Read this before anything else
A big electricity bill is not really made of the electricity you used. A large part of it is set by how much you were pulling during a few hours of the year — the hours when the whole grid was at its busiest. Miss those hours and you pay for them all year.
So the useful thing is not a warning. It is a warning that turns out to be right. Anyone can send warnings every week and be right eventually; every wrong one costs you production. This page exists so that when we do have a hit rate, it will cover a stretch that started before we ever tried to sell anything.
There is no sample file to download, and there is not going to be one. Other feeds here hand you a slice of a file to look at before you buy it. There is no file here and nothing to buy: the record below is the product, and the whole of what we hold is printed on this page.
Not for sale yet. Nothing on this page has ever been sold, there is nothing to subscribe to, and there is no form here that takes anything from you. If you want to talk about it, that is an email thread with a person.
What a call isread from the recorder · the New York state grid
Once a day, for the day after, we look at the grid operator’s own published forecast of how busy the whole system will be, and we write down one of two things: cut your load, or ordinary day. That is the call. It is one line, it is about one day, and it is written before that day happens.
We only say cut your load when the forecast sits near the very top of what the grid has actually done recently, and “near the top” is a written rule rather than a judgement on the day: 90% of the last 60 recorded days. Saying it rarely is the whole point — a warning you get every fortnight is one you learn to ignore.
No call is made at all until the record behind that rule is long enough to justify one: the recorder wants at least 10 days of real grid load on file first, and says so instead of guessing.
The call is sealed the moment it is made, before anybody knows how the day went. Each one is stamped with a fingerprint worked out from the call itself, so if we ever went back and changed one, its fingerprint would stop matching. Those fingerprints are in the table below, under “Fingerprint”, and they are re-checked every time this page is built.
Every call we have made: 7sealed for 2026-08-25 to 2026-08-31 · 7 calls
All of them, in one table, including the ones that turned out wrong. Nothing is summarised away and nothing is left off.
| Day the call is about | Grid | What we said | Busiest we expected the grid to be, and when | The level that makes it a cut day | How it turned out | Right or wrong? | Fingerprint |
|---|---|---|---|---|---|---|---|
| 2026-08-31 | NYISO | ordinary day | 22,044 MWat hour 19 | 28,460 MW | not scored yet | not decided yet | 6c117a59a49f |
| 2026-08-30 | NYISO | ordinary day | 20,748 MWat hour 19 | 28,460 MW | not scored yet | not decided yet | bffa8efb2b0a |
| 2026-08-29 | NYISO | ordinary day | 19,871 MWat hour 19 | 28,460 MW | correct quiet day | right | 7305dd5dd080 |
| 2026-08-28 | NYISO | ordinary day | 21,515 MWat hour 19 | 28,460 MW | correct quiet day | right | 3879684caf97 |
| 2026-08-27 | NYISO | ordinary day | 22,259 MWat hour 18 | 28,460 MW | correct quiet day | right | ef51ba8421b1 |
| 2026-08-26 | NYISO | ordinary day | 21,561 MWat hour 18 | 28,460 MW | not scored yet | not decided yet | 22a144be8c9b |
| 2026-08-25 | NYISO | ordinary day | 20,120 MWat hour 19 | 28,460 MW | correct quiet day | right | 4857ca0dd98b |
There are three answers under “Right or wrong?”, not two. 4 right, 0 wrong, and 3 not decided yet — a call whose day has not been measured yet is neither a win nor a loss, and folding it into either is how a thin record starts looking good.
We have not told anyone to cut anything yet. Every call so far has been an ordinary day: what we expected never got near the column headed “The level that makes it a cut day”. That is the rule working, not the rule sleeping — the two figures in each row let you see the gap for yourself.
How a call is scored
After the day is over, the grid operator publishes what the system actually did. We compare that with what we said, and the call gets one of these words. They come out of the scoring code itself, so this list cannot drift away from the rule it describes.
- correct quiet day — rightWe said it would be an ordinary day, and it was.
- false alarm — wrongWe said cut, and the day turned out ordinary. This is the expensive one: it costs real production and buys nothing.
- hit — rightWe said cut, and the day really was near the top of the record.
- miss — wrongWe said it would be an ordinary day, and it was near the top. You were not warned.
The wrong ones stay on the page. A false alarm is the expensive mistake — you cut production for nothing — so it is counted out loud rather than quietly dropped out of the total.
The number we are not printing yet
We do not have a hit rate, so there is no percentage on this page. No call has told anyone to cut yet, so there is nothing to be right or wrong about.
The bar was written down before the record was anywhere near it. The recorder will not treat this record as worth selling until at least 60 calls have been scored and at least 5 of them told somebody to cut. A percentage off a handful of calls flatters whoever prints it, which is why the bar was set while the answer was still no.
When there is a percentage, it will be on this line, and it will have the wrong calls inside it.
What the calls are measured against273 days of recorded grid load
The level in the table above is not our opinion. It is worked out from what the grid has really done, day after day, which we record and keep. Here is how much of that we hold.
| Grid | Days of load recorded | Oldest day | Newest day |
|---|---|---|---|
| NYISO | 273 | 2025-05-01 | 2026-08-29 |
That stretch has holes in it and we would rather say so. Between 2025-05-01 and 2026-08-29 there are 213 days with no recorded load at all, and the longest unbroken run we hold is 153 days. Two dates at either end of a record make it look continuous; counting the gaps is the only way to see whether it is.
We also keep the operator’s own forward forecasts: 36 of them on file. Those are what a call is made from.
The fingerprints, re-checked as this page was built7 of 7 calls checked
Every one of the 7 calls above still matches the fingerprint taken when it was made, before its day happened. That was worked out again just now, from the stored call, using the recorder’s own code — not read out of a note somebody kept.
If one of them stopped matching, this page would not be built at all. There is no version of this record that survives being edited after the fact, so a mismatch stops the page rather than appearing on it as a footnote.
The daily copy the recorder keeps for itself7 days written
Separately from this page, the recorder writes its own plain copy of the scoreboard each day it runs and stores what it said, so what the record claimed on a given day can be checked later. It holds 7 of those, dated 2026-08-24 to 2026-08-30.
Its own record says none of those files has been put anywhere a stranger can read it. That flag is about those stored files, not about this page. Written to disk and published are two different things, and the recorder refuses to count the first as the second.
Where the numbers come from
One place: the grid operator’s own public file service. We fetch the files they publish, we do not scrape pages, and we check before each source is read whether they allow it.
- mis.nyiso.com — allowedNYISO public market information system: load forecast and actual load csv. Checked 2026-08-24: robots: absent(404). We fetch at one request every 5.0s to this host.
A source we have not cleared is not on this page. The addresses above are read out of the recorder’s own code every time this page is built, and each one has to carry a written permission note saying yes. If one did not, the page would refuse to build rather than name it.
What this page is not
- It is not advice about your own equipmentIt says the grid is likely to be near its busiest. What you can safely switch off, and for how long, is yours to decide.
- It is not a promiseSome calls will be wrong. They stay on the page for exactly that reason.
- It is not for sale, and there is no sample fileNothing here has ever been sold. There is no file behind this page to send you, which is why the line at the top says the sample is not ready.
- There is nowhere here to send us anythingNo form, no account, no service that takes a file from you. If you want to talk about the record, it is an email thread with a person.
Tell us we have got one wrong
No pay button on this one yet. Email operations@ustechautomations.com. There is nothing to buy here and nothing to sign up to. The record above is the whole of it and it is free to read.
Email us about the recordIf a call above does not match what your own meter saw that day, say which day and we will put your reading next to ours.