Trust Center

Data status

Every number displayed on this site, where it came from, when it was observed, and which of three tiers keeps it current. If you want to check us, start here.

Tier 1 · Delayed

Market prices

The Market Pulse strip and company quotes refresh every 15s while US markets are open and every 600s when they are closed. They come from Finnhub and are delayed by about 15 minutes. Polling stops when the market closes — a price that changes when nothing is trading is a lie about liquidity. The badge in the strip always says which state you are looking at.

Tier 2 · Scheduled

Economic data

CPI, payrolls, the fed funds target and mortgage rates come from the statistical agencies on a scheduled job. Polling them per visitor would be waste: CPI changes once a month, and the release calendar is published a year ahead.

Tier 3 · Human

Everything written

Article text, chart series, valuation assumptions and anything with no dependable feed. A number inside a sentence is part of an argument. When a live figure drifts away from a written one, the build flags it for an editor instead of quietly rewriting the story.

Auto-refreshed fields

Pulled from public data APIs

Manual fields

Entered and checked by a person

Open drift items

Figures awaiting an editorial rewrite

Last refresh run: 2026-09-19T20:44:44Z

Decided

2 figures we have looked at and left as published. An article is a record of what we argued on the day. We add a correction note rather than edit the argument, because a quietly corrected argument is no longer a record of anything.

The economy is sending two signals at once. The Fed is listening to the louder one. — Nonfarm payrolls, July 2026 published as −23,000, now +21,000 (revised 4 Sep 2026). The piece was published on 17 August and argued from the July payrolls figure as BLS published it that month. BLS revised July to +21,000 on 4 September. The analysis is not rewritten — it described the evidence that existed when it was written — but the article carries a correction note naming the newer figure, and the July release page states the revision above its own headline.
Jobs — July 2026 — Nonfarm payrolls, July 2026 published as −23,000, now +21,000 (revised 4 Sep 2026). The release archive preserves what was published on 7 August. The page states the revision to +21,000 above its own headline and links to the release that made it.

Auto-refreshedDisplay-layer figures only: the market strip and the economy headlines.

Fields refreshed by fetch.py
Field	Value	Observed	Source	Mode
S&P 500	7,637.76	Sep 17, 2026	S&P Dow Jones Indices via FRED	Auto
Nasdaq Comp	26,522.55	Sep 18, 2026	Nasdaq OMX via FRED	Auto
Dow	51,682.64	Sep 18, 2026	S&P Dow Jones Indices via FRED	Auto
10-yr Treasury	4.940%	Sep 17, 2026	U.S. Treasury via FRED	Auto
VIX	15.44	Sep 17, 2026	Cboe via FRED	Auto
WTI Crude	$107.02	Sep 15, 2026	U.S. EIA via FRED	Auto
Bitcoin	$80,844	Sep 18, 2026	Coinbase via FRED	Auto
Inflation (CPI)	3.4%	Aug 1, 2026	U.S. Bureau of Labor Statistics via FRED	Auto
Jobs	+162,000	Aug 1, 2026	U.S. Bureau of Labor Statistics via FRED	Auto
Fed funds	3.75–4.00%	Sep 19, 2026	Federal Reserve via FRED	Auto
Housing	6.95%	Sep 17, 2026	Freddie Mac via FRED	Auto

A failed fetch never blanks a field — the previous checked value stays, with its original date. Anything older than its release schedule allows is flagged Stale rather than quietly shown.

ManualNo dependable free feed, so a person enters and checks these.

Fields maintained by hand
Field	Value	Observed	Source	Mode
Russell 2000	3,068	Aug 14, 2026	Investrade	Manual
Gold	$4,375.50	Aug 14, 2026	Trading Economics	Manual

What is never automated

Numbers inside article text, the rows of an indicator page, and the series behind a chart are not touched by the refresh job. A sentence such as “strip out energy and inflation is 2.5%” is an argument built around a figure, not a display of it. Swapping the number underneath would leave an old conclusion attached to new evidence — which is the exact failure our editorial standards exist to prevent.

So the build compares the two and reports drift instead. A machine refreshes the dashboard; a person rewrites the analysis.

When the feed fails

The page never blanks and never freezes silently. If the upstream stops responding, the last good value stays on screen with its real age attached and the badge changes to Delayed or Offline. If the site is served as plain static files with no API behind it, the published closing figures render exactly as they always have — the live layer is an enhancement, not a dependency.

Corrections

If a figure here is wrong, tell us and it goes in the permanent corrections log — not a silent edit.

The queue itself

This page lists what we publish. The newsroom lists what we have not fixed yet: the drift queue ranked by how badly each figure moved, the health of the data layer right now, and a computed audit of every article — sourcing, labelling density, and the share of claims that are evidence rather than judgement. It is not flattering, which is rather the point of publishing it.