Cases

Every partner call

Each card is one real partner payload replayed against production: end-to-end latency with the engine's share, then every covered variable — Axiom against the PolicyEngine golden and live PolicyEngine, with the difference. Variables Axiom cannot compute yet are folded below each card with the exact reason.

Evidence

State of the record

The latest run in one verdict: exact means equal within one cent; expected diffs are understood and documented (e.g. PE projects unpublished FY-2027 COLA values that Axiom deliberately omits); awaiting-encoding variables have no executable Axiom package yet.

Recorder

The tape

Every published run leaves a mark: exact agreement rides the upper trace, disagreements track beneath. Hover a point for that run's counts.

pens:exact matchesdisagreements

Runs

The docket

Comparison

Between two runs

Every variable whose value or coverage changed between the selected runs.

vs Compare runs

Service

Right now

Instrumentation

Where computation time goes

Measured inside the runtime for every real calculation, accumulated across instances and deploys.

planenginenormalize

No calculations recorded yet — run one and refresh.

Instrumentation

Latency by route

Busiest routes across live instances (in-memory, resets on deploy). Distribution buckets: ≤25, ≤100, ≤500, ≤2000, >2000 ms.

Ledger

What a full migration needs

Demand comes from PolicyEngine's production usage inventory: every output variable partners requested, weighted by the request volume of the clients that use it. Supply is what Axiom maps and executes today. The gap, ranked by unlocked volume, is the encoding roadmap.

Partner demand data is not bundled in this deployment.