p6-cost-loader: BOQ Into Primavera P6

# P6 Cost Loader (BOQ -> Primavera P6) Planners cost-load a P6 schedule by copying BOQ lines into activities by hand: which activity does this line belong to, which floor, what share. Hundreds of lines, redone every BOQ revision, and every typo is a wrong cash-flow curve. This skill does the transfer and shows its working. ## What it needs - **BOQ**: an .xlsx workbook, or a folder of them (one file per CSI division is fine). Header row with Description + Qty + Rate/Amount; headings can be a column or text rows above the lines. - **Schedule**: a P6 XER export of ONE project (File > Export > Primavera PM (XER)). - Python 3 + `openpyxl`. ## Workflow (always two steps - a person approves the matches before anything is loaded) 1. **Match** ``` py scripts/boq_to_p6.py match <boq.xlsx or folder> <schedule.xer> --out outputs/<job> ``` Writes `<job>-mapping.csv`: every BOQ line, the activity it matched, split rule, confidence, flags. 2. **Review** - open the mapping (Excel is fine). Change `activity_group` to an activity name (all its floors) or a single Activity ID; set `split` to `equal` or `Level 01:40;Level 02:60`. Fix every LOW. 3. **Build** ``` py scripts/boq_to_p6.py build <boq> <schedule.xer> --out outputs/<job> ``` Writes `<job>.xer` (import: File > Import > Primavera PM (XER) > Create New Project, or Update Existing), `<job>-p6-import.xlsx`, `<job>-review.xlsx`. Build refuses a mapping made for a different/changed BOQ. `run` = match + build in one go with no review (for testing only). `--expenses line|heading|auto`: one P6 expense per BOQ line, or per BOQ heading per activity (auto = per heading above 300 lines, so P6 gets hundreds of expenses, not tens of thousands). ## How to use this as Claude - Run `match` first. Read the review's **Summary** and **Heading Review** (LOW first, biggest money first) back to the user: totals, how many lines are HIGH/MEDIUM/LOW, and every BOQ error flagged. - Help the user fix LOWs in the mapping CSV - suggest the activity using the weak guess and the BOQ description, but let them decide. Never load a LOW line on your own judgment without saying so. - Then run `build`, confirm "Totals check: PASS" (or explain exactly which lines are unassigned), and give the P6 import steps. - If the schedule has no activity for a BOQ system (e.g. medical gases, UPS), say so - the fix is a new activity in P6, not forcing the money onto a near name. ## Confidence | | Meaning | |---|---| | HIGH | the heading's words match the activity's name; no flags | | MEDIUM | partial match, one-word match, cross-trade match, or a flag - glance at it | | LOW | no clear activity - **not loaded**; weak guess shown | Context used: CSI division of the line (`config/csi_divisions.json`), trade synonyms on the BOQ side (`config/synonyms.json`), rare words weigh more than common ones, BOQ boilerplate ("supply, installation and testing of") ignored, floor headings ("4TH FLOOR") route a line to that floor. ## Tests `py tests/regression.py` - 65 checks: a planner-made case study (48/48 activities equal the planner's manual loading), a real 1,332-line tower BOQ on a 688-activity schedule, floor routing, Excel-edited mappings. `py tests/boq_reader_check.py` - BOQ reader vs a planner's cleaned BOQ (442/444 lines, total diff 0.00). Test data is not bundled - see README. Run both after any change. ## Limits Text/number BOQs only (no scanned PDFs). Does not create activities, logic or durations - the schedule must exist. Matching is word-based: a schedule without an activity for some work leaves that money unassigned (by design). Lump-sum lines split equally unless you set a split.