Meridian Deal Partners
The opacity of private markets
is a filing problem.
Four decades of disclosure law pushed an enormous quantity of economic detail into the public record — who owns what, who crossed which threshold on what date, which power station exists and in what state of construction. None of it is secret. All of it is fragmented, because each register was built by a different institution for its own purpose and none was designed to be read against another.
A filing problem is an engineering problem. Meridian assembles those registers into a single estate on one identity spine, and answers questions that no individual register can.
Measured from the live stores, 2 November 2026. The adjacent half is held for reference and is never mixed into an analytical claim — only the resolved half can be sold as an answer.
The finding
Announcement is not capacity
Of 372 developers announcing generation capacity against datacentre demand, eighteen have confirmed any of it in the federal generator register. The remaining 354 have announced 162,985 MW and confirmed none.
Split by technology, the asymmetry is not subtle.
| Developer type | Count | Announced MW | Confirmed MW | Confirmed |
|---|---|---|---|---|
| Thermal / nuclear, confirmed | 16 | 30,473 | 30,473 | 100% |
| Thermal / nuclear, none | 2 | 2,600 | 0 | 0% |
| Renewable / storage, confirmed | 2 | 2,597 | 1,574 | 61% |
| Renewable / storage, none | 352 | 160,385 | 0 | 0% |
What is underneath
Four registers, read against each other
Read alone, each answers a narrow question. Read together, they resolve a project vehicle to the sponsor behind it, that sponsor to its ultimate owner, and that owner to the institutions exposed to it — which is the work that otherwise takes an analyst a week per name.
Method
We tried to fail first
A research programme that reports only its successes cannot be checked. Before publishing anything, the pipeline was pointed at a result somebody else had already published: Brav, Jiang, Partnoy and Thomas reported roughly a seven per cent abnormal return around an initial Schedule 13D, and no subsequent reversal.
On independently assembled filings and a different period, we recovered +8.16% (n=266, t=2.63) and +0.21% over (+1,+60) — both the magnitude and the absence of reversal.
Then the honest part. Four candidate signals were measured and all four looked strong. Under characteristic-matched controls, three lost between 75 and 82 per cent of their effect. One survived, and it is a crash-risk signal rather than a return forecast.
Access
The data room
Papers, briefs and the research record — including the work that did not survive testing.