Metering
Tokens, seconds, watts. Every call costed against three independent dimensions; reconciled into a triple-entry ledger.
MAP's billing model is three-dimensional. Every call is metered against tokens, seconds, and watts independently. MEAL records the metered events; MANA enforces tenant runway; MADE settles across organizations.
The three dimensions
Tokens
The unit of thought. LLM input + output tokens consumed by reasoning protocols (MARC, MAVEN, MACE, MAGE, MARI, MAGI).
- Token rates vary by protocol — heavy reasoners like
MARCcharge differently than light verifiers likeMAVEN. - Input and output tokens are counted independently.
- The token meter excludes overhead tokens (system prompts, tooling) — only intent-bearing tokens count.
Seconds
The unit of presence. Wall-clock time held against a deadline.
- Charged for every call. A protocol call that takes 1.4s and uses no tokens is still billed for 1.4 seconds.
- Useful for durable workflows (
MOON), physical actions (MOTE), and any operation that holds real time. - Sub-second resolution. The minimum chargeable unit is 1 millisecond.
Watts
The unit of work in the physical sense. Energy draw on bound compute.
- Charged when the operation runs GPU-bound work — heavy inference, training passes, large matrix ops.
- Measured at the substrate via vendor APIs (NVIDIA NVML on bare-metal; cloud GPU telemetry otherwise).
- Sovereign deployments can substitute their own measurement source.
The rate card
Rate cards are published per protocol and per tier. The current default rates (pricing):
| Dimension | Rate |
|---|---|
| Tokens | $3.00 / M tokens |
| Seconds | $0.018 / second |
| Watts | $0.42 / kWh |
| Cross-org settlement | 25 bps on MADE-settled value (15 bps at Institution, 5 bps at Sovereign) |
Heavy protocols carry multipliers on these base rates:
| Protocol | Token multiplier | Second multiplier | Watt multiplier |
|---|---|---|---|
MARC::reasoning_task | 1.0× | 1.0× | 1.2× |
MACE::deliberate | 1.5× (council members all consume) | 1.5× | 1.2× |
MAGI::trace | 1.0× | 2.0× | 4.0× (interpretability is GPU-heavy) |
MAVEN::attest | 0.8× | 1.0× | 1.0× |
| Most protocols | 1.0× | 1.0× | 1.0× |
Reconciliation
// MEAL operations
"meter" // record a metered event (tokens + seconds + watts) against caller account
"statement" // generate a statement for an account over a period
"reconcile" // reconcile metered events against settled payments; reports unbalancesThe triple-entry book is:
- Tenant ledger — every metered event debits the tenant's runway
- Settlement ledger — every successful
MADE::economic_contract_settlecredits the merchant - Audit chain — every meter event records to
MAX
MEAL::reconcile runs nightly (or on-demand) and asserts that:
sum(tenant_debits) = sum(merchant_credits) + sum(audit_records)Any drift produces an Unbalanced event recorded to MAX and routed to ops.
Runway enforcement
MANA::budget is consulted at the meter step (Stage 5 of the pipeline). If the tenant's runway is below threshold, the request is refused at MANA rather than MEAL:
let budget = mana::budget(tenant).await?;
if budget.remaining_in_dimension(Dim::Tokens) < estimated_tokens {
return Err(CoreError::ProtocolError(ProtocolError::PolicyDenied {
reason: "tenant runway exhausted (tokens)".into()
}));
}MANA can also throttle rather than refuse — reducing the rate-limit bucket for non-essential operations while still allowing critical ones (governance, audit reads).
Emergency holds
Under tenant-charter authority, MANA::hold can declare an emergency hold that stops all non-essential operations. The hold is multi-sig revocable (requires MACE quorum). All in-flight operations complete; new operations refuse.
This is the institution's emergency brake. It is rare but the institution always retains the option.
Pricing tier behavior
| Tier | Runway model | Settlement clearing fee |
|---|---|---|
| Sandbox (free) | 10k req/mo cap; no metering beyond that | n/a (no MADE access) |
| Charter ($499/mo) | 1M req included; metered per dimension above | 25 bps |
| Institution ($4,999/mo) | 10M req included; priority queue | 15 bps |
| Sovereign (custom) | Dedicated capacity; custom rates | 5 bps |
The charter fee covers the institutional surface (governance, audit chain, treaty maintenance, policy currency). Metered usage covers throughput. The clearing fee covers the cost of clearing cross-organization value. Three components, transparent.
See also
MEALprotocol — the meterMANAprotocol — the treasuryMADEprotocol — settlement- Pricing & rate card