The Cost Database

The economics engine is backed by a relational cost database — not constants in code. The database enforces provenance at the schema level and is fully open source: seed data, loader, and migrations live in the Jasper open-source repo under backends/cost-db/.

Design principles

• Provenance is a constraint, not a convention. Every fact row has a source_id foreign key and a basis_date, both NOT NULL at the database layer. An unsourced insert fails.
• Units are validated. Size and price units reference an allowlisted units table — a correlation can't silently mix m² and ft².
• Overrides are append-only. Price overrides are events in a log; a database trigger blocks UPDATE/DELETE, so the audit trail can't be rewritten.
• License-clean by construction. Coefficients enter only via open-source implementations or public datasets. No proprietary cost feeds (Richardson, ICIS, IHS) and no licensed index series (CEPCI) are stored — see Data Sources & Licensing.
• The repo is the source of truth. Seed data is versioned JSON loaded by an idempotent script; the database is a serving layer that can be rebuilt from git at any time.

Tables

Table	Holds
sources	One row per data source: type, license, URL, citation, access date
units	Allowlist of valid unit symbols with dimensions
equipment_correlations	Cost correlations: functional form, coefficients, size range, base index, bare-module factors, pressure model
material_factors	Material-of-construction multipliers (27-entry matrix)
chemicals	Chemical identities (CAS-keyed, linked to the component database)
chemical_prices	Dated chemical prices by region
utilities	Utility prices (electricity, gas, steam) incl. two-factor steam model
location_drivers	Composed location factors: labor rate, productivity, material index, FX, freight/duty
economic_factors	Contingency by AACE class, tax rate, discount rate
cost_index	The Jasper Cost Index: 304 monthly points, 2001 → today
unit_cost_map	Maps all 30 Jasper unit types to correlations (23 cost-bearing, 7 non-physical)
overrides	Append-only price/factor override events with scope and reason

Coverage

All 23 cost-bearing unit types in Jasper's catalog are covered by sourced correlations — 0 silent gaps. The 7 remaining types (Feed, Sink, Mixer, Splitter, Valve, PipeSegment, TextBox) carry no capital cost by definition. A queryable coverage report is served at GET /api/tea/coverage.

The Jasper Cost Index

Cost correlations are published against old base years (Turton: 2001, Seider: 2006), so costs must be escalated to today. Licensed indices like CEPCI can't be redistributed, so Jasper builds its own from public US Bureau of Labor Statistics series:

Jasper(t) = 100 · [ 0.36 · WPU10(t)/WPU10(2001-01)      (metals PPI)
                  + 0.25 · WPU11(t)/WPU11(2001-01)      (machinery PPI)
                  + 0.39 · CES2000000008(t)/CES(2001-01) (construction wages) ]

304 monthly points from 2001-01 (= 100) to today, rebuilt from BLS bulk files by cost-db/build_index.py. Validation: the 2001 → 2024 escalation ratio matches CEPCI's within 0.15% — CEPCI itself is used only as a test-time sanity constant, never stored.

Validation

The database ships with a benchmark suite (85 tests in CI):

• Per-unit exact match against the upstream open-source implementations the coefficients came from — including building and solving a real IDAES costing block and reproducing its result to 0.000%.
• HDA process benchmark (toluene hydrodealkylation, ~70 kt/yr benzene): total CAPEX lands inside the independently cited literature band.
• Capacity scaling: doubling capacity raises cost by 1.71× (exponent 0.78), inside the 0.6–0.9 textbook range.
• Corrupt-coefficient guards: a tampered coefficient fails the benchmark instead of shipping a wrong number.

Running it yourself

# Postgres + schema
export DATABASE_URL=postgresql+psycopg://costdb:costdb@localhost:5432/costdb
alembic upgrade head            # from backends/idaes/

# seed (idempotent — truncates and reloads fact tables)
python cost-db/load.py

# optional: rebuild the cost index from current BLS data
python cost-db/build_index.py

The IDAES backend mounts the /api/tea/* routes automatically when DATABASE_URL is configured, and boots stateless (simulation only) when it isn't.