The two lines that took New South Wales to the edge of the price cap

We tested which outage mattered on supagrid's dispatch model. It solves each five minutes the way AEMO's dispatch does, from the same published inputs, so it can run the afternoon again and change one thing at a time. We use three names.

The Backcast clears $23,107 at 14:20 and 14:30, as AEMO did, and 9 of the actual's 11 intervals above $5,000 from 14:20 to 15:35. It is one afternoon checked against itself, so it shows the Backcast is close enough to compare scenarios with, not how the model does on days it has never seen. Every figure from the model carries the flask .

To compare runs we use spot cost, New South Wales demand times the spot price over the sixteen intervals from 14:20 to 15:35, before any contracts. In the Backcast it is $82.5 million. Pick a scenario to put a line back in service.

With line 6 back there is no Canberra limit, and the afternoon is ordinary. The price stays between $32 and $300, and spot cost falls to $1.5 million. The Victoria link, which carried about 250 MW south in the Backcast, brings about 825 MW north instead. This run keeps the offers units behind the limit had at 13:45, before they raced to the floor, so it is the cautious case.

With line 86 back the Queensland link carries about 906 MW south where it carried 485. The Canberra limit is still full, but the extra power from the north is enough for every interval but one. At 14:30 the Armidale to Tamworth lines reach a limit of their own, at 1,178 MW with 874 MW from Queensland, and that interval clears $9,998.66. Spot cost is $8.5 million. With both lines back the price stays under $300 and spot cost is $1.1 million.

So neither outage would have done it alone. Line 86 back takes about $74 million off spot cost, and line 6 back about $81 million. Line 86's booking said it could be returned within 120 minutes if AEMO asked. A recall at 11:30 would have had it back at 13:30, while the price was still under $50, and from 14:00 the afternoon runs as if it had never been out, with a spot cost of $8.5 million. We found no market notice that day about recalling it.

Both bookings were public months ahead. Transgrid first booked line 6 on 30 June, and AEMO marked it likely to proceed on 31 August and again on 14 September. Line 86 was booked out from 06:30 to 16:00 on each day from Monday 12 to Friday 16 October, lodged on 8 July. Once line 6 was out, the Canberra limit bound on twelve of its first fifteen days, and at 07:20 on Saturday 10 October it took the price to the $23,200 cap.

So the bookings marked five afternoons with both lines out. Only the 12th spiked. On the other four the highest price from 11:00 to 17:00 was between $110 and $442.

Three things came together on the 12th, and no other afternoon had all three. Both lines were out, as booked. The afternoon was heavy and cloudy, with demand averaging 8,388 MW from 14:20 to 15:35, and the solar farms able to make under a quarter of what they could over the six days before. And at 08:20 that morning AEMO held the Queensland link to 500 MW south, a cap line 86's booking hadn't named. The 13th was nearly as heavy and cloudy, but the link kept its usual limit and the price peaked at $442. The 14th had the cap but was light and sunny, and peaked at $160.

We also ran the model as a forecast, with the outages as booked and only the offers published by then. Six days before, it put the 12th near the cap and kept the other four ordinary. AEMO's own seven day outlook first did so four days before. Our forecast knew each afternoon's actual weather, so that is no contest. Then both flagged the 13th as well, and until the day itself neither told the two apart.

The offers did. Give the model the offers generators actually lodged for each day and the 12th reaches $22,889 while the 13th peaks at $459. Add the morning's cap and the 12th has six intervals above $5,000. Swap the two days' offers and the 13th spikes too, reaching $23,107 with the 12th's offers and the link held. With the 13th's offers neither afternoon gets past $3,210. The weather and the network made both afternoons exposed, and the offers chose which one paid. Nobody could have named the 12th in advance, AEMO included.

A desk deciding whether to hedge doesn't need a name, though. It needs the odds, and those can be estimated before anyone has made an offer. We split each afternoon in two. First the physical afternoon, which the bookings, the limits, the weather and the units decide. We measure it as margin, how much more New South Wales could be supplied at 14:20 if every generator offered all it had at $0. Under 3,000 MW we call an afternoon exposed. At 14:20 the 12th had 2,901 MW to spare and the 13th 2,007 MW. Then the offers, which stay unknown until the day is over.

History can't give the odds, since the two lines had only ever been out together on those five afternoons. So we made more afternoons. We kept everything about 12 October as it was, the network, its limits, which units were running and the offers, and gave it the weather of 121 other spring weekdays. Each takes the demand, wind and sun of one real day together, so a cloudy afternoon is a heavy one too, as it tends to be.

With the network as it was on the day, five of the 121 afternoons fell under the line, and with the 12th's offers every one of them spiked. No afternoon with room to spare spiked, under any network or either set of offers. Every spike came when New South Wales demand less its own wind and solar was 5,438 MW or more at 14:20, and every quiet afternoon was at 5,060 MW or less. About one spring weekday in 24 is like that.

So from 8 July, when both bookings were in, the chance of a spike at 14:20 on any one of the five afternoons was about 1 in 50, or 2.1%. Across all five, the chance that at least one would spike was about 1 in 10. With both lines in service the same afternoon would have been 0.4%.

The chance sharpens as each piece becomes known. About a week out the weather forecast reaches the day, and on the morning AEMO sets the day's limits. We ran the 12th and the 13th as forecasts at each step, moving their weather by the errors AEMO's own forecasts make at that distance, 25 times over.

Two to three days out the 12th was 27%, about 1 in 4, and the 13th 3%. On the morning the Queensland cap took the 12th to 40%, and the 13th, with its link at the usual limit, fell to nil. The 12th sat on a knife edge. Without the morning's cap its weather left it 3,067 MW to spare, just above the line, so a forecast error of a couple of percent could tip it.

Run each morning, this becomes a lens on the days ahead, with each afternoon read Low, Medium or High. From 8 July it would have marked all five afternoons Medium. It first reads the 12th High on the morning of Monday 5 October, seven days ahead. It would not have said the 12th would spike, and nothing honest could have. It would have said which afternoon carried the risk, and why. Read it as a first test on one week, run afterwards with forecasts centred on the weather that came.

The same two lines are booked out together again, line 6 from 06:00 on 28 October to 16:00 on 16 November, and line 86 from 06:30 on 3 November to 15:00 on 8 November. In the outage schedule as it stood on 18 October, it is the only time the two are booked out at once. This time line 86 is out day and night, so the evenings fall inside it too. Of the three things that came together on 12 October, one is booked.

When the week is over we will add here what happened on each of the six days.

Outages stack. Line 6 had been out for twelve days and line 86 for one. Either alone left an ordinary afternoon. Together, on a heavy, cloudy day with the Queensland link capped, they took the price to $23,107.

Where a megawatt sits against a limit decides what it is worth. At 14:20, through the Canberra limit, a megawatt near Parkes was worth $16,387 to dispatch, one at Bango $1,747 and one at Tumut minus $1,000, while all three were paid $23,107. For anyone weighing where to build, a share in an equation like this one works as a locational price the market already sets.

For anyone trading around an outage, the bookings were public about three months before the day. Nobody could have named the 12th, but its odds could be read each morning, and they rose days before it came.

• Open the map on 12 October. Both lines that were out are there, with the limits they brought in and each limit's equation, with every unit's share.
• Move the clock through the afternoon and see what each unit behind the limit offered and ran, five minutes at a time.
• In Modelling, the Backcast is the run labelled Model. Run it beside the actual prices, and its panel shows how closely it matched over the days you choose.
• Set an alert for when a line near your assets is booked out, or when a limit you watch binds.