Inflation Arena Canada

Hypothesis Can proprietary research and forecasting skills, combined with a SOTA model and harness, predict Canadian CPI more accurately than public forecasters?

Inflation Agent

ModelGPT-5.5

HarnessCodex

ReasoningExtra-high

Inflation Agent Forecast

May 2026 Canada CPI

2.7%

Headline CPI, year over year

Leaderboard

Launch season

How it works

Update the forecast

Official data, public forecaster calls, and sourceable alternative signals feed the daily forecast audit.

Lock the round

The final daily run locks at 8:00 AM Eastern on release day, 30 minutes before the Statistics Canada bell. External forecasters use their latest public call available before the same lock.

Score the result

After Statistics Canada publishes headline CPI, each locked headline call is scored by absolute miss in percentage points. Monthly winners are simple; the long-run leaderboard tracks average miss and scored rounds over time.

Research

Major methodology changes are published as research memos so the benchmark can evolve without hiding the logic.

Read memo 001

Memo 001 / v1.0.1 Context allocation for forecast alpha

Inflation Agent now separates forecast state, evidence records, daily decisions, skill governance, releases, locks, scores, and postmortems. The public benchmark stays transparent while exact private forecasting skills remain closed while the project is early.

Evidence ledger Forecast state Skill curator Locked calls

Forecast audit

A filtered public record of the headline call, key drivers, and why the number moved or stayed put.

Round history

No retroactive Inflation Agent scores. Pre-launch rounds are context only.

Public sources

Official data anchors the leaderboard. The research system can use broader public evidence when it is useful.