Selected case study / QS automation

QS Takeoff and Tender Analysis Workbench

An auditable calculation path from explicit geometry to commercial review exceptions.

Validated vector fixtures become traceable quantity lines, versioned synthetic rates, bounded cost ranges, and line-level tender flags. Every formula, source geometry ID, and exclusion remains inspectable.

Open source Read evaluation

Actual local interface / synthetic plans, rates, and tenders

Problem

Quantity and tender review becomes difficult to audit when geometry assumptions, measurement formulas, rates, and exceptions lose their provenance.

System

Validated vector geometry, atomic edge classification, opening deductions, seven measured items, rate joins, uncertainty bands, and tender review flags.

Evidence

Three hand-calculated synthetic plans, 21 quantity lines, seven synthetic rates, three tenders, and five authored exceptions.

Boundary

No PDF, CAD, BIM, or IFC ingestion; no market pricing, bill of quantities, professional QS output, or award recommendation.

System

Geometry is validated before anything is priced

The workbench rejects invalid rooms, overlaps, openings away from walls, missing rates, and unit or currency mismatches. Only accepted geometry can produce measured quantities and a commercial review queue.

QS system map showing vector geometry validation, edge classification, quantity formulas, rate provenance, uncertainty bands, tender exception checks, and human commercial review — Generated from the current fixture evaluation. Synthetic pricing, the weaker naive perimeter baseline, and the no-award boundary remain explicit.

Measured behavior

Shared-wall classification fixes a visible baseline error

The deterministic fixtures are deliberately small enough to calculate by hand. Perfect matches show formula regression correctness inside the supported schema, not arbitrary drawing interpretation or pricing accuracy.

Quantity exact match1.00021 authored lines across three synthetic plans

Wall-length MAE0.000 mAtomic external and shared-wall classification

Naive baseline MAE6.333 mRoom perimeters over-count shared geometry

Tender flag F11.000Five authored exceptions across three tenders

Synthetic cost build-up with measured quantities, unit rates, provenance, and low to high illustrative ranges — Every priced line retains its quantity formula, geometry references, synthetic rate identifier, unit, and illustrative range.

Synthetic tender comparison with missing, low, high, extra, and within-band line-level review flags — Flags create a clarification queue. They neither rank bidders nor recommend an award.

Engineering decisions

Commercial outputs retain their calculation trail

The implementation emphasizes rejectable contracts and line-level provenance rather than presenting a screenshot as evidence of automated estimating.

Typed geometry contract

Scale, room rectangles, openings, IDs, and adjacency enter through a schema that rejects overlap and unsupported placement.

Atomic edge measurement

Collinear edges are split and classified so adjoining rooms share one partition instead of duplicating their full perimeters.

Dual provenance

Each cost line carries both source geometry references and a versioned synthetic rate source with unit validation.

Review, not recommendation

Missing, unusually low, unusually high, and extra items are surfaced for clarification without bidder scoring or award logic.

Retained limitation

Exact fixture results do not establish drawing understanding.

The workbench consumes a narrow rectangular JSON schema and synthetic rate library. It cannot interpret drawings, reconcile revisions, apply a standard method of measurement, calibrate market prices, or replace qualified commercial review.

Inspect limitations and human-review boundary

Run locally

Rebuild quantities, costs, and tender flags

The evaluator replays all three plans and tenders, compares shared-wall measurement against the naive baseline, and rewrites the machine-readable and visual evidence.

python projects/qs-takeoff-tender-analysis/evaluate_qs.py python -m pytest tests/test_qs_takeoff_tender_analysis.py streamlit run projects/qs-takeoff-tender-analysis/app.py