Foundations
Compute accountability is the practice of making digital compute measurable, attributable and legible to governance — so that claims and estimates about computing have traceable inputs, methods, assumptions and evidence limits. A structured record alone does not verify its claims.
Why it matters
Compute demand is growing faster than the ability to account for it. Energy is drawn in specific facilities on specific grids, but the resulting figures usually surface as aggregate corporate estimates: unattributable to a workload, uncheckable by a third party, and unusable as evidence. The problem is not that measurement is impossible. It is that no shared record structure exists for expressing what was measured, under what assumptions, on whose behalf.
Estimates need disclosed inputs, methods, assumptions and evidence limits. A structured record alone does not verify the workload or its claims.
Overheads, grid factors and allocation choices dominate results, yet are rarely disclosed alongside the number.
Institutions cannot compare, verify or act on figures that arrive in incompatible, unstructured forms.
Structure
A workload runs: training, inference, or general processing.
Compute activity is quantified in explicit units.
Location, grid and operating conditions are attached.
Impact is assigned to a responsible subject.
A structured Compute Accountability Record is produced.
Evidence becomes legible to institutional processes.
Layers
Build state is stated per layer. CSN has written specification work and a working demonstrator; it has no deployed measurement infrastructure.
Quantify the activity itself: how much compute ran, on what class of hardware, for how long, drawing how much power. Measurement is only meaningful when the unit and the origin of each figure are stated. A value with no declared unit and no declared source is not measurement — it is an assertion.
The same workload has a different consequence depending on where and when it runs. Context attaches location, grid conditions and operating environment to the measured activity. Without context, energy figures cannot become emissions figures without hiding an assumption.
Attribution assigns measured, contextualised impact to a responsible subject — an operator, a workload, a service. It requires a method that is written down, versioned, and inspectable step by step. Attribution without a visible method is not accountability.
A Compute Accountability Record is the artefact: a structured, machine-readable object carrying the inputs, the context, the method, the outputs, and the provenance of each. Records are the unit of exchange between technical systems and institutional processes.
Governance is what records are for. A record is governance-legible when a regulator, funder, auditor or operator can read it, understand what is claimed, see what is uncertain, and decide what to do. This is a documentation and schema problem before it is a technology problem.
