characters
Paste relevant drawing notes or keynotes for cross-reference contradiction detection
0 / 10,000 characters
v0 preview — for now, scans run by request. Paste your text, click the button to compose an email; typical turnaround is under 24 hours. The live in-browser tool is in development.
Scope summary
Generated only from verified findings. Every claim above traces to a finding in the sections below.
MEP-scope sections found
Owner-specific overrides
Cross-reference contradictions
Refuse list — what this tool couldn't ground
Refuse is first-class output, not an error. Two types: items the extraction pass couldn't cite verbatim (refused at source), and items initially extracted but whose citations the independent verification pass couldn't confirm (rejected post-verification). Both tell you where the tool's confidence ended.
How this works. Two independent AI passes. The first extracts MEP-scope findings and requires a verbatim quoted excerpt for every claim — no excerpt, no finding. The second pass independently checks whether each cited excerpt genuinely appears in your text; findings whose citations can't be confirmed are removed and logged in the refuse list. A third pass generates the scope summary paragraph only from findings that survived verification. The architecture is documented in the anti-fabrication essay. The full tool design is in AI for MEP estimating teams.
Privacy. Your pasted text is sent to an AI API to generate findings and is not retained beyond your session. The Hive does not store submitted text or use it to train models. See our privacy policy.