How the Fed Watched This Week

Signals from FOMC, regional banks, and econ data — as the record shows them

The Federal Reserve does not issue one signal a week. It issues many, through many voices, across many venues — statements, minutes, regional speeches, district surveys, the soft-power commentary of bank economists who read the same data the FOMC reads. The job of a structured record is not to summarize that noise. It is to preserve each claim, each source, each evidence type, and let the reader reconstruct the signal themselves.

This thread does that for one week in April 2026. The claims below are pulled live from the Synorb claim graph — the same rows an agent would receive via the API. No rewriting. No editorial softening of the numbers. If a source corrects a figure tomorrow, the block updates; if a record is retracted, the block marks it unresolved. That is the point of the exercise.

What the markets heard

The most audible voice in any Fed week is not the Fed itself. It is the weekly-update desk — the bank economists, the insurance-adjacent advisory shops, the newsletter writers who translate the FOMC's silence into a number someone can put on a slide. The numbers they choose to foreground tell you what the buy-side is actually watching.

A 12.4% rally off a March 30th low is the kind of figure that carries its own argument. It is a post-uncertainty number — the market deciding, for now, that whatever the Fed is going to do next is already in the price. Whether that conviction holds through the next CPI print is a different question. The claim is dated, sourced, and evidence-typed. The reader can verify it; the agent can ingest it.

What small business felt

The Fed can set a rate. It cannot set a mood. The firms that actually hire people, raise prices, and decide whether to extend inventory do not live inside FOMC communications — they live inside sentiment surveys, which are slower and noisier but arguably closer to the ground truth of policy transmission.

A 95.8 reading on the NFIB index is a specific number with a specific history. The long-run average sits near 98. A sub-96 print, persistent across months, is the kind of signal that used to precede labor-market softening in every cycle for which the series exists. Whether it does this time is unknowable from this data point alone. What is knowable is the number, the month, and the source.

What the advisory layer said about itself

A thing that happens quietly in every Fed week: the advisory and investment-advice industry re-states its own boundaries. These disclosures are not interesting as macroeconomic signals — but they are interesting as a case study in why structured extraction matters. An agent reasoning about the macro picture should know that the same record containing the S&P 500 figure above also contains the following kind of self-limiting language:

Both of these claims are extracted from the same weekly-update record as the 12.4% number. A downstream reasoning system that treats the whole record as one blob will mix editorial prose with regulatory boilerplate with hard market data. A system that treats each claim as its own typed assertion — with evidence_type, confidence, and a canonical source — can filter the boilerplate without throwing the data away.

What the week actually said

Put the claims side by side. A rally. A sentiment reading. A disclosure about the advisory's own status. That is not a synthesis — it is the material an analyst would use to build one. The structured record exposes the inputs. The inference is still the reader's, or the agent's, to make.

The Fed did not speak with one voice this week. It never does. What is new is that the thousands of voices that speak around it can now be captured atomically — claim by claim, source by source, date by date — rather than flattened into a newsletter paragraph. That is the step from "a summary of the week" to "the week itself, available for reasoning." Every claim block above is a door. Click the source link, read the record, audit the extraction. The thread is an argument; the claims are the evidence; the evidence is checkable.

That is what evidence-first publishing looks like when the reader is sometimes a person and sometimes a machine.