@Josh Turner agree 100%. The first run flagged activities marked complete with future dates and overdue items nobody had touched. The bigger lesson came from this simplified view I made for our director. One timeline for all houses: blue is construction to date, red is where we’re late against the buyer’s contract, orange where the contractor is late against his. Drag the blue line to any date and the table shows % complete, what’s being built and where each contract stands. Measured against the latest updated programme, everything looked “on track”. Measured against the contract dates, every house is late. An updated programme absorbs its own delays – so we now measure against the contracts ( see attached).

@Tim Fairley Thanks! Because we’re a developer, not a builder, so programme and cash flow are tied together. A sold house carries two things a builder doesn’t track: a delivery date in the buyer’s contract, and buyer payments due at construction milestones. Unsold houses bring in no cash at all; their cost is funded until they sell. So the split shows three things at a glance: - Sold houses: delay against the buyer’s date (red) and against the contractor (orange). A slip there also pushes back the next buyer payment. - Unsold houses: only the contractor’s programme applies. The question is how far ahead of a sale we build and finance them. - For the director: when cash actually comes in versus when we pay the contractor. What gets measured, gets managed, and for a developer that means progress and money together.

Great thread. Sharing a real-world use case from our villa project: we used Claude + Revit (via MCP) to turn the mechanical 2D drawings into a 3D duct model, drop it onto the structural IFC model, and check every ceiling void for clashes before installation. Workflow: calibrate the PDF/DWG layouts to the structural grid → Claude traces the ducts and places them at the highest allowed level above the ceiling → automatic clearance check against beams, slab and other ducts → coloured 1:20 sections and a clash register in Excel/PDF for the engineer and subcontractor. First house: 22 structural clashes found. Second house was done in a day: 16 clashes, e.g. extract ducts running straight through beams. Cheap to fix on paper, expensive after the ceiling is closed. Red = concrete beams/slab, blue = ducts with insulation.

@Tim Fairley Mostly automatic. Claude reads the vector linework from the PDF and matches it against the structural IFC: walls, columns and the lift core. It finds the offset by image matching and gets to within about 15 mm. All sheets for the same floor (HVAC, drainage, water) share one frame, so once one sheet is calibrated the others follow. When we had DWGs instead of PDFs it was even simpler: we used the insertion point of the architectural xref. The manual part is the check. I look at an overlay of the drawing on the structure to confirm the columns and walls line up before any ducts are modelled. That takes about 5 minutes per floor.

Hi Tim — happy to share. I’ve attached a generic version of the template we built for a residential villa (example data only, our project figures removed). How we prepared it 1. Take-off from the drawings. The designer’s electrical sheets were vector PDFs, so instead of counting by hand we extracted the symbols programmatically (AI-assisted, Claude): each symbol was classified against the drawing legend (socket, socket with lid, USB, 1/2-gang, two-way, intermediate, appliance isolator, etc.), adjacent symbols were grouped into wall locations, and each group was allocated to a room. We cross-checked the result against an independent manual count and against two supplier quotations — the supplier offers turned out to be well off the drawings, which was the main reason for doing this. 2. One register as the single source. One row per group on the wall (room, drawing ref, composition, mounting height, count per item code). Every summary — by room, by floor, by item, frames, back boxes — is a formula off that register. 3. Composite rate per point = [(fitting + back box) × (1 − discount) + (1st fix h + 2nd fix h) × labour rate] × (1 + overhead) × (1 + margin). In our case the supplier priced “complete points” (box + insert + plate + frame), so we split the back box out — the boxes are needed on site at 1st fix, months before the fittings. 4. Two finishes. The client wanted the design finish only where it’s visible, with a cheaper finish behind appliances and inside joinery. So the finish is set per room by default and can be overridden per group. 5. Separate extracts. The full workbook is ours. The client and the interior designer got a one-page, plain-language list by room, with no item codes and, for the designer, no prices. Lessons learned - Count groups and points. With BS sockets in some premium ranges every point has its own box and frame (no multi-gang frames), so 4 points = 4 boxes = 4 frames. Check the manufacturer’s data before assuming otherwise. - Flag missing prices explicitly instead of leaving zeros in the total. - The labour hours in the template are placeholders — replace them with your own norms.

Yeah... Opus 5.5 seems like a better option.