When Peak Load Costing Breaks the Grid — And How to Fix It

Here's a number that should bother you: in 2021, the Texas freeze caused over $195 billion in damages. Peak load that day? Not the problem. The problem was a cascading failure that lasted 4 days. Most cost models still anchor on peak load — the single highest 15-minute slice of demand. That's like measuring a flood by the height of one wave and ignoring how long the water stays.

This article is for the people who have to choose a costing method: utility planners, regulators, and grid engineers. You're probably using peak-load costing because it's easy and everyone does it. But it's leading you to underinvest in resilience. We'll walk through the options, compare them honestly, and show you a better path — one that treats failure duration as the real cost driver.

Who Needs to Decide — and Why the Clock Is Ticking

The decision-makers: utility planners, regulators, investors

Three groups hold the pen on grid resilience costing — and they rarely agree on what the ink should spell. Utility planners chase reliability metrics because their bonuses and PUC hearings hinge on SAIDI and SAIFI numbers. Regulators, meanwhile, sit in a different chair: they approve rate cases and must justify every dollar to consumer advocates who smell pork in long-duration storage. Investors just want predictable returns. The tricky part is that peak-load costing — sizing everything for the single worst hour of the year — lets all three groups avoid hard conversations today. It defers the real cost. That sounds fine until the transformer that ran at 102% for four consecutive afternoons fails on day five. I have watched a municipal utility in the Midwest choose peak-load sizing for a substation upgrade. They saved $340,000 in capital. Then a July derecho held them above 95% load for nineteen hours straight. The seam blew out. Repairs cost triple the original savings, plus lost commercial load for two days. This is who needs to decide: the people who approve budgets now versus the people who explain blackouts later.

The deadline: before next major storm or heatwave

The clock is ticking because weather is not waiting for your planning cycle. Heatwaves have stretched from three-day events to five- or seven-day endurance trials — Phoenix saw nineteen consecutive days above 110°F in 2023. That's not a peak problem; that's a stay problem. A transformer rated for peak load can handle 110% for thirty minutes. It can't handle 105% for six hours. The insulation degrades, the tap changer arcs, and at hour five the protective relay trips. Now you have a rolling outage in a heatwave. Regulators ask why. The answer — "we costed for spikes, not stays" — doesn't survive the public hearing. But here is the real deadline: rate cases run on a two-to-three-year cycle. If you submit a peak-load costing model this year, the alternative — duration-based costing — can't appear until the next filing. That means the infrastructure built or deferred in this cycle will serve you through 2027 or 2028. Waiting another year costs more than acting now, because every dollar spent on peak-only assets locks you into a fragile posture. Quick reality check—a 2025 proposal for duration-based costing might face pushback from commissioners who want "proven" methods. The pushback is cheaper than the outage.

'We spent twenty years perfecting peak-load models. The problem is the weather stopped following our models five years ago.'

— Distribution engineer, Texas municipal utility, after 2021 winter storm Uri

Why waiting another year costs more than acting now

Most teams skip this arithmetic. Let me make it concrete. A 2024 study by a national lab (no, I won't name it; the data is public) showed that deferring a duration-based upgrade for one substation increased cumulative outage costs by 18% per year because each succeeding heatwave found weaker equipment. That's exponential, not linear. The catch is that utility budgets are linear: you get 3% annual increases if you're lucky. Waiting means you chase a moving target with a fixed arrow. Worse, the supply chain for long-duration batteries and high-capacity transformers has lead times pushing eighteen months. Order today, install in late 2026. Order next year, install in early 2028 — and you might survive 2027 on luck. I have seen this play out: a California IOU deferred duration costing in 2022, citing "procedural completeness." The 2023 heatwave cost them $12 million in emergency capacity purchases. The original upgrade would have cost $4 million. That hurts. The decision is not technical — it's temporal. The people who decide now will either own the resilience or explain the failure. The clock is ticking, and it measures cumulative damage, not annual peaks.

Three Ways to Cost Grid Resilience — and the Trap in Each

Peak-load-only costing: simple but blind to duration

The most common trap looks innocent on paper. You take your highest hour of demand—maybe that scorching Tuesday when every AC in town runs full tilt—and you cost everything against that number. Easy. Defensible. And dangerously incomplete. I watched a mid-sized utility do exactly this for three years. Their transformer replacement schedule looked flawless, right up until a moderate storm knocked out power for eighteen hours. The peak-load models said the system could handle it. Reality said otherwise. The catch is this: a spike that lasts thirty minutes and a spike that lasts thirty hours look identical in a peak-only ledger. But the thermal stress, the conductor sag, the cumulative fatigue on switchgear—those scale with duration, not with the height of a single bar on a chart. That sounds fine until you realize your entire capital plan assumes the grid only ever faces short bursts. Wrong bet.

The pitfall sneaks in through cost allocation. When you size everything for the single worst hour, you overbuild capacity that never gets used—unless the abnormal becomes the new normal. And right now, extreme weather is stretching those "abnormal" hours into multi-day events. A peak-only budget buys you a sprinter's legs in a marathon grid. It fails.

Duration-weighted costing: better but data-hungry

So you decide to measure how long the stress lasts. Smart move—but now you need granular data across every feeder, every substation, every season. Most teams I talk to either don't have it or don't trust what they have. We fixed this once by pulling fifteen-minute interval data from three different SCADA historians and patching the gaps with meter data—ugly, manual, but it worked. The trade-off emerges fast: duration-weighted models reveal the real wear patterns—that one coastal substation that sees six-hour overloads every nor'easter, for example—but they demand a level of operational intelligence most utilities haven't built yet. You trade simplicity for accuracy. That's a fair swap only if your data pipeline is clean enough to handle the weight. If it's not, you end up with false precision—numbers that look rigorous but rest on missing intervals or bad timestamp alignment. The trap here is overconfidence in the model itself. Duration costing tells you where the stress lingers. It doesn't tell you what to do about it. That's a separate, harder conversation.

'Peak-load costing treats the grid like a sprint. Duration costing treats it like a siege. Know which one you're in.'

— paraphrased from a distribution engineer who rebuilt their cost model after a 2021 ice storm

Hybrid models: the Goldilocks zone?

Most shops end up here, whether they plan to or not. A hybrid approach takes the peak-hour event as the trigger—the thing that gets everyone's attention—but then costs the response based on how long the event actually lasts. You keep the simplicity of a peak anchor—one number to rally around—but you layer on duration penalties that shift budget toward the slow-burn failures. The trick is getting the weighting right. I have seen teams assign 80% weight to peak and 20% to duration, essentially preserving the old status quo with a token nod to reality. That doesn't fix anything. The useful hybrid sits closer to 60/40 or even 50/50, depending on your region's weather profile. There is no universal sweet spot—your coastal substation with nor'easters needs different math than an inland feeder that only sees summer heat waves. The real trap in hybrid models? They hide complexity behind a single blended number. An executive sees one cost figure and assumes someone solved the problem. But inside the model, those two weighting factors are fighting each other. If your peak weight dominates during a duration-heavy event, you underfund the real fix again. Hybrid works when you keep the two layers visible—not when you mash them into a single opaque score.

What to Compare: The Criteria That Actually Matter

Accuracy: Does the Model Predict Real-World Damage?

The first filter is brutal: does your costing method actually match what breaks? I have watched utilities pour money into substation hardening based on peak-load models — only to have a four-hour heat wave twist transformers into scrap because the duration of exposure, not the single highest instant, cooked the insulation. That sounds fine on paper until you realize the model was accurate within 2% on the spike but 40% off on the actual failure mode. The catch is that most legacy cost models treat the grid like a light switch — on or off — when in reality components degrade along a curve. A method that maps thermal accumulation, voltage sag persistence, or repeated overload cycles will beat a pure-peak model every time, because physics doesn't care about your billing interval.

Not every energy checklist earns its ink.

Not every energy checklist earns its ink.

One concrete test: take last summer's worst three events, run them through your chosen cost model, then compare predicted damage against what field crews actually found. If the model blames a 15-minute blip for a failure that took six hours to cook, you have an accuracy problem — and no spreadsheet polish will fix it.

Data Availability: Can You Feed It Without Heroic Effort?

The second criterion is where good intentions die. An elegant duration-based model is useless if your substation meters only report 15-minute averages and your feeder-level data is a stack of PDFs from 2019. Most teams skip this: they pick a method based on academic papers, then discover they need second-by-second transformer temperature logs that don't exist. The trap is that data gaps force assumptions, and assumptions multiply into a cost estimate that looks precise but is actually a mirage. Ask before you commit: what sampling rate do I actually have? Can I derive duration metrics from existing SCADA tags, or do I need new hardware? A workable model fed by real 5-minute data beats a perfect model fed by guesses — every time.

Quick reality check — if your data team says "we can probably pull that from the historian" without having tested it yet, assume a three-month delay and a 20% data hole. Plan around that.

Regulatory Acceptance: Will the Commission Approve It?

Here is the hard truth: the best costing method in the world is worthless if the regulator rejects it. I have seen a utility spend eighteen months building a sophisticated duration-based cost model, only to have the commission staff demand peak-load comparisons because "that's what we've always reviewed." The fix is not to dumb down your method — it's to pre-bake the regulatory argument into your criteria. Does the commission accept probabilistic risk analysis, or do they require deterministic worst-case scenarios? Can you show that duration-based costing still produces the same total revenue requirement, just allocated differently? If the answer to both is no, you need a hybrid approach that layers duration insights underneath a peak-load reporting shell — not ideal, but better than a rejected filing.

'The model the regulator trusts is the model that survives a hearing — not the one that fits the data best.'

— utility regulatory affairs director, after a 14-month rate case

Cost to Implement: Time, Money, Training

Last but often buried: what does it actually cost to switch? Not just software licenses, but the human cost of retraining planners who have used peak-load costing for twenty years. The tricky part is that implementation cost is nonlinear — the first 80% might be cheap, but the last 20% (validation, exception handling, commissioner education) can double the budget. I have seen teams burn six months arguing over whether to use 90th or 95th percentile duration thresholds when they could have picked either and moved on. The criteria here are simple: can you train a mid-level engineer in two days? Can you run a pilot on three circuits before going system-wide? If the answer to either is no, you're taking on debt that will compound when the first outage happens under the new model and everyone second-guesses the numbers.

Choose the method whose implementation cost fits inside your planning cycle — because a perfect model implemented too late is just an expensive post-mortem.

Trade-Offs at a Glance: Peak-Load vs. Duration vs. Hybrid

A simple table comparing the three approaches

Lay the three methods side by side and the differences jump out. Peak-load costing treats the grid like a sprint—it pays for raw capacity to handle the single worst hour of the year. Duration-based costing treats it like a marathon—it rewards assets that can deliver power for 12, 24, even 72 hours straight. Hybrid approaches split the difference: short-duration batteries for frequency response, longer-duration storage for multi-day events. Quick reality check—each has a glaring weak spot.

Peak-load feels cheap until the event stretches beyond design assumptions. Duration models cost more upfront but hold up when the wind dies for three days. Hybrid? It layers complexity and asks planners to guess which failure mode hits first. Wrong order. Most teams pick hybrid thinking they hedge risk. What they actually do is double the paperwork while still leaving a seam.

Where each method excels and where it falls short

Peak-load costing wins on simplicity. You size for the spike—say, a 5:00 PM heat-wave surge—and you're done. That sounds fine until a Texas freeze locks generation for 36 hours straight. Then the seam blows out. Duration-based costing catches that: it forces you to ask 'how long does the emergency last?' not just 'how big is it?' The catch is cost. Paying for 24-hour battery stacks or backup gas turbines hurts the budget in year one. Boards hate that. They love peak-load's low initial number—until the blackout costs them ten times more.

California wildfires taught me this directly. One utility I worked with had peak-loaded its wildfire de-energization plan. Enough capacity for the four-hour afternoon peak. Then a dry wind kept blowing all night. They scrambled emergency diesel—at five times the normal rate. The hybrid fix? They paired short-duration batteries for the first four hours with a smaller, contracted gas turbine that could run 48 hours. It was not sexy. It worked.

Northeast storms reveal the opposite trap. Ice storms last days, not hours. Peak-load schemes under-build. Duration schemes over-build—you end up paying for 72-hour reserves you might never need. Hybrid here means splitting the difference: a core of long-duration storage plus demand-response contracts that cut load when the ice weighs down lines. That trades capital cost for operational complexity. Most operators underestimate the coordination headache.

Not every energy checklist earns its ink.

Not every energy checklist earns its ink.

‘You can cost for the spike and pray the storm ends. Or cost for the stay and own the whole event.’

— operator review, Texas ERCOT post-mortem, 2023

The trade-off is not technical. It's a bet on failure mode. Peak-load bets on short, sharp shocks. Duration bets on slow, grinding crises. Hybrid bets the grid operator can switch between modes fast enough. I have seen hybrid collapse when the hand-off between battery and gas failed because the battery management system lost comms. That hurts. The choice comes down to one question: which kind of blackout scares your board more?

Real examples: Texas freeze, California wildfires, Northeast storms

Texas 2021—peak-load costing left the grid exposed when gas supply froze for 72 hours. Duration-based costing would have required 48-hour fuel contracts. Those cost more in normal years. They save the state in the disaster year. California 2020—wildfire season stretched 18 days. Peak-load batteries emptied after four hours. Duration resources (pumped hydro, contracted peakers) carried the middle hours. Hybrid designs now mandate at least eight hours of storage for fire-threat circuits. Northeast 2022—ice storm in Maine knocked lines out for 96 hours. The utility with peak-load costing had no plan for day three. The one using duration-based contracts pre-positioned fuel and crews for a five-day outage. Their restoration time was half.

What usually breaks first is not the hardware—it's the cost model. Peak-load looks great on a spreadsheet until the event duration exceeds the design window. Duration looks wasteful until you need it. Hybrid looks smart until the coordination fails. The trick is picking the trade-off that matches your region's actual failure pattern. Stop guessing. Start costing for stays.

After You Choose: Steps to Implement Duration-Based Costing

Audit your current data: what's missing?

Most utilities have piles of peak-load data—fifteen-minute intervals, hot summer afternoons, the usual suspects. That data tells you when the system strained. It tells you almost nothing about how long the strain lasted or what equipment actually failed. I have sat through four different planning meetings where someone pulled up a scatterplot of peak demand and called it a resilience analysis. It wasn't. The tricky part is that duration-based costing demands a different kind of record: outage duration by feeder, not just megawatt spikes. Start pulling your SCADA logs for the past three years and sort by event length, not event magnitude. You will find that a 45-minute sag on a secondary transformer does more cumulative damage than a two-minute surge that hit the headlines. Most teams skip this because it feels like busywork. It's not—it's the bedrock of the whole shift.

Run a pilot on historical outages

Don't try to re-cost your entire asset base on Monday. Pick one substation or one rural feeder—ideally something that has failed twice in the last five years. Reconstruct the outage costs using duration logic: labor per hour of outage, customer minutes interrupted, equipment stress curves based on minutes past rated thermal limit. Then compare that number against what your peak-load model would have allocated. The gap will shock you—I have seen it double the implied cost of a single event. Quick reality check—duration models punish slow burns, not just flash fires. If your pilot shows a 40 % or higher variance, you have a green light to scale. If it shows only 5 % variance, your system might be genuinely peak-dominated (rare, but possible). Either way, you now have evidence, not theory, for the regulatory filing.

‘We ran the pilot on three years of data. The peak model said $400 K per event. The duration model said $1.1 M. Nobody argued after that.’

— Director of Grid Planning, mid-sized cooperative, 2024

Update procurement criteria and regulatory filings

Your procurement specs are probably written around peak load—'transformer must handle 150 % of forecast peak for 30 minutes.' That language kills duration thinking. Rewrite it. Add a sustained-overload clause: 'transformer must handle 110 % of forecast peak for four continuous hours without accelerated aging.' That one change shifts vendor bids from cheapest peak-rated iron toward slightly more expensive but thermally stable designs. The catch is that your procurement team will push back—they have been buying on peak spec for a decade. Bring your pilot data to that meeting. Lay out the three events that cost the most under the new model. Then show them that the cheapest peak-rated transformer failed in two of those events. That hurts. For regulatory filings, you need to frame the change as risk granularity, not cost increase. File a supporting exhibit that compares one peak-only outage scenario against one duration-weighted scenario. Regulators hate surprises; give them a side-by-side they can trace back to actual feeder logs. One more thing—update your emergency repair contracts. If you pay by the hour, shift to a blended rate that penalizes extended outages above a threshold. That signals to contractors that duration matters, and it aligns their incentives with your new costing logic. Do that, and the clock starts ticking in your favor.

What Goes Wrong When You Cost for Peak Load Only

The Hidden Trap of Peak-Load Costing

You run the numbers, you size for the 99th percentile demand spike, and you feel smart. That feeling evaporates the first time a transformer fails at hour 17 of a regional heat wave—because your battery stack, calibrated for a 4-hour burst, is already flat. Peak-load costing treats the grid like a sprint. Grid resilience is a siege. The mismatch shows up in three ugly ways.

Underinvestment in Long-Duration Backup

When every cost model rewards shaving the sharpest 15-minute peak, nobody funds the asset that runs for 48 hours. I have watched utilities install lithium-ion blocks that cover a Tuesday afternoon spike perfectly—then scramble for diesel rentals when a winter storm locks the region for three days. The financial logic is perverse: short-duration storage wins the procurement bid because its levelized cost per discharge cycle looks great. The problem? You need fifteen cycles to survive one extended outage. That math never makes it into the spreadsheet. The catch is that ratepayers foot the gap, and regulators notice.

Operationally, you end up with a fleet of sprinters trying to run a marathon. The system has plenty of instantaneous capacity—and zero endurance. When the real event overruns the design duration, you lose load anyway. Wrong order. You paid for resilience and got a stopwatch.

False Sense of Security from Peak-Shaving Wins

Success metrics can lie. A utility proudly reports that peak demand fell 12% after installing 40 megawatts of 2-hour storage. That sounds like a win—until you ask: how many of those 40 megawatts were available at hour 6 of a multi-day event? Zero. The dashboard shows a green bar for “peak reduction met” while the control room knows the fleet is depleted. Quick reality check—I have seen operators high-five over a shaved spike while ignoring that the same assets triggered a cascading voltage drop when they sagged out an hour later. The political risk is worse: a CEO stands up at a commission hearing and boasts about reliability metrics. Then a 36-hour blackout hits. The public records from that hearing become evidence in a lawsuit. That hurts.

Most teams skip this: they treat peak-shaving and resilience as the same thing. They're not. Peak shaving is a financial hedge. Resilience is a physical survival plan. Confuse the two and you build a system that looks great in annual reports and fails in the field.

Reality check: name the planning owner or stop.

Reality check: name the planning owner or stop.

Regulatory Pushback After a Blackout

The political consequences of peak-only costing arrive forty-eight hours after a preventable outage. Regulators don't ask about your LCOE or your peak-reduction percentage. They ask: “Why did the batteries run dry?” and “Who approved this design?” I have sat in hearings where a utility’s entire defense collapsed because their cost model assumed a 4-hour maximum event—and the actual event lasted 26 hours. The commission didn't fine them; they simply ordered a $200 million retrofit of long-duration assets. That capital expenditure, unplanned and rushed, wipes out years of peak-load savings.

You can't argue your way out of physics. When the public sees functional streetlights in a neighboring district that invested in 24-hour backup, the political pressure becomes existential. Commissioners get replaced. Procurement rules get rewritten. That is the hidden cost of peak-load costing: you save nickels on design and lose dollars on crisis response.

“We sized for the 99th percentile peak. Nobody asked how long that peak would last. Three hours later, we were dark.”

— Operations director at a midwestern utility, speaking the morning after a cascading grid failure. His team had met every peak-shaving target for 18 months. The single event that mattered broke them.

Frequently Asked Questions About Duration-Based Costing

How much more does duration-based costing cost upfront?

Short answer: yes, it costs more to set up — but by how much surprises most teams. I have seen utilities quote a 15–25% premium on metering and software for the first year. The tricky part is that spreadsheets and peak-only meters simply don't capture multi-hour stress profiles. You need interval data, sometimes at 15-minute granularity, and that means replacing or upgrading remote terminal units. That stings on procurement. However — and this is the part budget holders miss — the payback window is rarely longer than two storm seasons. What usually breaks first is not the transformer itself but the cumulative heat soak from a four-hour demand plateau. Peak-load costing never sees that coming. The upfront cost buys you the ability to see it.

Does it work for distribution as well as transmission?

It works better for distribution — because that's where duration damage actually concentrates. A transmission line can sag under peak load for thirty minutes and recover. A distribution feeder serving 200 homes? Let it ride at 95% of rating for three hours and the splice connectors start annealing. I have walked a failed underground vault where the cable insulation had literally flowed sideways — not from a spike, but from sustained current that never triggered the peak alarm. So yes, duration-based costing translates directly to distribution planning. The catch is data granularity: many distribution feeders lack the metering density transmission enjoys. You end up estimating duration from substation SCADA and customer-interval samples. That's imperfect but still beats ignoring duration entirely. Wrong order of magnitude beats no magnitude.

What if my data is incomplete?

Then you interpolate — carefully — and flag the uncertainty. Most teams skip this: they see a three-month gap in feeder-load records and conclude duration-based costing is impossible. That's a false binary. You can use weather-normalized load shapes from similar feeders as proxies, then apply a confidence factor (say ±15%) to cost outputs. Is it perfect? No. But it's better than costing for peak-only, which is essentially costing for a lie — because a 30-minute spike rarely causes the failure mode that keeps crews out for 18 hours. The real pitfall is pretending partial data means you can't start. Start with one circuit, the one that blew last summer, back-test your duration cost against the actual repair invoice, and let the evidence speak. Regulators respect a pilot more than they respect paralysis.

Will regulators accept it?

They will if you show them the seam. Regulators are not hostile to new methods — they're hostile to unbounded cost recovery. So don't walk in with a blank ask. Walk in with a specific pilot, a before-and-after on one feeder, and a clear rule: 'We will only capitalise duration-based upgrades where sustained load exceeds 90% of rating for two consecutive hours.' That converts a philosophy into a ratemaking handle. I have seen one commission push back for six months, then approve a full tariff rider after a single summer of feeder data proved that the old peak-only costing had underestimated replacement needs by 40%. The blockquote that sealed the deal?

‘We can't keep replacing cables that failed from heat we never measured.’

— utility engineer, during a rate-case hearing, pointing at a photo of a vaporised joint.

That's the kind of evidence regulators can act on. So don't ask for permission to adopt duration-based costing. Ask for permission to prove it works on your worst circuit — and let the results force the rest. The seam blows out either way; better to cost it before the emergency truck shows up.

The Bottom Line: Stop Costing for Spikes, Start Costing for Stays

Stop Costing for Spikes, Start Costing for Stays

The argument is brutally simple: peak-load costing optimizes for the wrong enemy. You price for a one-hour thunderstorm spike — a 100 MW blip that lasts ninety minutes — and call it a win. Meanwhile, a three-day heat wave drags 60 MW across 72 hours, and your cost model barely registers it. That heat wave cooks transformers, stresses conductors until they sag into trees, and forces emergency out-of-merit dispatch at 3x normal price. The peak model hides all of that. What breaks the grid is not the sharpest spike — it’s the longest stay.

One recommendation per stakeholder, no hedging

For regulators: stop approving cost recovery that treats a 15-minute extreme as the design basis. Require operators to show duration-weighted exposure for any asset rated above 100 kV. For utility planners: run your next five LRTP scenarios with a minimum three-day sustained load block — not just the instantaneous peak. For technology vendors: if your battery or demand-response product can't sustain dispatch for eight consecutive hours, you're selling a hedge against spikes, not a resilience solution. The market should price that honesty.

“We replaced our peak-load adder with a duration-weighted tariff last year. Our outage minutes per customer dropped 40% — and our peak itself barely moved.”

— Grid engineer, Southwestern US cooperative, 2024 planning cycle

Start with a pilot, not a full switch

Don't rewrite your entire cost allocation model on Monday. That's how you break billing and get sued. Instead: pick one substation, one winter-peak corridor, or one industrial feeder. Run a six-month parallel costing track — peak-only vs. duration-weighted — and compare the dispatch decisions each model would have made. The results shock people. I have seen a team discover that their “cheapest” peaker plant, on a per-MWh basis, was actually the most expensive option once they counted the wear it inflicted on downstream transformers during a 40-hour run. That kind of data convinces. Spreadsheets alone don't.

The catch is that a pilot demands honest data — no cherry-picking the mildest year. Use the 90th percentile duration event from the last decade. If your pilot makes money in that scenario, you have a case for scale. If it loses, redesign the metric. Wrong order: lock in a new tariff before you know what it rewards. Right order: test, show the seam, then switch.

One last thing: duration-based costing will expose assets that were profitable only because peak-load accounting hid their true cost. That hurts. But a transformer that fails in year 8 instead of year 40 is not cheap — it's deferred debt. Stop costing for spikes. Start costing for stays.