Public production release · indexing enabled
Interpretation library
Four long-form guides isolate the category errors that most often overstate or understate AI-compute infrastructure: collapsing state transitions, crossing registers, assigning the whole chain to one actor and treating a public-record gap as an outcome.
Reading protocol
Start with the executive summary. The conclusion, boundary and risk are visible before the archive detail.
Use the diagram as a model. It organizes evidence classes; it does not create missing observations.
Inspect rival tests. Contrary evidence determines whether a stronger explanation survives.
End with maintenance. Reopen only when a specified evidence delta appears.
Long-form reports
Each guide is written for both decision readers and record owners, with a navigable argument, evidence file and bibliography.
C0–C20
Primary risk · State inflation
Why every transition requires a different kind of public evidence — and why the chain can stop, branch or reverse.
Four registers
Primary risk · False equivalence
How load, supply, money and AI-share measures become misleading when collapsed into one total.
C7–C10 fork
Primary risk · Single-actor attribution
How companies, governments, utilities, regulators, courts and suppliers control different links.
C18–C20
Primary risk · Invented downstream bridge
Why operating status does not by itself establish contracted use, measured utilization, revenue or sustained economic capacity.
Reading boundary
Explainers synthesize the locked archive and carry claim IDs, source IDs, ceilings, contrary evidence, negative findings and update triggers. They do not rank projects, estimate market totals or predict outcomes.