The numbers carried in the model that I had never verified myself. All five survive. Three need an amendment, and two of those change what the model claims rather than just its precision.
Model said 80–85% of pre-crisis.
Confirmed and now precise. Destatis puts production in energy-intensive branches 17.8% below 2021 in 2025, which is 82.2% — inside the model's range. Chemicals fell 18.1% from February 2022 to March 2026, the index going from 99.4 to 78.7.
Amend the duration. The model calls this a roughly two-year episode. It is now four years and still depressed, which makes it structural rather than a shock being absorbed. That strengthens R1 rather than qualifying it.
Model said roughly 10× on data-centre demand.
Confirmed almost exactly. The 2024/25 delivery year cleared at $28.92/MW-day against $269.92 for 2025/26, a factor of 9.3. Constrained zones went further: BGE at $466.35 and Dominion at $444.26.
New and useful. The attribution is now quantified. Data centres drove 63% of the 2025/26 increase, about $9.3bn, and accounted for 40% of the $16.4bn cost in the December auction for 2027/28. The crowding-out arrow in v4 now has a price tag.
Model said 23%, with a 2022–25 moratorium.
The share is confirmed and rising: 7,663 GWh in 2025, 23% of metered consumption, up from about 21% in 2024.
Two corrections. The moratorium began with a CRU direction in November 2021, not 2022. More importantly it was lifted in December 2025, so the model's past tense is now right for the wrong reason.
The replacement condition matters more than the lift: new connections must install on-site generation or storage covering their full demand. That converts data centres from firm-demand free riders into loads that bring their own capacity, which is a different node in the diagram.
Model said dozens to a few hundred per site.
The quotation is confirmed, and the detail is sharper: colocation runs 20–30 permanent staff per 100 MW, and a large site is roughly 1,500 construction jobs against 100 operational.
But the framing needs softening. Brookings also finds counties receiving their first large data centre see total private employment rise 4–5% over five to six years. The effect is small and easily overstated — naive estimates inflate it threefold — but it is not zero, and the model currently reads as if it were.
Model said 16 → 2,580 GWh, about 8% of district heat.
Confirmed at the primary source rather than via the Substack. Energiateollisuus reports electric boilers doubling to 8% of produced district heat in 2025, on district heating consumption of just over 31 TWh. Eight per cent of 31 TWh is about 2.5 TWh, which reproduces the 2,580 GWh independently.
One extra point for the series. E-boilers produced 750 GWh in 2023, so the path is 16 → 750 → 2,580 GWh. The curve is convex, which is what a policy step looks like when capital takes two years to arrive.
Germany's decline is persistent rather than episodic, so R1 should be drawn as a completed transition rather than an ongoing shock. Ireland's self-firming condition is a genuinely new instrument and belongs in v5 as a modifier on the crowding-out arrow. And the data-centre jobs claim should carry the 4–5% county effect, because omitting it is the kind of overstatement that gets a piece attacked on its weakest sentence.
PJM and Finland come through clean and both are now anchored to primary or near-primary sources. The German range and the Brookings quotation were accurate as written; only their surrounding framing needed work.
Germany: Destatis and Clean Energy Wire. PJM: IEEFA and Utility Dive. Ireland: Irish Times and Energy Connects. Jobs: Brookings. Finland: Energiateollisuus.