Verified-correctness forecasting

Forecasts you can audit.
Decisions you can defend.

Phronesis mints probabilistic forecasts across eight verticals. Every forecast is primary-source-cited, time-stamped, persisted to a canonical ledger, and rendered publicly on a calibration scorecard from the moment it issues. API-forward by design. Agents are first-class customers.

5 of 8 Verticals live
$0.0084–0.0151 Inference cost / forecast
24/7 Production uptime
MCP + REST Co-equal at every endpoint
What Phronesis is

A forecasting platform for capital allocators in domains where rules cannot be mechanically applied.

Aristotle distinguished phronesis — practical wisdom — from episteme (systematic knowledge) and techne (craft). Phronesis is the wisdom that guides decision-makers in complex, variable, uncertain domains. The platform names that function precisely: a decision-support intelligence engine for energy infrastructure, AI compute, healthcare, climate physical risk, regulatory environments, supply chains, space economy, and robotics.

Forecasts are probabilistic, not deterministic. Every output carries an issue date, an evaluation window, and canonical-quantile estimates (p10, p50, p90). The substrate maintains an audit trail every customer can inspect.

Architecture

Verified correctness as a substrate property.

Three commitments enforced at the code layer, not asserted in marketing copy.

01

Primary-source citation always

Every forecast traces to a primary source: FERC filings, EDGAR, USPTO, state PUC dockets, DOE grant awards, peer-reviewed research, satellite telemetry, IFR World Robotics annual disclosures. No second-hand citations of secondary citations.

02

Timestamp discipline

Every forecast carries an issue date and an evaluation window. The forecast ledger is the canonical record of what was known when. Calibration accrues publicly over time. Past performance is observable; no cherry-picking.

03

API and MCP co-equal

Every endpoint that exists for human-driven REST clients exists at parity for autonomous-agent MCP clients. Idempotency keys are required, not optional. Errors are structured with stable machine-recoverable codes.

04

Sub-cent meter event precision

Per-call pricing reports inference cost-attestation with Decimal precision. Stripe meter events fire at canonical persistence to the Mnemosyne ledger. No floating-point drift in the billing path.

05

AP2 mandate at the door

Autonomous agents authenticate via Google Agent Payments Protocol mandates verified at a blocking middleware gate. Mandate signature verification budget is approximately two to five seconds. Agents act within mandate scope; the substrate enforces the boundary.

06

Per-tenant isolation at the boundary

Cross-tenant data access is architecturally prevented at the boundary publish layer. Customer queries never see other customers' data. Sacred Invariant #8 enforced through the substrate, not through policy.

Vertical axis

Eight verticals. One substrate.

Roughly seventy percent of the architecture is vertical-agnostic. The per-vertical layer specializes the prompt library and the data plane.

  • V#1Energy TransitionLive
  • V#2Computing + AI InfrastructureLive
  • V#3Healthcare (HIPAA Safe Harbor)Live
  • V#4Climate Physical RiskPre-launch
  • V#5RegulatoryPre-launch
  • V#6Supply-ChainPre-launch
  • V#7Space + Space EconomyLive
  • V#8RoboticsLive
Read the vertical detail
Engage

Two paths in.

If you are an agent acting on behalf of a capital allocator, start at the contract surface. If you are a human, the calibration scorecard is the most useful place to begin.