signal triage for macro and ai feeds

see also: liquidity regime framework · decision boundary checklist for weekly publishing

intake funnel

Most feeds are ambient noise. The first job isn’t to read everything — it’s to filter fast and accurately.

Layer 1: Source classification I tag sources by reliability and horizon before I read the content:

• Economic data releases — high signal for the session, short half-life.
• Central bank communications — medium signal, multi-day echo.
• Industry reports and research — variable, depends on incentive alignment of the source.
• Social media and narrative feeds — low signal alone, useful as sentiment temperature.
• AI tooling news — high noise ratio, signal often buried in hype.

Layer 2: Claim extraction After source classification, I extract the actual claim, not the framing. The headline almost never matches the substance. I ask: what is the specific assertion? What would change my mind about it?

Layer 3: Confidence grading

• Confirmed — multiple independent sources, consistent with observable market behavior.
• Likely — one credible source, plausible mechanism, no contradicting evidence.
• Possible — interesting angle, but missing key confirmation.
• Noise — narrative without mechanism, or mechanism without evidence.

time-horizon sorting

I sort everything into buckets based on actionable horizon:

Trade week (1–5 days) — primarily data releases and positioning shifts. I care about the direction, not the story.

Month (5–30 days) — policy signals and trend confirmation. I care about whether the narrative is gaining or losing institutional sponsorship.

Quarter (30–90 days) — regime shifts and structural resets. I care about whether the framework needs updating.

Anything that can’t be placed in one of these buckets usually isn’t actionable yet, regardless of how interesting it sounds.

escalation rule

A signal gets promoted to publication when:

1. It changes a position-relevant view, not just adds texture.
2. It survives at least one disconfirming check.
3. It can be stated clearly without the hedging that comes from uncertainty.

If I can’t state it cleanly, I hold it as a working note until the signal strengthens or the uncertainty resolves.

ai feed specific problems

AI news has structural overhype problems that macro feeds don’t:

• Vendor announcements are not market signals.
• Benchmark results are not deployment reality.
• Demo quality is not product quality.
• Regulatory concern is not regulatory action.

I apply a higher bar to AI signals before treating them as regime-relevant. Most of what lands in AI feeds is either already priced or won’t arrive for years.

my take

Signal triage is a discipline, not a talent. The people who are best at it aren’t smarter — they just have stricter filters and faster rejection.

linkage

• [[liquidity regime framework]]
• [[decision boundary checklist for weekly publishing]]
• [[ai compute bottlenecks chips power and deployment lag]]

ending questions

which signal type has the highest rate of reversal within 30 days of publication?