What the five findings do when you put them together.
What the five findings do when you put them together.
Kenneth · 2026-08-12
Source: LBNL Queued Up 2026 Edition (CC BY 4.0), through end-2025. Analysis my own.
Six notes, each with one finding. Held apart they are interesting. Multiplied
together they change what a queue position is worth.
| Note | Finding |
|---|---|
| 001 | The 68% composite hides a 34–98% spread by technology |
| 002 | It rose from ~68% pre-2014 to 93% for 2018–21 |
| 003 | Not PJM — 16 of 19 operators worsened. ERCOT went the other way |
| 004 | Time to build roughly doubled: 1.5 years to 5.0 |
| 005 | Failure rises with size: 76% under 20 MW, 89% at 100–299 MW |
| 006 | Battery converts at 0.5% of requested MW. Offshore wind at 0.0% |
Take the base rates as they stand rather than the composite:
Every axis points the same way, and they are not independent — a large recent
battery is in the worst bucket of all four. A defensible planning assumption is
below one in ten, against a composite figure of roughly one in three.
The same project at 40 MW, in ERCOT, is a different business. Note 005 says
small does better; Note 003 says ERCOT converts at 61% failure against PJM's 93%.
Neither of those is a small adjustment.
It is not a model. These are base rates from history, multiplied by
inspection rather than by regression. A real underwriting model would need
interconnection costs, upgrade allocations and developer identity, none of which
this dataset carries.
What it is: a much better prior than 68%, and a demonstration that the number
everyone quotes is wrong in at least four independent directions at once.
*Method: every figure traceable to Notes 001–006, each a count from the published
LBNL file. Free at emp.lbl.gov/queues.*
*This is published research. Every figure is a count from a free dataset at emp.lbl.gov/queues — check it yourself rather than taking my word for it.