What people ask before they subscribe

NukeWorker has been the nuclear industry’s outage schedule for more than twenty years. It is built for the people who plan and staff outages, and it is where the underlying data comes from. NukeGryd is that same data rebuilt for energy markets: generation availability by ISO, formatted for a trading desk rather than a maintenance planner. It exists because traders kept arriving at NukeWorker needing something NukeWorker was never designed to be. Same data, same people, two very different audiences.

Every operating reactor in the United States, attributed to its balancing authority: the seven organized markets (PJM, MISO, ERCOT, CAISO, ISO-NE, NYISO, SPP) plus the vertically integrated utilities in the Southeast and West. Ontario (IESO) is covered as a separate fleet: Bruce, Darlington and Pickering, reported per unit. Ontario is always shown alongside the U.S. numbers and is never added into them, because the two come from different sources measuring different things.

The U.S. fleet updates once each morning, from the NRC Daily Power Reactor Status Report. That is the regulator’s publishing cadence, not ours: the NRC issues one report a day. Scrams and trips arrive the same day through a separate feed, ahead of the next morning’s status report. Ontario updates hourly, because IESO republishes its market data roughly every hour. Neither is a live tick, and we would rather say so than let you find out.

Three public sources, and we name them on every page. The NRC Daily Power Reactor Status Report for U.S. per-unit power level. EIA for net summer capacity and balancing-authority assignment. IESO’s Generator Output and Capability Report for Ontario. On top of that sits NukeWorker’s own outage schedule and projection engine, which is the part nobody else has. No confidential utility data, nothing under redistribution restriction.

Daily unit-level records for the U.S. fleet back to 1 January 1999. Ontario runs from May 2019, about 7 years of per-unit hourly output. How much of that you can see is set by your plan: 90 days free, then 6, 12 and 18 months as you move up, so the window back always mirrors the window forward.

No. It is a Monte Carlo simulation. Published outages are held fixed in every trial, so the band is tight in the near term where the schedule is known. Units with no published date have their start and duration sampled per trial from machine-learning quantile models, so the band widens further out, where the model is doing the work. Percentiles are taken across simulations rather than by summing per-unit quantiles, because summing quantiles assumes every outage moves together, which they do not. Output is P10/P50/P90 out to six months on Individual, twelve on Pro and eighteen on Team / Desk.

Because a shorter horizon hides units. Every reactor has a predicted next outage, but they do not fall evenly: at a twelve-month horizon only about 59% of units have had their next outage yet, so a one-year curve leaves roughly forty units’ next outages out of view. Eighteen months covers materially more of the fleet. We stop there rather than going further because published utility schedules run out well before it, and past that point the curve would be almost entirely model output with no filed date to check it against.

Correct, and we are explicit about it: Ontario outages are inferred from the output signal, not declared by a regulator. A unit is treated as offline when its output falls below 10% of rated capacity for three consecutive hours. Because the underlying data is hourly, the onset carries a real timestamp: we can say a unit came down at 21:00 on a given evening rather than just naming the day. Every inferred row is flagged as inferred, and a reason is only ever shown when it comes from a sourced announcement.

Because a rating is a reference figure, not a ceiling. Published net capacity is stated at design conditions, and real output rises when condenser cooling water is colder, which lowers turbine back pressure and yields more megawatts from the same reactor. A percent or two above the reference value is ordinary. We track both numbers: the sourced registry capacity, and the observed full-power output measured from the feed. Availability arithmetic uses the observed figure, because what matters when a unit trips is what it was actually delivering.

Yes, on the Ontario page. It is a different model from the U.S. curve rather than the same one pointed north: CANDU reactors refuel on power, so they have no refueling shutdown cycle and the cycle signal the U.S. projection leans on does not exist here. The Ontario model holds announced refurbishments to their published return dates, excludes permanently retired units on a sourced basis, and samples everything else from the observed duration distribution back to 2019.

Being straight about how good it is: it backtests at 80.6% P10-P90 coverage against a nominal 80%, but individual windows have ranged 67-93%, so the band is better centered than it is uniformly sharp. Two refurbishment return dates have settled so far and both units came back about seven months early, so a unit returning sooner than the curve shows is the expected error rather than a surprise.

Partly. Most of the model runs on a wholesale on-peak hub feed carrying seven hubs, which gave PJM, ISO-NE, MISO and CAISO a price layer and left NYISO, ERCOT, SPP and Ontario without one. We now take NYISO prices directly from the ISO (hourly day-ahead LBMP for all eleven zones, so each plant is priced at the zone it injects into rather than against one proxy hub), and Ontario prices directly from IESO.

Being straight about what that does and does not buy you: NYISO now has a real price series and a measured sensitivity, and that sensitivity comes out close to zero, because New York nuclear is about 3.3 GW against a market that peaks near 32 GW, so a nuclear loss there does not visibly move the zonal price. Ontario has a price series for context but no sensitivity, because the demand series the model regresses against covers U.S. balancing authorities only. ERCOT now has one too, taken directly from ERCOT’s public settlement-point feed (day-ahead, hub average plus the zonal hubs). One honest caveat: that source publishes the current operating day only, with no archive, so ERCOT history begins when we started collecting rather than running years back like the others.

SPP still has no price layer, and TVA is not applicable: TVA is a vertically integrated federal utility selling at regulated rates, so there is no wholesale market price to attach. Its units appear in the outage data with no $/MWh column by nature, not by omission.

It is a historical association, conditional on a tight grid: how the on-peak hub price has moved when nuclear supply dropped on high-demand days, scaled to the current gas regime. In mild conditions it is close to zero. Treat it as a sensitivity, not a price forecast, and not investment advice.

The free tier needs no card, and it is there so you can satisfy yourself the history is real before you pay for the forecast. It gives you the national MW-offline curve, today’s snapshot, the largest outages by impact and a seven-day look ahead. It does not include the per-ISO breakdown, the probabilistic forward curve, Ontario, alerts or the API, because those are the parts you would actually position against. Beyond that, Individual is $165 a month, billed annually at $1,980, and access unlocks the moment you subscribe. There is no sales call in the way and no trial to negotiate: pay, and you are in. Plans are annual because the product is the daily refresh, not a one-off download. See pricing.

Yes, on the Team / Desk tier and above: a REST API returning JSON, full historical datasets, bulk export and a documented signal library as CSV. See the API documentation for the available datasets.

Ask. NukeGryd is actively being developed, and specific requests from people trading these markets carry more weight than our guesses about what a desk needs. Tell us what would make it useful for you: get in touch.