Clean energy capital fails when models, money, and schedules disagree.

Project Epsilon unifies least-cost portfolio sizing, 8,760-hour grid stress, project finance pro-formas, and EPC critical-path scheduling under one shared scenario engine. Built on transparent open-source science, not proprietary black boxes.

Open the Interactive Platform View GitHub Repository
Physics to Cash Flow

No Copy-Pasting

One typed scenario definition drives capacity expansion, hourly dispatch, bankable debt sculpting, and Primavera P6 schedule binding.

Open-Source Science

Zero Black Boxes

Powered by PyPSA, HiGHS, NREL-PySAM, pandapower, and Mesa. Inspect every LP constraint, dual variable, and physical assumption.

Execution Realism

BOM ↔ COD Binding

128-week transformer lead times translate into exact carrying costs, loan covenant DSCR slips, and credit cliff float.

1/6 · Quantitative Decision Engine

What Epsilon can do: High-value insights, not an event register

Traditional tools record siloed point-estimates. Epsilon computes the trade-offs across physics, financial leverage, and procurement realities to answer board-level clean energy investment questions.

PyPSA + HiGHS LP

Hyperscale 24/7 CFE Optimization

Solves the non-linear cost curve of achieving hourly carbon-free energy matching for large power consumers (e.g., 250 MW AI data centers). Identifies the exact overbuild ratio between local PV, long-duration BESS, and remote wind to minimize blended LCOE while avoiding exorbitant curtailment penalties.

min Σ(CAPEX_r · Cap_r + OPEX_r · Gen_rt) s.t. Gen_rt + Discharge_rt ≥ Load_t, Emitted_CO2 ≤ Budget
94.2%
24/7 CFE Score
$54.20
Blended LCOE ($/MWh)
1.82 : 1
PV/Storage Ratio

Key Insight: Jumping from 90% to 98% hourly CFE matching increases portfolio CAPEX by 38% due to storage duration limits, highlighting where clean firm power or flexible compute dispatch beats battery overbuilding.

ISO-RTO + Monte Carlo

Extreme Weather & Dunkelflaute Stress

Subject portfolios to multi-day winter freezes, Dunkelflaute events (zero solar, low wind, peak heating demand), and correlated generator outages. Quantifies tail-risk unserved energy and peak peaker reliance beyond average-weather assumptions.

EUE = ∫ (Load_t - AvailableGen_t) dt | Dunkelflaute_T & P(Outage_k) ~ Binomial(n, p)
0.002%
Unmet Load Fraction
42 MWh
P90 Unserved Energy
84 hrs
Critical Lull Duration

Key Insight: Synthetic average-week profiles miss 90% of blackout exposure. Epsilon's Monte Carlo outage injector reveals that 4 hours of storage leaves a 14-hour deficit during a regional freeze unless paired with on-site dispatchable backup.

ProjectFinanceAdapter

Revenue Stacking & IRA Credit Cliffs

Models 20-year project finance pro-formas stacking PPA contracts, wholesale VPP capacity, demand charge clipping, and retail energy arbitrage. Simultaneously models Inflation Reduction Act (IRA) Section 6418 transferability, domestic content bonuses (30% vs. 40%), and construction-start credit cliff float.

CFADS_t = Rev_PPA + Rev_VPP + AvoidedTariff - OPEX; DSCR_t = CFADS_t / DebtService_t ≥ 1.25x
14.8%
Sponsor Equity IRR
1.34x
Min Debt DSCR
10.5 mos
Credit Cliff Float

Key Insight: VPP frequency regulation revenues lift project IRR by 240 bps, transforming a marginal 8.5% standalone solar array into a bankable 14.8% levered asset that qualifies for DSCR-sculpted senior debt.

Theta Core CPM + BOM

Equipment Slip to Debt Covenant Breach

Binds a 100+ SKU hardware parts catalog (BOM) directly to Primavera P6 Critical Path Method (CPM) tasks. Translates long-lead equipment delays (e.g. 128-week utility-scale transformers) into precise schedule slip, construction carrying cost interest, and debt service covenant default risk.

COD_slip = CPM_CriticalPath(Task_procurement + LeadTime_delta); CarryingCost = CAPEX · r_borrow · Δt
+14 wks
Transformer Slip
$1.84M
Carrying Cost Accrual
1.14x
Post-Slip DSCR (Default)

Key Insight: A 14-week transformer shipment slip shifts commercial operation into the low-tariff winter season, wiping out $1.84M in interest reserve and breaching the 1.25x lender DSCR covenant before commissioning starts.

2/6 · Using Verified Tooling

Engineered on battle-tested open scientific tooling

Instead of building brittle proprietary solvers or relying on opaque vendor calculations, Epsilon orchestrates the leading open-source libraries in power systems, mathematical optimization, and network graph theory.

PyPSA

v0.26+
Capacity Expansion & Dispatch LP

Python for Power System Analysis. Formulates linear programming models to determine optimal solar, wind, battery, and transmission investments across thousands of timesteps subject to emissions and network constraints.

Why it matters: Full visibility into system nodal equations, shadow prices, and dual variables without expensive commercial solver fees.

HiGHS

highspy v1.6+
High-Performance LP/MIP Solver

State-of-the-art open-source linear, mixed-integer, and quadratic optimization solver. Solves large-scale multi-period dispatch and portfolio expansion matrices in seconds with enterprise-grade numerical stability.

Why it matters: Rivals proprietary solvers (Gurobi, CPLEX) on large LP benchmark problems, ensuring lightning-fast runs inside standard Docker containers.

NREL-PySAM

NREL SAM v5.1+
Solar & Battery Physics Engine

National Renewable Energy Laboratory's System Advisor Model Python wrapper. Computes irradiance transposition, module temperature coefficients, inverter clipping, and physical battery degradation schedules.

Why it matters: Bank-accepted generation profiles based on peer-reviewed government lab performance curves.

pandapower

pandapower + linopy
AC/DC Power Flow & Feeder Analysis

Combines pandas with MATPOWER-based AC and DC power flow solvers. Validates radial distribution feeder voltage profiles, transformer thermal ratings, and IEEE 33-bus injection limits when DERs feed back into the grid.

Why it matters: Verifies that a financially optimized portfolio will not trigger physical voltage excursions or line overloads on local utility feeders.

Mesa & Nashpy

Mesa ABM + Game Theory
VPP Multi-Agent Market Clearing

Agent-based modeling of distributed aggregators bidding flexible capacity into 8-stage wholesale day-ahead and real-time markets. Uses game-theoretic equilibrium methods to model strategic bidding and gaming.

Why it matters: Replaces static revenue multiplier assumptions with dynamic market clearing where aggregators compete for ancillary service awards.

NetworkX

Theta Core CPM
Critical Path Method Graph Theory

Constructs directed acyclic graphs (DAGs) representing Primavera P6 construction activities. Calculates Early/Late Start, Total Float, and topological critical path sorting when equipment lead times shift.

Why it matters: Enables sub-second CPM recalculation and Monte Carlo schedule risk simulation without relying on proprietary Oracle P6 desktop installs.
Geospatial Diligence

Folium & HIFLD

Calculates precise geodesic distances in statute miles to the nearest Homeland Infrastructure Foundation-Level Data (HIFLD) transmission substations and highway access.

Financial Mechanics

numpy-financial & Scipy

Drives nominal/real Fisher equation discount rates, debt amortization schedules, working capital draw S-curves, and Monte Carlo probability distributions.

Visualization

Plotly & Streamlit

Interactive 8,760-hour dispatch heatmaps, duration curves, Gantt charts, and audit gauges rendered natively in high-performance web interfaces.

3/6 · Comparative Architecture

Open Stack vs. Black-Box Enterprise Suites

Energy modeling has historically been trapped between opaque $100k/year enterprise desktop software and fragile disconnected Excel spreadsheets. Here is how Epsilon differs.

Dimension Project Epsilon (Open Stack) Legacy Suites (PLEXOS / Aurora) Isolated PM Tools (P6 / Procore alone)
Optimization & Solvers PyPSA + HiGHS: Open, reproducible LP formulation. Inspectable mathematical constraints, dual variables, and shadow carbon prices. Proprietary black-box solver wrappers. Opaque penalty heuristics and uninspectable objective formulations. No optimization. Only accepts manual dates entered into spreadsheets.
Physics & Generation NREL-PySAM: Transparent, peer-reviewed solar irradiance, thermal derating, and battery degradation curves. Coarse historical shape libraries or third-party proprietary weather add-ons. Zero physical awareness. Energy output must be imported from external tools.
Supply Chain ↔ Cash Flow Theta Core CPM + BOM: Equipment lead times (100+ SKUs) bound directly to CPM tasks, updating project IRR and DSCR in real time. No construction or supply chain awareness. Assumes COD happens automatically on scheduled date. P6 tracks task dates, but lead times live in disconnected emails and Excel logs. No finance feedback.
Electrical Grid Feeder pandapower: AC/DC load flow, IEEE 33-bus feeder simulation, and N-1 contingency line screening. Requires expensive separate transmission licenses; often limited to bulk node approximations. No electrical network awareness whatsoever.
Auditability & Trust Full Provenance Ledger: SHA-256 assumption hashing, explicit confidence scores (0–100), and automated caveat registry. "Trust the vendor" model. Difficult to trace which internal coefficient created an anomalous spike. Scattered across change orders, meeting minutes, and conflicting spreadsheet versions.
Deployment & Cost Permissive Open Source: Fully containerized, zero per-seat solver taxes, deployable locally or on cloud containers. $50,000 to $150,000+ per seat annually. Windows-bound, proprietary binary license servers. Heavy enterprise license lock-in with steep learning curves for non-schedulers.
4/6 · End-to-End Decision Flow

The Scenario Decision Journey: From assumptions to bankable board reports

Instead of manual copy-pasting between optimization scripts, dispatch spreadsheets, and P6 schedules, Epsilon runs an auditable seven-step flow on a single typed Scenario envelope.

System Architecture: Scenario-First Decision Spine
Step 1

Define Scenario & Siting

Assumption Editor · Preset Library

Input physical load (e.g. 250 MW hyperscale or 10 MW campus), grid ISO region (CAISO, ERCOT, PJM), utility rate tariff, target CFE %, and financial debt terms.

Step 2

Size Clean Portfolio

PyPSA · HiGHS LP Solver

Optimizes least-cost MW of solar, wind, and battery storage. Generates an immutable CapacitySeed envelope passed downstream automatically.

Step 3

Stress-Test Grid Reliability

ISO-RTO Adapter · Monte Carlo

Runs hourly dispatch against weather freezes, Dunkelflaute anomalies, and generator forced outages to compute unserved energy, curtailment, and P50/P90 distributions.

Step 4

Evaluate Project Finance

Microgrid Adapter · IRA Section 6418

Stacks PPA power revenues, wholesale VPP reserves, and demand-charge savings into a 20-year pro-forma. Evaluates debt DSCR sculpting and 30–40% ITC tax basis.

Step 5

EPC Schedule & Lead-Time Realism

Theta Core · NetworkX CPM · BOM Binder

Expands capacity into a 100+ SKU hardware parts list. Ingests Primavera P6 schedules, binds equipment lead times to tasks, and measures COD slip and carrying cost accrual.

Step 6

Site Diligence & Geospatial Screening

HIFLD Dataset · OpenStreetMap · Folium

Screens land constraints and computes proximity to the nearest high-voltage transmission substation in statute miles, flagging expensive interconnect extensions early.

Step 7

Generate Audited Decision Report

Reporting Engine · Audit Registry

Exports audience-tailored deliverables (Executive Brief, Investment Committee Memo, Technical Review) accompanied by confidence scores and delta comparisons.

5/6 · Supporting Decision Making

Commercial dilemmas solved by Epsilon

Clean energy developers and investment committees do not need another passive log of activities. They need quantitative answers to high-stakes dilemmas.

"Can our project survive a 16-week transformer delay without triggering debt default?"

The Epsilon Answer: Epsilon binds the transformer lead time to the P6 schedule CPM, detects a 14-week slip in commercial operation, recalculates interest during construction ($1.84M added carrying cost), and checks the first-year debt coverage. It flags that the debt service coverage ratio drops from 1.34x to 1.14x, breaching the 1.25x covenant unless an interest reserve is funded upfront.

Resolved by: Theta Core CPM + Project Finance True-Up

"Should we overbuild solar + storage by 20% or contract for remote clean firm generation?"

The Epsilon Answer: PyPSA capacity optimization models the hourly 24/7 matching curves under localized solar/wind profiles. It shows that overbuilding local solar and 4-hour batteries to reach 98% CFE drives system LCOE from $54/MWh to $89/MWh due to curtailment, whereas contracting a remote 30 MW geothermal or clean firm PPA caps the blended LCOE at $62/MWh.

Resolved by: PyPSA Linear Capacity-Expansion LP

"Will our community solar portfolio forfeit IRA tax bonus depreciation if COD slips past year-end?"

The Epsilon Answer: The credit cliff evaluator maps beginning-of-construction evidence against the P50/P80 placed-in-service schedule. It quantifies the float remaining before statutory deadlines, models Section 6418 transfer discount rates, and computes the exact dollar variance ($1.2M) between keeping the 10% domestic content adder versus missing the placed-in-service window.

Resolved by: Microgrid Financial Engine + Credit Cliff Float Model

"Will our distributed VPP export violate feeder thermal limits during peak injection hours?"

The Epsilon Answer: pandapower runs an AC power flow simulation across the radial distribution topology. When 5 MW of rooftop solar and battery systems backfeed during midday minimum load, Epsilon calculates reverse power flow and identifies bus voltage rising beyond 1.05 p.u. at Bus 18, generating an alert before expensive utility interconnection studies begin.

Resolved by: pandapower AC Power Flow & Feeder Analysis
6/6 · Transparency thru What-if Scenarios

Built-in auditability, confidence scores, and provenance

Traditional models hide assumptions inside hidden spreadsheet rows. Epsilon exposes a living audit registry where every output carries explicit caveats and confidence scores.

Provenance Ledger

Assumptions Cite Sources

Every lead time, unit CAPEX cost, and discount rate carries provenance: verified vendor quote, NREL ATB index, or explicit user override. Lenders see exactly what backed every number.

Confidence Scoring

0–100 Model Health

Each engine output computes a transparent confidence score based on assumption materiality, profile fidelity (168h vs 8,760h), and cross-engine convention alignment.

Mathematical Normalization

Cross-Engine Consistency

Enforces Fisher real-to-nominal discount rate conversions and battery one-way round-trip efficiency (√RTE) across all engines so comparison reports never compare apples to oranges.