Breaking down how Arcus’s high-probability approach made capturing IESO GA peaks possible

Arcus Power successfully identified the Global Adjustment peak hours within its High Probability range. Arcus’s internal comparison shows the minimum of the top five peaks is about 10.64% above the 2024 minimum, and the smallest 2025 peak exceeds the largest 2024 peak by roughly 1,104 MW. This season was noisy to say the least. Increased curtailment activity and behind-the-meter effects complicated peak identification. Our models still delivered clear signals because it quantified uncertainty, recalibrated in real time, and tied probability windows to clear actions.

‍

‍

Global Adjustment recovers the difference between contracted or regulated resource costs and market revenues, along with other system programs. Large customers that qualify for Class A are charged GA based on their Peak Demand Factor, which depends on their contribution to the five Ontario demand peak hours in the base period. Arcus aligns peak identification to IESO’s use of Ontario Demand for ranking those hours.

The IESO defines the ICI base period as May 1 to April 30, with the subsequent adjustment period from July 1 to June 30. Organizations track the evolving top hours and confirm official values with IESO tools and documentation.

Why identification quality matters? Many teams curtail broadly on hot days. That can raise production costs, operation issues, and overtime without improving GA savings if the actions miss the five hours that set the Peak Demand Factor. A good CP strategy aims to capture all five while avoiding unnecessary curtailments. This is a probability management problem informed by clear rules and fast updates.

Takeaway: The CP mechanism rewards selective action focused on the five right hours, aligned to IESO definitions and data.

‍

What Arcus confirmed in 2025

Arcus captured every 2025 GA coincident-peak hour within its High Probability range. The five confirmed hours, all reported as hour-ending in Eastern Time using Ontario Demand, are listed below.

‍

‍

Arcus’s internal year-over-year check shows the 2025 floor moved higher: the minimum of the 2025 top five sits about 10.64% above the 2024 minimum, and the same 2025 minimum exceeds the 2024 maximum by around 1,104MW. The implication shows that the thresholds that worked last year were not sufficient this year without dynamic adjustment.

How the IESO framework anchors the analysis

IESO determines the five hours based on Ontario Demand and provides a peak tracker and related materials that organizations can use to monitor and verify the evolving top hours. Aligning strategy to these program definitions keeps the analysis consistent with how Peak Demand Factor is set.

Why 2025 differed

This summer produced the right mix of heat and humidity to push risk into the late afternoon and evening. Cooling load stayed elevated, while solar output declined late in the day. Curtailment showed up earlier for many portfolios, bending the load trace and creating "decoy hours" if masking effects were not accounted for. With more look-alike hours and tighter decision windows, the penalty for false positives increased.

Arcus’s high-probability approach

Arcus uses probability windows rather than a single-number forecast. The system integrates multiple short-horizon load models with weather-to-load data that gives weight to apparent temperature and persistence. At a high level, it estimates solar, wind, interties and curtailment signatures by hour, and it recalibrates in real time as new traces arrive. Hours are classified into identified windows using thresholds that move with the season’s evolving evidence. The High probability is a go-signal for operational actions. Other probability windows prepares assets and automation. Low remains in monitoring. The promotion and demotion logic reflects intraday weather updates, import and supply conditions, outages, and actual load bias.

Technical challenges in 2025 and how the method handled them

Curtailment: When many customers curtail early, system load can dip before a true peak is formed. The model treats curtailment as a measurable feature with expected depth and timing by sector. That reduces the risk of calling a false peak created by early action and keeps attention on hours when underlying stress is still building.

Behind-the-meter solar: Late-day solar output can hide rising 'cooling demand', then fade quickly and expose the ramp. Hourly estimates of behind-the-meter solar by region and weather keep that pattern visible in probability space.

Weather uncertainty: A small error in apparent temperature or a modest shift can swing megawatts during heat events. Ensemble meteorology and frequent intraday updates move hours between windows when warranted.

Supply, outages, and imports: Changes in available supply or imports affect risk on tight days. Real-time system data and constraint indicators feed the forecasts and real-time so probability adjusts before the decision window closes.

2025 versus 2024 - what actually changed

Arcus’s internal analysis shows a higher 2025 base and a wider gap to 2024’s top end. The minimum of 2025’s top five peaks is about 10.64% higher than the 2024 minimum, and the same 2025 minimum is roughly 1,104 MW above the 2024 maximum. Also, 2025 favored later-day peaks with stronger evening ramps. That increased the cost of late curtailment and rewarded operations that promoted to higher probability windows.

Peak-day walk-throughs: five mini case notes

June 24, HE 19, 24,862 MW, Rank 1. The hour stood out because cooling demand stayed strong into the evening while solar tapered, and real-time bias showed the ramp outperforming earlier traces in a way that fit a top rank. The High probability window stayed intact through confirmation.

August 11, HE 18, 24,788 MW, Rank 2. This hour reflected mature summer conditions with firm cooling load and less support from imports during the GA period. Intraday weather held steady on dew point, which removed a reason to demote probability.

June 23, HE 19, 24,712 MW, Rank 3. The day featured back-to-back heat, and early curtailment produced 'decoy' troughs. Curtailment inference kept focus on the hour with real system stress rather than the earlier dip.

July 24, HE 18, 24,521 MW, Rank 4. Demand outperformed the median through the afternoon and into the evening. The High probability held through the decision window.

July 28, HE 16, 24,210 MW, Rank 5. This was the earliest of the five. Solar support rolled off while temperatures stayed high, and adjacent hours did not sustain the same normalized ramp once behind-the-meter adjustments were applied.

What this means for 'Class A' C&I operations

Treat peaks as a program with triggers, assets, and verification. Response should map to probability so that the high probability windows executes a full plan and Medium windows sets assets without wasting effort. Contract alignment matters because demand response commitments and GA savings should reinforce each other rather than compete. Measurement discipline prevents rebound penalties. Metering should be tight. Restart sequences should be staggered to avoid pushing load into another costly hour.

How Arcus fits into operational workflows

Forecasts must live inside decisions. Arcus exposes probability bands through API, dashboards, and notifications so human and automated steps use one source of truth. Governance is explicit. After the season, we back-test probability bands, verify asset performance, and update thresholds to reflect the new floor.

Ontario’s demand growth and evening stress patterns will likely continue to raise the value of flexible evening assets, especially HVAC optimization, thermal storage, and process timing. Behind-the-meter activity will keep masking parts of the load trace, so operators need estimates of DER and curtailment to avoid 'decoys'. Decision windows will stay tight, which means fast updates, dynamic thresholds, and clear signals will remain essential.

‍

Three lessons stand out from 2025. Dynamic thresholds beat static rules. Treating curtailment and DER as measurable features prevents false positives. Probability windows work because they tie confidence to specific actions. Practical next steps are straightforward. Run a readiness review using this year’s playbook. Map assets to probability windows and confirm triggers and limits. Align demand response contracts with GA objectives so incentives reinforce each other for forecasts.