Why I built Recession Tracker — and why I made it argue with itself

I'm obsessed with the economy and economic data. I kept catching myself doing the same thing: opening five different tabs, cross-referencing the yield curve on FRED, checking the VIX on one site, hunting for the latest ISM numbers on another. There was no shortage of recession data — the problem was it was scattered everywhere, and nothing told me how to weigh any of it against anything else.

So I built Recession Tracker for myself first. But the more I used it, the more I realized the real design challenge wasn't fetching the data — it was stopping myself from treating it too confidently.

Here's what bothered me about every dashboard I found: they all just show you red, yellow, and green lights and let you draw your own conclusions. But not all indicators are equal. The 3-month/10-year yield curve has a strong historical track record. ISM Manufacturing PMI is mostly a coincident indicator — by the time it turns red, you're already in trouble. If you weight them the same, you get a meaningless average.

So I went deep on the historical literature and tiered the signals. Tier 1 indicators (the TED spread equivalent, the 3M–10Y curve, HY credit spreads, the Sahm Rule) get 3× weight because they have genuinely strong pre-recession track records. Tier 5 signals like the 10-year yield level get 0.5× — useful context, not a leading signal.

The other thing I built in that I'm most proud of: four AI panels that actively challenge the score. Not just "here's the analysis." Literally a bear case arguing the score is too low, a bull case arguing it's too high, and a blind spots panel listing what the dashboard can't see at all — geopolitical tail risks, policy paralysis, off-balance-sheet credit deterioration. The dashboard should make you think, not just reassure you.

Right now the dashboard is tracking something genuinely interesting: the yield curve and Sahm Rule (the two strongest leading indicators) are both green, but consumer sentiment just hit an all-time record low at 48.2 — below levels seen at the worst of the 2008 crisis. The dashboard holds those two things in tension rather than averaging them away.

Happy to answer any questions about methodology, data sources, or what I'd build next. What recession signals are you watching that you wish were easier to track?