Claims and limitations
Every load-bearing claim, tiered: sourced, derived, measured, open. And the honest boundaries.
Following the discipline the framework prescribes for agents: the paper's own claims, settled against their evidence, never against enthusiasm.
The claim ledger
| Claim | Tier |
|---|---|
| Tokens track compute, energy, carbon to first order | sourced |
| Token bills cannot denominate business value | sourced |
| Measured proxies distort effort; outcomes must be the measure | sourced |
| Mint rule: Q = V · B; scoring, settlement, IR are one computation | derived |
| Meter separation: inference volume cannot mint quirqs | derived |
| Tenure: cost non-increasing under non-decreasing memory | derived |
| Self-report mints fiction; state-minting does not | measured (E1, mock) |
| One surface check: silent gaming to full detection | measured (E2, mock) |
| Cost decays to floor in persistent environments | measured (E3, mock; consistency only) |
| Dashboard arithmetic | illustrative |
| Real agents reproduce E1 to E3 (H1, H2; E4) | open |
| QER* predicts program outcomes (H3; E5, E6) | open |
| Budget inflation is governable (H4; E7) | open |
| Frontier applications: builder pricing, market microdata, quantum metering | open (exploratory) |
Tiers: sourced (established in cited literature), derived (follows from stated definitions), measured (exhibited by the harness; mock mode unless noted), open (conjectured, falsifier stated, bound to a scheduled experiment in validation).
Limitations
The budget is the model's entry point for human error: a mispriced outcome mints mispriced quirqs, and budget governance is institutional, not mechanical. Definitions of done are incomplete contracts; work whose acceptance is irreducibly judgmental enters the ledger only through the owner's verdict, which restores a human bottleneck exactly where the work is fuzziest. The energy bridge inherits the wide measured variance of per-token energy. The measured tier is mock mode: it validates machinery, not agents. And a unit of account reshapes the behavior of what it measures; the mitigations are designed and exhibited in miniature, and unproven at scale, which is why the validation program, not the paper, is the product.
Ready to run it? Jump ahead to the playbook: definitions of done, budgets, and the ledger schema.
With the claims tiered, the mechanics: how to calculate it.