Token ledger
What a token-ledger row contains
A usable token-ledger row records more than a token count. It carries the timestamp, the model and endpoint invoked, the input and output token counts, the priced cost, and, decisively, the project or resource identifier that attributes the request to a specific build. Without that identifier the row cannot support capitalisation, because it cannot be tied to a recognised asset. With it, the row becomes primary evidence of directly attributable development cost.
From ledger row to general-ledger entry
The token ledger and the general ledger are different books serving different readers. The token ledger is granular, high-volume and owned by engineering. The general ledger is periodic, summarised and owned by finance. Token capitalisation exists to bridge them: development-phase rows for a build are aggregated for the period and posted as an addition to the intangible asset, with the ledger retained as the supporting detail.
- 2026-03-04 gpt fine-tune job
- project = support-copilot
- input 4.2M tok, output 1.1M tok
- priced cost, development phase
- Dr Intangible asset (support-copilot)
- Cr Cost capitalised / clearing
- period addition to carrying amount
- ledger rows retained as support
Why the ledger is the audit base
An auditor challenging a capitalised amount asks what it is made of. The token ledger answers at the level of the individual request, which is what lets the reliable-measurement criterion be met at source rather than by estimate IAS 38 §54-62. Retention matters: if the ledger is not kept for the life of the asset and the audit period, the nexus of cost weakens and the amount becomes harder to defend.