Pre-submission vetter
Showcase · not yet tested on real applicationsA consent application PDF goes in. It is scrubbed, its scrubbed text is read by Claude against the council's checklist, and it comes out as a gap list with page evidence, the likely RFIs and a readiness score.
- 1UploadPDF, council, type
- 2ScrubNo AI: find and remove
- 3CheckClaude reads the set
- 4ResultsGaps, RFIs, score
Checking the server…
Council
Queenstown Lakes District Council
Application type
38 checklist items for QLDC · New dwelling · 1 with an unverified basis
- allForm 2 application completed and signedblocker
- allEvidence of ownership attachedblocker
- allAgent's authority from the owner statedmajor
- allBuilding Code clauses and means of compliance listedmajor
- allPIM (if one was obtained) referenced or attachedminor
- allDistrict plan / resource consent status identifiedminor
- allRBW declared and LBPs named with licence class and numberminor
- allCertificate of design work (Form 2A) from the LBP designerblocker
- allFull set of plans and specifications for the workblocker
- allSite plan: boundaries, siting dimensions, levels, drainagemajor
- allFloor plans, elevations and sectionsmajor
- allProject specification (not a generic one)major
- allProduct appraisals / CodeMark and installation instructionsmajor
- allNZS 3604 wind, exposure, earthquake and snow zones statedmajor
- allBracing plan and bracing calculationsmajor
- allFoundation plan and slab/footing detailsmajor
- allGround suitability: good ground or geotechnical reportmajor
- allFraming, lintels and truss designmajor
- allPS1 (and PS2 where required) for specifically designed workmajor
- allE2/AS1 weathertightness risk matrix score for each facemajor
- allCladding, roofing and junction/flashing detailsmajor
- allWet-area waterproofing and mechanical ventilationminor
- allH1 compliance method and calculation providedmajor
- allClimate zone stated; R-values shown for each envelope elementmajor
- allStormwater (E1) disposal shownmajor
- allFoul drainage (G13): connection or on-site wastewatermajor
- allWater supply (G12) and hot water systemminor
- allNatural hazards on the site identified and addressedmajor
- allFire safety for houses (C/AS1) incl. smoke alarmsmajor
- allSafety glazing and barriers (F2/F4)minor
- QLDCCorrect QLDC checksheet completed (CS19R, CS19SR, CS19C...)major
- QLDCLodged on the QLDC Community Portal with QLDC file namingminor
- QLDCSite altitude and snow zone stated; SED if above 400 mmajor
- QLDCFloor levels match PIM / consent notice; hazards ticked on CS19Rmajor
- QLDCOn-site wastewater: QLDC application form and designmajor
- QLDCStormwater soak pit design with site testing (CS 4.4)major
- QLDCConnection to Council Services / building-over-services formsminor
- QLDCNo alternative options or conflicting details in the setminor
Drop a consent application PDF here
Up to 4 MB and 100 pages in this demo. Scrubbed first; only the scrubbed text goes to a model, and only when you say so.
A fictional 10-page application for a new house in Arrowtown, with six gaps left out on purpose.
What this proves / what it does not
- The pipeline is real: a PDF goes in and is scrubbed by deterministic code; Claude reads the scrubbed page text against a council checklist held as data, and answers with a status per item, quotes and page numbers.
- Every quote is checked by code against the page text the model was given, so a quote the model made up, or put on the wrong page, is shown as such, and a "present" item whose quotes are not found is marked partial.
- The readiness score is arithmetic on the statuses and the checklist severities. The model never sets the score.
- The fictional sample has six gaps left out on purpose, and the results show whether a run caught each one.
- It has not been tested on real applications, and nothing here is calibrated against real council RFIs yet.
- It does not check Building Code compliance (whether the bracing is adequate, say). It checks whether the information a council needs is there.
- Checklist items marked "basis unverified" could not be confirmed against their source during research, and council requirements change. Each item links to its source.
- For your own file the model sees only the scrubbed text, never the PDF: drawings, scanned sheets and images are not read, and there is no OCR yet. Only the fictional sample is sent as a PDF, so the demo can show the model reading drawings.
- Without an API key on the server, results are a recorded run of the sample, clearly labelled. Your own file is never analysed in that case.
- It is an aid to the designer's own check, not a substitute for council vetting.
Testing on a real application
This vetter has not yet been tested on a real application. The honest test is one NJS has already lodged:
- 1Pick a past application that drew a council RFI, and have the RFI letter to hand.
- 2Check it first on the scrubber page (no AI involved) and read its report, especially what it could not remove. Its PDF is a viewing copy only: the boxes cover the details but the text underneath is unchanged.
- 3Upload the PDF here, choose the council and application type, and run the check. The vetter scrubs it again and sends only the scrubbed text to the model, never the PDF.
- 4Compare the gaps and likely RFIs with what the council actually asked for: what it caught, what it missed, and what it flagged that the council did not care about.
- 5Repeat on five to ten applications before trusting the hit rate, and feed the misses back into the checklist.
With a live key, only the scrubbed page text and the checklist are sent to Anthropic's API for that one request; your PDF is not. Anything the scrubber missed would go with the text, so read its report first. Only the fictional sample is sent as a PDF. This app does not store the file or the result.