← Dashboard

Performance

Live since June 2026

Scoreboard · model vs market

The metric that decides whether this project lives. Beating the BLS print is not the objective — the market already predicts it well. The objective is beating the market.

T−7 days · n=22

-0.027

log-loss edge · market ahead

model 1.041 · market 1.014

T−2 days · n=22

-0.002

log-loss edge · market ahead

model 1.039 · market 1.038

T−1 days · n=22

+0.035

log-loss edge · model ahead

model 1.016 · market 1.051

How to read this: both the model and the market put a probability on the bucket that actually printed; log-loss punishes overconfidence. Edge > 0 means the model's distribution was better than the market's. Kill criterion: if the T−7 edge is still negative after a trailing 9 live prints, the model has no edge over the market and the project should be killed or rethought. Retro months are scored point-in-time (only data visible on that day, curve fitted only on months already resolved by then). Updated 2026-09-23.

Per print — probability on the realized bucket (T−7)

MonthActualModelMarketBetter
November 20242.7%41%46%market
December 20242.9%48%69%market
January 20253.0%51%79%market
February 20252.8%34%28%model
March 20252.4%30%26%model
April 20252.3%44%38%model
May 20252.4%35%34%model
June 20252.7%74%73%model
July 20252.7%31%29%model
August 20252.9%41%42%market
September 20253.0%38%36%model
October 20253.0%61%67%market
November 20252.7%4%3%model
December 20252.7%93%91%model
January 20262.4%98%97%model
February 20262.4%41%42%market
March 20263.3%99%99%model
April 20263.8%20%29%market
May 20264.2%42%42%market
June 20263.5%2%2%model
July 20263.4%42%44%market
August 20263.4%39%40%market

Live record · since June 2026

Real forward forecasts, scored against the BLS print using the model's locked call from the day before release, next to what the market said at the same moment. May 2026 is the first scored cycle.

Forecasts scored

4

Beat market

0/4

higher prob on realized bucket

Bucket hit rate

3/4

realized = model's top bucket

Avg abs error

0.13pp

point estimate vs actual

May 2026 CPI

model called 4.24% · actual 4.2% · miss +0.04pp

bucket hitmarket was sharper
model 36.8% on 4.2%
market 99.5% on 4.2%

June 2026 CPI

model called 3.92% · actual 3.5% · miss +0.42pp

bucket missmarket was sharper
model 0.4% on ≤3.6%
market 1.9% on ≤3.6%

July 2026 CPI

model called 3.43% · actual 3.4% · miss +0.03pp

bucket hitmarket was sharper
model 38.3% on 3.4%
market 38.5% on 3.4%

August 2026 CPI

model called 3.38% · actual 3.4% · miss -0.02pp

bucket hitmarket was sharper
model 42.1% on 3.4%
market 43.0% on 3.4%

September 2026 CPI

model calling 3.59% · releases 2026-10-15

pending

Live paper trades (1)

june-inflation-us-annual-20260610151727862

Entered 2026-07-03 · Settled 2026-07-15 · hcal-v1.0 · live

$-125.00

actual 3.5%

BucketEntry pxSizeEdgeP&L
3.9%10.0¢$59+22.8pp$-58.96
4.0%3.1¢$66+27.5pp$-66.04

Backtest · simulated

not live

The current model and entry rules replayed point-in-time on historical markets (1 entries). Most months correctly produce no trade — the market usually prices these prints well, and a small sample of monthly binary events cannot prove an edge by itself. That is what the scoreboard above is for.

Backtest P&L

$-125.00

on $1000 bankroll

Return

-12.5%

Markets entered

1

0 won

Avg P&L / entry

$-125

Cumulative P&L (paper bankroll $1,000)

Simulated trade history (1)

may-inflation-us-annual

Entered 2026-06-03 · Settled 2026-06-10 · hcal-v1.0 · backtest

$-125.00

actual 4.2%

BucketEntry pxSizeEdgeP&L
4.1%5.7¢$125+18.5pp$-125.00