A Streamlit + Snowflake app that re-estimates an existing ADR project estimate for a user-selected Location and Time Period by applying EMMA market factors (MFC for materials, LRC for labor), then shows an original-vs-updated comparison and exports a CSV.
It shares its foundation with the sibling
data-quality-app: the same Snowflake client,
env-driven settings, mock/snowflake data-source switch, global theme,
session/router plumbing, and pytest + ruff + CI harness.
pip install -r requirements.txt
cp .env.example .env # only needed for snowflake mode
streamlit run app.py # or: make runDefault DATA_SOURCE=mock ships deterministic synthetic ADR + EMMA data, so
the full 3-step flow works with no Snowflake connection.
- Project - pick a project that has ADR estimations loaded (latest snapshot is used).
- Location & Period - choose from the (Location, Period) pairs present in both the MFC and LRC reference data.
- Estimation - see total cost & hours (original vs updated, absolute + %), a per-category breakdown (Specialty Subcontractor, Vendor Shop, Base Material, Field Shop Fabrication, Field Labor), grouped bar charts, and two CSV downloads (line-level + category summary).
For the selected location/period the engine pulls one labor factor + USD rate (LRC) and a per-code material factor table (MFC):
| Category | Type | Formula |
|---|---|---|
| Specialty Subcontractor | labor | SPEC_H_NEW = DB_SPEC_H × F_lrc ; SPEC_COST_NEW = SPEC_H_NEW × USD_R |
| Field Shop Fabrication | labor | FSF_H_NEW = DB_FSF_H × F_lrc ; FSF_COST_NEW = FSF_H_NEW × USD_R |
| Base Material | material | BASE_MATERIAL_COST_NEW = DB_BM_C × F_mfc[base_code] |
| Vendor Shop Fabrication | material | VENDOR_SHOP_FAB_COST_NEW = DB_VSF_C × F_mfc[vsf_code] |
| Field Labor | pass-through | carried from ADR databook (no factor) |
| Total Cost | VSF + SPEC + BM + FSF + FIELD_LABOR |
|
| Total Hours | SPEC_H + FSF_H + FIELD_LABOR_H |
Confirmed interpretations of the spec: Field Labor is a pass-through (the doc lists it only as a totals input), and the single LRC factor per (location, period) applies to both labor categories (no labor-type code in LRC). See CLAUDE.md for the full assumptions log.
DATA_SOURCE=mock pytest -q # 39 tests, ~95% coverage
ruff check .CI (.github/workflows/tests.yml) runs both with
--cov-fail-under=90.
app.py # Streamlit router (current_step -> renderer)
config/
settings.py # env-driven Settings (DATA_SOURCE + Snowflake)
schema.py # canonical column names, ADR table names, categories
src/
models.py # ProjectRef / FactorSelection / Comparison / EstimationResult
snowflake_client.py # connector wrapper (fetch_table / fetch_query + shared client)
mock_data.py # deterministic ADR 4-table + EMMA MFC/LRC
adr_repository.py # list projects + join 4 ADR tables (latest snapshot)
emma_reference.py # MFC/LRC load + selections + factor lookups
estimation_engine.py # vectorized calc + run_estimation -> EstimationResult
csv_export.py # line-level + category-summary CSV
ui/
_theme.py # one global stylesheet (status colours via sentinels)
step_project_selection.py # Step 1
step_parameters.py # Step 2
step_results.py # Step 3 (comparison + charts + downloads)
utils/
colors.py / helpers.py # status hexes + money/hours/% formatting
session_state.py # slim re-export shim over utils/session/*
session/ # state / navigation / sidebar
tests/ # engine, emma, adr, csv, mock, helpers, snowflake, AppTest
adr_repository and emma_reference branch on SETTINGS.is_mock. The
Snowflake path is wired (it reads the 4 ADR tables and the MFC/LRC references
through the shared client) but the exact source column names are placeholders -
reconcile the raw→canonical maps in config/schema.py
(MFC_RAW_RENAME / LRC_RAW_RENAME) and the ADR table projections against the
real schema. The calculation engine never changes.