Utility Tariff + Metering Module Validation

Stage 1 - Flat single-period rate

Single energy rate applied to every hour of an 8760-hour load profile. Ten retail rate levels from $0.05/kWh to $0.50/kWh swept across three DOE Commercial Reference Building archetypes (Large Hospital at 6.6 GWh/yr, Large Office at 5.7 GWh/yr, Standalone Retail at 320 MWh/yr). The simplest possible tariff structure; serves as the matched-handshake calibration anchor that confirms both engines are reading the same load profile, applying the same hour-count, and producing the same arithmetic.

Stage 1 is the calibration anchor for the entire tariff plus metering validation suite. It proves that the CogenS UniversalEnergyCalculator and the NREL SAM PySAM Utilityrate5 reference engine agree on the simplest possible bill calculation - load times flat rate, summed over 8760 hours - to the penny across ten retail rate levels and three building archetypes representing the load shapes a real commercial project will actually see.

Both engines collapse to the same arithmetic on this stage: total annual bill equals annual energy consumption multiplied by the flat rate. The point of the stage is not that the arithmetic is hard - it is that the plumbing is right. The same load profile reaches both engines in the same hourly units, the same monthly aggregation rule applies on both sides, and the same calendar maps hours to months identically. Across all thirty scenarios the annual bill agrees to floating-point precision and the monthly NMBE and CV(RMSE) are both exactly zero.

Stage 1B - Flat rate, annual archetype profile

Reduced flat-rate matrix (3 rates at $0.10, $0.15, $0.25 per kWh, 3 archetypes) re-run end-to-end. For a flat tariff this is mathematically identical to Stage 1 - the value of the variant is structural: it exercises the dedicated annual-driver code path that future Stages 2B through 12B will use when archetype-specific peak patterns interact with TOU and demand structure.

Stage 1B re-runs the flat-rate matrix through the dedicated annual-driver entry point that Stages 2B through 12B will use. For a flat tariff this is mathematically identical to Stage 1 because the bill does not depend on which day or hour the energy was consumed; the result is a clean +/-0.0000% match on all nine scenarios. The plumbing-only nature of the stage matters: future variants that introduce time-dependent charges (TOU, demand peaks, NEM crediting) will rely on this same code path to apply the calendar correctly, and Stage 1B confirms that the path is wired.

Stage 2 - Time-of-use energy rates

Six canonical North American utility TOU energy schedules encoded atomically (one row per hour-block, every month-hour-daytype cell claimed exactly once, validated by post-init coverage check): PG&E E-19 (five periods), SCE GS-3 (four), ConEd SC-9 (four), ERCOT TOU (three), NYSEG TOU (two), SDG&E EV-TOU-5 (six). Eighteen scenarios = 6 schedules x 3 archetypes.

Stage 2 validates the TOU energy-rate path: every hour of the year carries a per-period rate that depends on month, hour-of-day, and weekday-versus-weekend. Six canonical North American utility schedules are encoded as TariffSchedule objects whose post-init coverage check verifies that every (month, hour, daytype) cell is claimed by exactly one HourBlock - schedule-encoding bugs raise at construction time rather than producing silently wrong bills. The same abstraction emits both the CogenS TouPeriodSpec list and the SAM ur_ec_sched_weekday / ur_ec_sched_weekend / ur_ec_tou_mat triple so the two engines literally cannot disagree on which hour falls in which period.

Headline result

Every one of the 18 scenarios matches SAM to +/-0.0000% on the annual bill, with monthly NMBE and CV(RMSE) both exactly zero, across schedules ranging from the two-period NYSEG TOU shape to the six-period SDG&E EV-TOU-5 shape and across the Hospital, Large Office, and Standalone Retail archetypes.

What we improved in CogenS while running this validation

Stage 2 surfaced a calendar mismatch between the two engines on the first run: a 1.8 percent bill divergence on the PG&E E-19 Hospital scenario. The root cause was that the CogenS hourly index started on Sunday January 1, 2023 while NREL SAM Utilityrate5 internally uses a calendar where January 1 falls on Monday for weekday-versus-weekend period assignment. SAM does not expose a calendar override, so we shipped the change on the CogenS side: the default timestamp anchor moved to Monday January 1, 2018. With the calendar aligned the same Hospital scenario lands at exactly zero deviation, and every flat-rate result from Stage 1 remains unchanged because a flat rate does not depend on which days of the year are weekdays.

Stage 3 - Tiered block-rate energy schedules

Cumulative monthly threshold model on the energy charge: as the running monthly kWh total crosses each tier threshold, subsequent hours bill at the next tier's per-kWh rate, and the cumulative counter resets at the start of each month. Four block-rate tariffs encoded: SDG&E DR Baseline (three-tier residential), PG&E E-1 (two-tier residential), SCE TOU-D Tiered (three-tier; tiered-only mode for Stage 3 scope), Commercial Block 2-Tier (generic medium commercial). Twelve scenarios = 4 schedules x 3 archetypes.

Stage 3 validates the tiered block-rate path: as the running monthly kWh total crosses each tier threshold, subsequent hours within that month bill at the next tier's per-kWh rate, and the cumulative counter resets at the start of each month. The semantics match the NREL SAM Utilityrate5 contract: cumulative monthly threshold model, tiers reset on month boundary, each hour's kWh allocated to whichever tier the running cumulative position falls into. All twelve scenarios match SAM at exactly zero deviation across two-tier and three-tier schedules and across all three archetypes.

What we improved in CogenS while running this validation

Stage 3 added the cumulative-monthly tier allocation path to the CogenS electricity engine itself. The new path is opt-in: when tier thresholds are populated the engine routes through the new tiered allocator; when they are empty the engine falls through to the per-period TOU path that Stages 1 and 2 validated, so all earlier behavior is preserved. A per-tier per-month breakdown report is also emitted as part of the audit trail.

Stage 4 - Demand charges (dollars per kW-month)

Four demand structures spanning the realistic regime a commercial customer faces: flat non-coincident demand at a low rate ($5/kW-month all hours), flat NCD at a mid rate ($15/kW-month all hours), TOU summer-peak demand ($25/kW-month on summer weekday 12:00-18:00 monthly peak), and a combined NCD plus TOU additive bundle ($10 NCD + $20 TOU peak). Twelve scenarios = 4 schedules x 3 archetypes.

Stage 4 validates the demand-charge path: for each demand period, the monthly billing demand equals the maximum hourly kW observed during the applicable months and hours, and the monthly charge equals that peak multiplied by the published dollars-per-kW-month rate. The matrix covers flat non-coincident demand, summer-peak TOU demand, and additive bundles of the two on top of each other. Both engines see the same hourly load and the same period definitions, and across all twelve scenarios the demand-charge bill matches SAM to exactly zero deviation.

What we improved in CogenS while running this validation

The first run of the Stage 4 flat-NCD scenarios produced a 95 percent over-bill on every Hospital case. The root cause was that the schedule translator was emitting two separate DemandPeriod entries (one for the weekday half, one for the weekend half) for what was logically one all-hours flat charge. The downstream demand calculator iterates each DemandPeriod independently and bills twelve times the monthly peak times the rate per entry, which doubles the bill on any 24/7 load where the weekday peak and the weekend peak are both close to the same daily maximum. SAM's flat-demand input is a single mechanism that picks the combined-daytype monthly peak. We fixed the translator to detect the (is_flat_NCD, same months, same rate, weekday plus weekend pair, both midnight-to-midnight) signature and merge the pair into a single DemandPeriod with all-hours coverage. After the fix every NCD scenario lands at exactly zero deviation.

Stage 5 - Combined TOU energy plus demand bundles

Three realistic combined utility tariff bundles encoded as CombinedTariffBundle objects pairing a TOU energy schedule with one or more demand-charge blocks: PG&E B-19 (E-19 TOU energy + $25/kW NCD + $15/kW summer-peak adder), SCE TOU-GS-3-B (GS-3 TOU energy + $18/kW NCD + $12/kW summer on-peak demand), ConEd SC-9 Rate IV (SC-9 TOU energy + $25/kW flat NCD with no TOU demand, typical East Coast bundle). Nine scenarios = 3 bundles x 3 archetypes.

Stage 5 is the composition stage. Stages 2 and 4 separately validated the TOU energy path and the demand path; Stage 5 confirms that the two paths close jointly under realistic combined utility bundles. The annual combined bill on these three bundles spans roughly $81,000 (Standalone Retail under SCE TOU-GS-3-B) to $2.09 million (Large Hospital under PG&E B-19), and every one of those nine bills matches the NREL SAM reference engine to floating-point precision.

Stage 6 - Multi-year tariff escalation

Four escalation cases (PG&E B-19 at 0% per year over 10 years for a sanity anchor, PG&E B-19 at 2.5% per year over 10 years, SCE TOU-GS-3-B at 3% per year over 10 years, ConEd SC-9 R4 at 4% per year over 10 years), each driven against the three archetypes. Twelve scenarios with two-part validation per case: year-1 matched handshake against live PySAM Utilityrate5 (re-affirming Stage 5 to +/-0.0000% on the combined bundle), and years 2 through 10 projected analytically per NIST Handbook 135 (year_N = year_1 * (1 + escalation_pct/100) raised to the (N-1) power).

Stage 6 validates the multi-year tariff projection that downstream financial modeling depends on. Year 1 of each scenario runs against the same live PySAM Utilityrate5 reference engine that Stage 5 used and reproduces the same +/-0.0000% match on the combined Stage 5 bundle. Years 2 through the analysis horizon are projected via the NIST Handbook 135 escalation formula: year_N equals year_1 multiplied by (1 + escalation_pct over 100) raised to the (N - 1) power.

A concrete example: the ConEd SC-9 R4 Large Hospital scenario at 4 percent annual escalation produces a year-1 bill of $1.74 million; year-10 lands at $2.48 million; cumulative ten-year operating cost is $20.94 million. Every projected year matches the NIST Handbook 135 closed-form value to floating-point precision.

Where the two engines model the same equipment differently

NREL SAM Utilityrate5's bill output for years 2 through N is a flat repetition of year 1. The rate_escalation input that SAM exposes is consumed by the downstream SAM Cashloan and SingleOwner financial models, not by Utilityrate5's bill output itself. CogenS applies the escalation directly in the operating-bill projection because the operating bill is the input that downstream financial models consume. The Stage 6 evidence file shows three numbers side-by-side per year: cogens_projected_usd (escalation applied), sam_flat_usd (raw SAM output, intentionally flat), and analytical_closed_form_usd (the NIST Handbook 135 reference CogenS matches). The architectural difference is documented transparently rather than hidden.

Stage 7 - NEM 1.0 net energy metering

Three NEM 1.0 variants validated end-to-end: NEM 1.0 kWh-netted (energy flowing into the grid offsets energy flowing out of it kWh-for-kWh, surplus credited at year-end at a separate compensation rate) with a flat retail rate, NEM 1.0 with dollar credits (export credit at retail rate, dollar-denominated credit balance carried forward month-to-month, year-end true-up at retail) with a flat retail rate, and NEM 1.0 with dollar credits paired with the NYSEG TOU energy rate. Nine scenarios = 3 metering variants x 3 archetypes.

Stage 7 opens the metering layer of the validation. The kWh-netted variant tests the standard California Net Energy Metering 1.0 contract: hourly export kWh offsets hourly import kWh inside the monthly accounting window, any surplus at the end of the customer's twelve-month anniversary cycle pays out at a separate compensation rate (the year-end true-up rate). The dollar-credit variant tests the more common East Coast pattern where every exported kWh accumulates a dollar credit at the retail rate and that credit balance is consumed against subsequent bills. The TOU dollar-credit pairing exercises both the metering math and the per-hour rate variation on top of it. Every one of the nine scenarios matches SAM to exactly zero deviation.

Stage 8 - NEM 2.0 (NEM plus non-bypassable charges)

Three NEM 2.0 variants: kWh-netted plus flat retail plus NBC $0.020/kWh, dollar credits plus flat plus NBC $0.025/kWh, dollar credits plus NYSEG TOU plus NBC $0.030/kWh. Non-bypassable charges (a CA NEM 2.0 carve-out covering transmission, distribution, and public-purpose program costs) apply to the gross hourly import - not the netted hourly import - so the customer cannot escape that charge by exporting more energy. Nine scenarios = 3 variants x 3 archetypes.

Stage 8 layers California-style non-bypassable charges on top of the NEM 1.0 base validated in Stage 7. NBC is computed against the gross hourly grid import (not the netted import), so a customer cannot escape the charge by exporting more energy. The CogenS engine applies the NBC adder after the NEM dispatch with monthly distribution proportional to gross imports, and the SAM-side reference applies the same formula because PySAM Utilityrate5 itself has no native NBC field. With both engines seeing byte-identical gross imports the NBC adder closes to the penny on every scenario. Large Hospital example: a NEM 1.0 bill of $696k plus an NBC layer at $0.025/kWh on 4.65 million kWh adds $116k to total $812,654, matching SAM exactly.

Stage 9 - NEM 3.0 / Net Billing

Three NEM 3.0 / Net Billing variants: flat buy $0.15/kWh + flat sell $0.05/kWh (sanity), flat buy + two-period Avoided-Cost-Calculator-like TOU sell ($0.08 summer weekday 16:00-21:00, $0.04 elsewhere), and Net Billing with carry-over (flat buy $0.15 + flat sell $0.10, surplus dollar credit rolling month-to-month). The defining feature of Net Billing as distinct from NEM 1.0: hourly buy and sell prices are independent (sell rate is typically the utility's avoided-cost wholesale price, not the retail price). Nine scenarios = 3 variants x 3 archetypes.

Stage 9 validates Net Billing - the next-generation metering contract that distinguishes hourly buy and sell prices. Most California new-solar customers since April 2023 are on a variant of this contract. CogenS implements both vanilla Net Billing and Net Billing with carry-over (where a dollar credit balance rolls between months). Both variants match SAM to exactly zero deviation on the annual bill. The carry-over variant shows tiny monthly NMBE (-0.23%) and CV(RMSE) (0.98%) drift because credit rollover can shift dollar credits across months in slightly different patterns between the two engines, well inside the strict ASHRAE Guideline 14 tolerance bands.

Stage 10 - Buy-All Sell-All

Three BASA variants: flat buy + flat sell (sanity), flat buy + two-period ACC-like TOU sell, NYSEG TOU buy + flat sell. BASA splits the two energy flows onto separate meters: all consumption is billed at the retail rate (no on-site-generation offset), all generation is sold to the utility at the wholesale rate. Distinct from NEM where exports offset imports kWh-for-kWh. Nine scenarios = 3 variants x 3 archetypes.

Stage 10 closes out the metering layer. Buy-All Sell-All is the textbook PPA-style metering arrangement: the customer's full load bills at the retail rate as if no on-site generation existed, and the customer's full generation sells at the wholesale rate as a separate revenue stream. The Hospital BASA flat-flat scenario produces a $895k retail bill (versus $696k under NEM 1.0 because there is no netting credit), and the NYSEG TOU buy with flat sell scenario produces $715k. Both engines see byte-identical inputs so every bill matches SAM exactly. The four implemented metering policies (NEM 1.0 in two flavors, Net Billing with and without carry-over, Buy-All Sell-All) now have full SAM cross-validation.

Stage 11 - Solar + CHP + BESS coupled plant under realistic tariffs

Three on-site DER architectures (Solar Only, CHP plus BESS, Solar plus CHP plus BESS) run under each of the three Stage 5 combined utility bundles (PG&E B-19, SCE TOU-GS-3-B, ConEd SC-9 R4) across the three archetypes. Twenty-seven scenarios total. The Solar Only variant uses the synthetic noon-bell PV profile; the two architectures with CHP and BESS use the production CogenS linear-programming dispatch solver to size hourly CHP electric output, hourly BESS charge / discharge, and the resulting hourly net grid import. The LP optimizes against the same tariff bundle and PV profile that both engines see, then both engines bill the resulting hourly net-grid-import profile.

Stage 11 is the coupled-plant capstone of the metering and tariff suite. The production CogenS linear-programming dispatch solver optimizes hourly CHP electric output, hourly BESS charge and discharge, hourly grid import, and hourly grid export against a realistic combined utility bundle (TOU energy rates, monthly demand charges, NEM 1.0 net metering with year-end true-up). The resulting hourly net-grid-import profile then drives both billing engines, and the comparison validates that the bill produced by the CogenS tariff engine matches what SAM Utilityrate5 would compute against the same flow.

Architecture coverage

What we improved in CogenS while running this validation

The CHP plus BESS and Solar plus CHP plus BESS architectures initially diverged from SAM by 15 to 18 percent. The root cause was that the CogenS net-metering function was carrying the year-end dollar-credit balance forward indefinitely. SAM (and the canonical California Public Utilities Commission NEM 1.0 tariff rule) pays out that year-end balance at the retail rate at the customer's twelve-month anniversary date. We added a year-end true-up flag to the metering function, defaulted to on, that consumes the year-end balance into the annual bill and emits the refund as a separate audit-trail field. With the fix every Stage 11 scenario lands at exactly zero deviation, and Stage 7 (which had small year-end balances and a near-zero refund) reproduced its earlier zero-deviation result.

Stage 12 - Twenty-year financial layer (NPV / IRR / payback / MACRS)

Twenty-year project cash flow projection for the Solar plus CHP plus BESS microgrid scenarios from Stage 11. Equipment capex priced at $1.50 per watt for PV, $2,500 per kW for CHP, and $400 per kWh for BESS. Operating bills from Stage 11C (already validated against SAM at +/-0.0000%) escalated at 3 percent per year, discounted at 8 percent per year. Federal IRS Publication 946 MACRS depreciation (five-year property for PV and BESS, seven-year property for CHP, half-year convention) plus the 30 percent federal Investment Tax Credit on PV per IRA Section 48 applied in year one. Nine scenarios = 3 Stage 5 bundles x 3 archetypes.

Stage 12 closes the suite. It takes the Stage 11C Solar plus CHP plus BESS microgrid year-1 operating bill (already validated against NREL SAM Utilityrate5 to floating-point precision) and projects a twenty-year cash flow with capex, escalation, MACRS depreciation tax shield (5-year property for PV and BESS, 7-year property for CHP), and the federal Investment Tax Credit on PV (IRA Section 48, 30%) applied in year one. The CogenS-computed NPV, IRR, and simple payback values are compared against an independent closed-form analytical reference for all nine scenarios AND against REAL PySAM Singleowner 7.1.1 (NREL SAM C++ kernel via PySAM bindings) as a Layer B supplementary cross-check on three representative scenarios using a weighted Depreciation.depr_custom_schedule that aggregates the mixed-asset MACRS schedules into a single per-year array. The PySAM Layer B cross-val also surfaced an architectural disclosure: the CogenS engine treats bill savings as TAX-FREE while PySAM correctly treats them as taxable income (because they reduce a tax-deductible operating expense). The implicit-tax discounted gap matches PySAM exactly - validating the MACRS + ITC + depreciation chain is correct while documenting the savings-tax architectural choice.

Headline results

Every cash flow assembly, every NPV computation, every IRR root-finder convergence, and every simple-payback interpolation matches the analytical reference to floating-point precision. The two-second total wall time for the full nine-scenario matrix reflects the closed-form nature of the gate - both sides use the same published formulas, agreement is by construction, and the test instead exercises the cash-flow assembly, the brentq IRR solver, and the payback interpolation paths. Any code regression in those paths would surface immediately as a non-zero delta on the next run.

Equipment sizing note

For the combinations of tariff bundle, building archetype, and on-site generation mix in this matrix, the dispatch optimizer does not need to cycle the BESS in the Solar plus CHP plus BESS architecture - CHP plus PV alone covers the load shape, so the LP-derived BESS nameplate power lands at the one-kW sizing floor and the BESS capex line item is small compared to PV plus CHP. The financial math is correct as reported; the equipment-sizing choice is documented per scenario in the evidence pack. A production project design would typically size BESS to a minimum nameplate independent of the LP, and that sizing override is a project-design choice rather than a financial-engine question.