Nexus brain markNexus brain
Read the report

An answer engine hands you a verdict.
A decision needs a distribution.

Nexus brain runs a team of AI agents on your question. They investigate the live web, argue the sources out, and file an ORACLE report: what is proven, what is guessed, and what would change the answer. Every claim dated.

You can run your own question here. The proof below was the founding run, by hand, Oct 17, 2026.

One question, two engines

“Will the Federal Reserve cut rates at the October 2026 FOMC meeting?”

The answer engine said

The Fed is highly unlikely to cut. Hold probability 83.9–87.3%, hike probability 15–17%. Core PCE remains elevated, around 2.8–3.7% YoY. Conclusion: hold is the base case.

Real output, Brave AI Answers, Oct 17, 2026. No dates on any figure.

The ORACLE report said
  • PROVEN Fed raised 25bp on Sep 16 to 3.75–4.00% — first hike since Jul 2023. federalreserve.gov
  • PROVEN FOMC meets Oct 27–28. Fed calendar. Hot Sep CPI (Reuters, Oct 14) → core PCE est. ~3.1%.
  • GUESSED 84.5% hold — market pricing, CME FedWatch, Oct 17. A guess, not a fact.
  • DISPUTED Reuters, Oct 14: a hike is “back on the table.” CNBC, Oct 16: markets “dangerously complacent.” Against Williams: “no need for urgency.”
  • GUESSED What would change the answer: the Oct-vs-Dec fork for the one more 2026 hike the Fed itself projected; FedWatch below ~75%; Warsh rhetoric before the blackout. The decisive Sep PCE print lands Oct 30 — after the meeting.

The engine’s headline range, split by date

“Hold probability 83.9–87.3%” — one number, no day attached. The sources behind it:

  • 84.5% holdCME FedWatch, Oct 17
  • 15% hikeCNBC, Oct 16
  • 78.2% no change through DecSchwab — stale: still quotes the pre-hike 3.50–3.75% range

Three days, one blend. A reader who must defend the decision needs the days.

One answer-engine call, four web searches. The bottom line agreed. The audit trail did not exist. Full run: docs/demo in the company repo.

What our report shows that theirs does not

  • Verdicts. “84.5% hold” is market pricing, not a fact. The answer engine states it as one. We stamp it GUESSED.
  • The disagreement. The two freshest sources argue the market is complacent on hike risk. The engine averaged that fight away. We surface it, dated.
  • Stale sources, exposed. A widely-quoted range still claims the pre-hike rate. Blended ranges quietly mix such snapshots. Dated claims make it visible.
  • What would change the answer. The one section a decision-maker uses: which datapoint, on which date, flips the call.
  • Primary sources. The rate and the meeting date from federalreserve.gov, not commentary.

The limits

Same web, same facts. Our claim is structure — verdicts, dates, conflicts, sensitivities — not better facts. On this question both engines reached the same bottom line. The difference is what you can defend when someone asks how sure you are.

What would you ask?

The method above is real. Your question gets its own ledger — fresh from the live web, every claim dated.

Run your question