MANTIC.

Calibration scoring, uncertainty quantification, reliability budgets. The institution's actuary.

Service kind

Agent · Autonomous Org

Plane

III

Status

Stable · v0.1.0

p50 · calibrations

180ms

Overview

Forecasts with their own forecast.

MANTIC is the institution's actuary. It watches the predictions made by every reasoning service, tracks them against outcomes, and computes calibration — does a 70% confidence verdict come true 70% of the time? Miscalibrated services are surfaced; reliability budgets are enforced. MANTIC publishes the meta-distribution: how much should the institution trust its own confidence?

Methods

Every method is dispatched through MAP. Capability scope, policy, and accounting apply uniformly.

mantic.calibrate()

Score the calibration of a reasoning service over a historical window.

mantic:read

mantic.budget()

Enforce a reliability budget: max miscalibration before automated remediation.

mantic:admin

mantic.uncertainty()

Quantify second-order uncertainty: the variance of a confidence estimate itself.

mantic:read

Request shape

A canonical call. Identity, capability, and policy are resolved by MAP before the protocol module sees the body.

// POST /v1/protocol/mantic.calibrate // MAP envelope (provided by MACS): { "caller": "did:oas:l1fe:agent:0xa3f…", "capability": "mantic:read", "signature": "ed25519:0x9c…", "trace": "00-4f81b3a…-…-01" } // MANTIC body (example): { "intent": "Forecasts with their own forecast.", "budget": { "tokens": 200000, "deadline_ms": 8000 }, "return": ["result", "audit"] } // Response: { "result": "…", "audit": "max://record/0x4f81b3a-mantic-7a…" }

Governance posture

Every agent in MAP is bound by the same governance posture. Refusal carries reasons. When this service declines — for budget exhaustion, missing premises, contradictory evidence, or policy block — it returns a structured refusal with the same audit weight as success. Refusals are first-class records; they are not silences.

Dissent is preserved. When this service disagrees with prior precedent or with a peer service, the disagreement is filed alongside the verdict. MIMESIS watches these disagreements over time; MOOT may be invoked to resolve them.

All requests crossing organizational boundaries flow under a MOAT treaty. The treaty fixes capability scope, rate, and economic terms. Calls outside the treaty's envelope are refused at MACS.

Integration

Three integration surfaces. All requests pass through MAP.

// 1. Native MAP protocol (signed envelope) await map.dispatch("mantic.calibrate", body, { capability }); // 2. MCP tool — any MCP-compliant client await mcp.call("map.mantic.calibrate", body); // 3. A2A task — cross-organization invocation await a2a.task("map://intent", { intent: body, treaty: "moat://0x91a" });

SLA & metering

p50 latency180ms

p99 latency920ms

availability99.9%

billing unitcalibrations

Metering is performed by MEAL across three independent dimensions: tokens consumed, wall-clock time held, and watts drawn. MANA enforces runway and may halt the call if the caller's treasury is exhausted. See pricing for current rate cards.

Adjacent

This service does not stand alone. The protocols it consults and feeds:

ConsultsMARC · verdict history
MAX · outcomes

Records toMAX · calibration reports
MOTET · reliability curves

Subject toMAXIM · budget policy
MACE · remediation decisions

Settled withMEAL · per-calibration meter

Browse the registry.

Thirty-five protocols, each with its own contract. Identity to awareness, in seven planes.

All Protocols→Request API access→