supagridWhat coal plants tell the market when something breaks
Whenever a coal unit changes its offer to the market, its owner has to say why in a few words. We read all 288,869 of those notes from August 2024 to September 2026 and matched each one to the megawatts it moved.
In short.
The question
Are Australia's coal plants breaking down more often? From July to September 2026 they sent the market 71% more notes naming a fault than a year before.
What we found
No. Almost all of the rise was automatic notes from one unit, Kogan Creek. Without them the count rose 4%, and the output lost to faults fell 18%.
Why it matters
A rebid note is the earliest public word of a fault, usually within minutes. The outlook owners publish for the days ahead can lag by a week. Read the notes by the megawatts they move, not by how many there are.
Sixteen days at Vales Point 5.
Vales Point 5 is a 660 MW unit on Lake Macquarie in New South Wales. In the middle of August 2025 its notes read like a shift log. Mill and feeder limits trim it by tens of megawatts most days, a fabric filter limit takes a little more, and an auxiliary plant failure comes and goes.
At 18:22 on Sunday 17 August it reports a water leak. Eleven minutes later the note says tube leak, and by 04:00 on Monday the unit is offline. It comes back on the 24th after a slow climb, a trip on the way up and an evening of coal limits. Two days later its turbine starts to vibrate, and by that night it is out again.
AEMO's dispatch figures show the megawatts falling to zero. Owners declare each unit's availability for the days ahead in MT PASA. Vales Point 5's entry for that week was last updated before the leak, and it still showed the unit fully available. The forced outage first appears in the following week's outlook. For most of that week, the notes were the only public record of why.
Tue 12 August, 07:04 Brisbane, note 1 of 117
Mill/Feeder limit SL
- On offer
- Generating
- A note naming a fault
- Any other note
- Forced outage in MT PASA
Half hourly averages of AEMO's five minute dispatch figures, and every rebid that moved MW or named a cause, in Brisbane time. The bottom strip is the state MT PASA carried for each day, as the owner declared it ahead. The figures and notes are AEMO's, as published. Which part a note names is our reading of it.
A count that rose by 71%.
Notes like these are where this post began. From July to September 2026 coal units lodged 13,762 rebids whose words name a plant fault, 71% more than in the same three months of 2025. Read that way, the fleet looks to be breaking down faster.
It isn't. 5,387 of those notes came from one unit. Between 17 June and 17 July 2026, Kogan Creek lodged a note reading Vacuum limit about every 3 minutes, day and night, 9,587 times in all, each signed AT. It had already lodged Ambient temperature the same way 78,737 times. We read these as notes written by software, not by a person. That is an inference from their rhythm. Without them, the quarter's count rose 4%.
Meanwhile what the faults cost fell. On an average day from July to September 2025, coal units were 2.8 GW short of their full output on days a fault was the cause. A year later it was 2.3 GW, 18% less. The capacity owners declared ahead as forced or maintenance outage fell further, from 1.4 GW to 0.8 GW, 43% less.
Over the whole year to September 2026 the picture is the same, if milder. Operators wrote 11% fewer fault notes than in the year before, the output short on fault days was 3% lower and the capacity declared unplanned ahead 15% lower.
Counting notes measures how much a plant writes. The megawatts measure how much it breaks.
Rebids whose words name a plant fault, each month
July to September 8,048 in 2025, 13,762 in 2026
Output short of full on days a fault was named, GW on an average day
July to September 2.8 GW in 2025, 2.3 GW in 2026
Capacity owners declared as forced or maintenance outage ahead, GW on an average day
July to September 1.4 GW in 2025, 0.8 GW in 2026
- Written by the operators
- Kogan Creek's automatic notes
- July to September, a year apart
All 44 coal units, August 2024 to September 2026. Rebid words are published by AEMO, and which of them name a fault is our reading. Output short of full comes from AEMO's dispatch availability against each unit's own full output. Declared capacity is from MT PASA, as published.
Every unit, every day.
Here is the whole fleet for two years, a row for each of the 44 units and a column for each day. Across all of it, a unit spent 51% of its days running near full, 9% in planned outages and 38% short of full for a fault its notes named.
The long bright runs are units out for weeks. The scatter between them is the everyday wear of mills, fans and filters. Pick a part to see only its days, point at any day for what the owner said, and click it to read that day's notes in order.
- Running near full
- Short for a stated cause, 5% up to all of it
- Planned outage
- Off or short, no reason given
One row per unit, one column per trading day, 1 August 2024 to 30 September 2026. A day's cause is the part named by the notes that took the most MW off offer that day. A unit that goes offline after a fault keeps that cause until it returns, up to fourteen days. Short of full is against the unit's own full output, from AEMO's dispatch availability. The notes and figures are AEMO's, published. The causes are our reading of the notes.
What breaks.
Mills, and the coal they grind, are named most. A mill limit is the everyday fault of a coal plant. A mill or feeder is down for repair, or the coal is wet or burns poorly. The unit gives up a slice of its output, on average 21% on the days it is the cause, and usually gets it back the same day.
Boilers, turbines and trips come less often and take far more. On a day a boiler fault is the cause, a unit is 46% short on average, 63% for its turbine or generator and 72% after a trip. One boiler run in ten lasts 8 days or more, long enough to cool a unit, find a leaking tube and bring it back.
Unit days on which a note named each part
With the operators' commonest words for it, the station that used them most and how many times
Mills and coal
On its days a unit was 21% short on average. One run in ten lasted 5 days or more.
- Fuel/Mill/CV LimitationMillmerran, 8,960
- Mill -Mill Limit-SLCallide B, 2,048
- 010 Unexpected/plant limits~101 Milling LimitsBayswater, 1,055
- 010 Unexpected/plant limits~103 Coal Quality IssuesLoy Yang A, 831
Air and ash
On its days a unit was 25% short on average. One run in ten lasted 4 days or more.
- Fan(s) LimitationMillmerran, 1,393
- Change in avail - Fabric filter limitation revised SLEraring, 853
- Emissions -Fabric filter-SLGladstone, 520
- Furnace -ID Fan-SLGladstone, 429
Boiler
On its days a unit was 46% short on average. One run in ten lasted 8 days or more.
- Clinker deload - SLCallide C, 390
- Furnace -Flame Stability-SLCallide B, 359
- Boiler -Other-SLGladstone, 242
- Clinker management - SLCallide C, 151
Water and steam
On its days a unit was 30% short on average. One run in ten lasted 5 days or more.
- Condensate Polisher Inlet Temperature LimitationMillmerran, 1,872
- Boiler -Feed Pump-SLGladstone, 1,129
- Steam feed pump - SLCallide C, 57
- Furnace -Spray Valve-SLCallide B, 51
Condenser and cooling
On its days a unit was 33% short on average. One run in ten lasted 4 days or more.
- Vacuum Limit - ATKogan Creek, 9,587
- Condenser -Backflush-SLGladstone, 2,170
- Backpressure LimitationMillmerran, 308
- Cooling Water System-Delta Temperature-SLGladstone, 269
Plant, no part named
On its days a unit was 48% short on average. One run in ten lasted 6 days or more.
- Availability ChangeBayswater, 3,941
- 030 Increase in avail cap~301 plant limit liftedLoy Yang A, 909
- revised unit availability. SLLoy Yang B, 183
- 030 Increase in avail cap~plant limit liftedLoy Yang A, 112
Controls and auxiliaries
On its days a unit was 53% short on average. One run in ten lasted 6 days or more.
- Control System - Stabilise Unit Output -SLKogan Creek, 1,282
- Control System - Match Unit Output -SLGladstone, 772
- 010 Unexpected/plant limits~106 Aux/Plant failureBayswater, 704
- Aux plant failureVales Point, 517
Turbine and generator
On its days a unit was 63% short on average. One run in ten lasted 6 days or more.
- Turbine LimitationMillmerran, 508
- Turbine -Vibration-SLGladstone, 479
- Testing-Turbine -SLCallide B, 35
- Turbine Valve Stroke Testing - Fixed Load Condition RequiredMillmerran, 24
Trip
On its days a unit was 72% short on average. One run in ten lasted 5 days or more.
- Unit trip - rebid energy contFCAS and regFCAS to zero - ATGladstone, 86
- Unit commitment-Unit Trip-SLCallide C, 59
- Unit trip - SLVales Point, 39
- Unit TripMillmerran, 36
All 44 coal units, August 2024 to September 2026. A unit day counts once for a part however many notes named it that day. The words are published by AEMO exactly as lodged. Which part each names is our reading.
Station by station.
Each station has its own complaint. Millmerran's notes are mostly about mills and coal quality, and Eraring's and Vales Point's about fabric filters and ash. Gladstone's are about condenser backflushes, Loy Yang A's about controls and auxiliary plant, and Callide C's about its boilers. Loy Yang B had more of its full output on offer than any other station and barely writes at all.
The dots are what the notes moved, so they show what a station talks about as much as how it runs. For how it runs, read the columns on the left.
Scroll the table sideways for what each station's notes name.
| Station | Closes | On offer | Declared unplanned | Mills and coal | Air and ash | Boiler | Water and steam | Condenser and cooling | Turbine and generator | Controls and auxiliaries | Trip | Plant, no part named |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Loy Yang BVic, 1,170 MW | 2047 | 96% | 0.0% | |||||||||
| Mt PiperNSW, 1,430 MW | 2040 | 85% | 3.3% | |||||||||
| StanwellQld, 1,460 MW | 2043 | 84% | 0.2% | |||||||||
| Loy Yang AVic, 2,210 MW | 2035 | 84% | 2.3% | |||||||||
| BayswaterNSW, 2,715 MW | 2033 | 82% | 4.9% | |||||||||
| TarongQld, 1,400 MW | 2036 | 80% | 1.0% | |||||||||
| MillmerranQld, 870 MW | 2051 | 80% | 5.9% | |||||||||
| Callide BQld, 700 MW | 2031 | 77% | 2.3% | |||||||||
| Vales PointNSW, 1,320 MW | 2033 | 75% | 7.0% | |||||||||
| Kogan CreekQld, 772 MW | 2042 | 74% | 4.6% | |||||||||
| EraringNSW, 2,800 MW | 2029 | 73% | 11.4% | |||||||||
| YallournVic, 1,501 MW | 2028 | 70% | 11.0% | |||||||||
| Callide CQld, 844 MW | Not given | 69% | 9.6% | |||||||||
| Tarong NorthQld, 443 MW | 2037 | 66% | 5.8% | |||||||||
| GladstoneQld, 1,680 MW | 2029 | 61% | 9.3% |
Notes naming it took 296.9 GWh off offer for each GW of the station, each year.
- Fuel/Mill/CV Limitation8,960
- Mill or Feeder Limitation2,103
- Mill/CV limitations. SL233
Stations from most to least of their full output on offer, on average over the two years, from AEMO's dispatch availability. Declared unplanned is the share of the station's capacity MT PASA carried as forced or maintenance outage, as declared ahead. Closing years are those owners gave AEMO for its generator information, August 2026. The notes are published by AEMO. Which part each names is our reading, and how much a station writes shapes its dots as much as how it runs.
What it means.
The notes come first.
For faults that took at least 50 MW off offer and gave the time they happened, half the rebids reached AEMO within 2 minutes and nine in ten within 6. MT PASA can lag by a week. The notes are the earliest public word of a fault and its cause.
Weigh words by megawatts.
A note is a sentence, not a measure. One unit's automatic notes lifted the fleet's count by more than two thirds. Match every note to the megawatts it moved before reading anything into how often a plant writes.
Read a plant against itself.
Owners write in their own codes and habits, and some sign every note. A station's record says most beside its own past, and least beside a neighbour that writes differently.
In supagrid.
- Open any coal unit and point at any five minutes. Its offer sits beside what it generated, with the time it was lodged and the reason its owner gave.
- Ask the Analyst what a unit was offering on a day and why. It reads every rebid that was in force and quotes the owner's own words.
- Move the clock back to any day in this post and every unit is there as it was that day.
Method and data.
The data
Everything here comes from AEMO's published records for the trading days from 1 August 2024 to 30 September 2026, covering 44 coal units at 15 stations. That is every offer and rebid with the reason its owner gave, what each unit had available and generated, and what owners declared for the weeks ahead. The files behind each figure are frozen with this post.
How we read it
We sorted every note by what its words say, such as a fault and the part it names, hot weather, an outage or the market. Then we matched each one to the megawatts it moved. The rules are our own and approximate. Where we read something from a pattern rather than the words, we say so.
Where it falls short
A note is what an owner chose to write, so it shows what a station talks about as much as how it runs. The outlook owners publish for the weeks ahead can lag by days, so a short fault may never show as declared. Nothing here judges whether a plant is safe or forecasts how it will run.
Times
Times are market time, which is Brisbane time all year round. A trading day runs from 4am to 4am, so a rebid lodged at 1am belongs to the day before.
More from Research.
Our dispatch model, within a dollar of the real market
We tested supagrid's dispatch model on nine days it had never seen. 98.2% of its five minute prices landed within $1 of AEMO's, with an average gap of 8c, so every scenario in supagrid starts from the market as it really was.
The machines that bid our batteries
Batteries now lodge nearly three in four of the NEM's rebids, and two pieces of software write most of them. We sorted 9.7 million rebids into families, then scored a year of 45 batteries to ask whether the choice of software shows up in what they earn.
The two lines that took New South Wales to the edge of the price cap
On 12 October two lines in New South Wales were out for maintenance, and the price reached $23,107 while Victoria's sat below zero. We show how the limits set the price, run the afternoon again with either line back, and work out how likely it was.
See it in supagrid.
Book a demo and we'll show you the evening, the line or the change on your own corner of the NEM.