notcfo runs two live forecasting systems in the open: an on-demand Oracle you can question yourself, and a standing Orchestra that publishes calls on a schedule and is judged against what actually happened. Built and maintained by a twenty-year debt capital markets banker who put his name on the record.
One entry per domain, distilled from the same live research that feeds the Oracle and the Orchestra below — refreshed daily, not on request.
Ask it anything forecastable, and it runs itself. An autonomous web search gathers live signal for your question — Sensing. Five independent reasoning passes argue it out with no cross-talk between them — Deliberating. A synthesis pass converges on one consensus forecast and names its strongest disagreement rather than hiding it — Speaking. Four horizons, every time: 24 hours, one week, one month, one year.
Kept in this tab's memory only — never saved to disk, never sent anywhere but Anthropic's API directly. Clears on reload. Get a key at console.anthropic.com/settings/keys.
The same three-stage pipeline as the Oracle above — Sensing, Deliberating, Speaking — but no one has to ask, and fifty voices run it instead of five. A scheduled job runs it automatically once a day against five fixed questions, one per coverage domain, start to finish with no person triggering it. What isn't automated, on purpose: whether a call was right. That's recorded by hand below.
Short, dated notes on the events actually moving markets and mandates — written, not generated. This is where judgment shows its work.
Growth, inflation, and employment across the US and EU — tracked through the data releases that actually move rates.
Equity indices, credit spreads, yields, and FX — the funding environment DCM actually operates in.
Bitcoin price structure, leverage, and capital flows across ETFs and stablecoins.
Conflicts, elections, tariffs, and the regulatory shifts that reprice risk overnight.
Model releases, compute buildouts, and the power infrastructure now underwriting the AI industry.
A forecast is only as good as who's willing to be wrong about it. Here, that's a human and a machine — both by name.