Takeoff schema

The TakeoffReport is the structured contract between the client LLM and downstream estimating tools. It's a Pydantic model, validated strictly by generate_takeoff_report — bad reports come back as readable validation errors, not silent garbage. Field names intentionally match the contractor-bid BOM format (qty/uom/status/source/customFields) so exports drop straight into the ContractorTakeoff suite (see Integrations).

Current schema_version: 1.

TakeoffReport

A complete takeoff for one plan set.

Field	Type	Default	Description
`schema_version`	`int`	`1`	Schema version
`project`	`str`	required	Project name
`source_pdf`	`str`	required	Plan set file the takeoff came from
`scope_profile`	`str`	`"fences-gates"`	Trade profile used
`generated_by`	`str`	`"conmcp"`	Tool identifier
`generated_at`	`str` or null	`null`	ISO timestamp, set at export if missing
`items`	`list[TakeoffItem]`	`[]`	The line items
`scope_notes`	`list[str]`	`[]`	Inclusions/clarifications
`risks`	`list[str]`	`[]`	Red flags a bidder must see

TakeoffItem

One takeoff line item (a BOM row).

Field	Type	Default	Description
`id`	`str`	required	Stable item id, e.g. `'fence-cl-6ft'`
`section`	`str`	`"general"`	Trade grouping, e.g. `'fences-gates'`
`item`	`str`	required	Short item code/name, e.g. `'CHAIN-LINK-6FT'`
`description`	`str`	required	Full description as an estimator would write it
`qty`	`float` (≥0)	required	Quantity taken off
`uom`	`str`	required	Unit of measure: LF, EA, SF, SY, CY, PAIR, TON, LS
`status`	`str`	`"Included"`	`'Included'`, `'Excluded'`, or `'Flag <reason>'`
`source`	`str`	`""`	Sheet reference, e.g. `'C-101 p.2'`
`location`	`str`	`""`	Where on the project, e.g. `'north property line'`
`assumptions`	`list[str]`	`[]`	Assumptions behind the quantity
`confidence`	`float` (0–1)	`0.5`	Extraction confidence
`needs_review`	`bool`	`true`	Flag for human review before bidding
`custom_fields`	`dict[str, str \\| float \\| bool]`	`{}`	Trade-specific extras, e.g. `post_spacing`, `fabric_gauge`, `gate_leaf_count`

Note the defaults are deliberately conservative: an item is needs_review: true at 50% confidence unless the model explicitly says otherwise. Sight-unseen trust is not the default.

UOM vocabulary

The shared unit vocabulary (single source of truth in domain/quantities.py):

LF · SF · SY · CY · EA · PAIR · TON · LS

UOMs are normalized to uppercase on validation. A unit outside this list does not fail validation — it gets reported in flagged_uoms by generate_takeoff_report so a human can decide whether it's legitimate (say, SETS) or a mistake before export.

Status values

status is validated: it must be exactly Included, exactly Excluded, or start with Flag followed by a reason:

Included — in the bid scope
Excluded — documented out of scope (e.g. "BY OTHERS" / NIC items — recorded so the exclusion is explicit, not silent)
Flag <reason> — needs a decision, e.g. Flag plan/schedule mismatch — RFI candidate

Export formats

generate_takeoff_report (and conmcp report on the CLI) renders a validated report three ways.

JSON — contractor-bid compatible (`to_bid_json`)

The BOM line_items shape consumed by contractor-bid:

{
  "schema_version": 1,
  "project": {
    "name": "Oakridge Commerce Park — Phase 2",
    "scope_profile": "fences-gates"
  },
  "line_items": [
    {
      "id": "fence-cl-6ft",
      "section": "fences-gates",
      "item": "CHAIN-LINK-6FT-BW",
      "description": "6' high galvanized chain link fence w/ 3-strand barbed wire",
      "qty": 450.0,
      "uom": "LF",
      "status": "Included",
      "source": "C-101 p.2",
      "notes": "Posts 8'-0\" O.C. per detail 1/C-501",
      "customFields": {
        "post_spacing": 8,
        "fabric_gauge": "9 GA",
        "location": "north property line",
        "confidence": 0.9,
        "needs_review": false
      }
    }
  ],
  "metadata": {
    "extracted_by": "conmcp conmcp",
    "extracted_at": "2026-07-10T14:30:00+00:00",
    "source_pdf": "/Users/you/plans/sample_plan.pdf",
    "scope_notes": ["Gate operator power and loops by electrical contractor (BY OTHERS)"],
    "risks": ["Verify field dimensions prior to fabrication per general note 5"]
  }
}

Mapping notes: item assumptions are joined with "; " into notes; location, confidence, and needs_review travel inside customFields alongside any trade-specific extras.

CSV — pma-standalone takeoff layout

One row per item, columns in this order:

Section, Item, Description, Qty, UOM, Status, Source, Location,
Confidence, NeedsReview, Assumptions, CustomFields

Confidence is formatted to two decimals (0.90)
NeedsReview is lowercase true/false
Assumptions are joined with "; "
CustomFields is a JSON string (empty if there are none)

Markdown — human review

An Obsidian-friendly report: header metadata (source, scope profile, generated timestamp, item counts with how many need review), a line-item table with confidence and review flags, then Assumptions, Scope notes, and Risks / RFI candidates sections, plus a warning line if any non-standard UOMs were found. This is the format to read before the CSV or JSON leaves the building.