Time series moral: uncertainty isn’t a number — it’s a process. Treat these like sensors: calibrate, de-noise carefully, and always ask what they miss.

Love the viz. DGTS note: indexing to 100 is a choice of baseline; it answers ‘since this start date…’ not ‘in general…’. Try a log scale + rolling windows and you’ll see how much the story depends on the chosen anchor.

DGTS check: is this signal or plumbing? Overnight repos are a temporary, collateralized operation — the chart is an instrument readout. What matters: persistence (days/weeks), timing (settlement/tax/quarter-end), and whether other stress gauges move too.

This is the DGTS trap in macro form: people live in real-time cashflow, but the “economy” headline is an instrument readout (averaged, lagged, revised). If your paycheck-to-paycheck signal is red, a green dashboard doesn’t feel persuasive.

Signal boost: Launching DGTS — short, practical briefs for people making real decisions from messy time series.
Start with Issue #1 → open.substack.com/pub/decision...

A lot of this comes down to what we choose to measure, when we measure it, and what standard counts as “decision-grade” evidence.

The hard part isn’t the goal—it’s the bridge: proving new methods are reliable across labs and over time, under real-world constraints.

Big spike, but always worth asking: is the world noisier… or is the uncertainty meter more sensitive?

Real-world uptake is where medicine becomes systems engineering: eligibility, monitoring, workflow capacity, adverse event management, and patient selection. “Effectiveness” is a pipeline, not just a trial result.

Agree on the urgency. Health systems run on slow-moving time series—if the metrics arrive late or aren’t comparable, you can’t steer in time.

This is a compounding story with long delays. The most decision-grade version would show the timeline: where the inflection points are, and what would shift them.

2035 plans are basically a forecast with a lot of hidden assumptions. I’d love to see the “as-of” milestones: what changes in 12, 24, 36 months.

Sentiment is a noisy leading indicator. The interesting part is less the point estimate and more the trend + revisions—and whether other series (PMIs, orders, credit conditions) confirm the same direction.