Designed for the bigger screen.Open the mobile app →
Meridian is not a model that predicts markets. It is a synthesis layer that reads what the best independent voices are already saying, weighs them against their track record, and identifies when their views converge into something worth acting on. Selectivity is the entire point.
We continuously read the work of approximately 320 independent macro voices, sector specialists, and primary sources across the financial internet. Newsletters, podcasts, X posts, research notes, central bank communications, regulatory filings, exchange data, and more. The breadth is non-trivial: most of what is published is noise; what we are looking for is the small fraction that is signal.
Our monitoring layer is automated where it is honest to be automated — extracting positions, theses, and price levels from public writing — and curated by editors where judgment is required. We never paraphrase a source's view; we summarise it and link to the original. The Knowledge layer of the publication is, in part, a transparent record of what we are reading.
Not every voice carries the same weight. We maintain a track record of every public position from every monitored source, going back as far as their public history allows. Weights are computed from calibrated accuracy — not just whether a source was directionally right, but whether stated conviction matched realised outcomes across comparable calls.
Weights are recomputed quarterly. Sources are added and removed. No single source has a weighting above 4% of the network. The point is independent alignment across many framings, not concentration on one star analyst.
The synthesis layer is where the publication exists. We do not aggregate by majority vote — we look for independent alignment across framings. A signal at 96 means that several of our highest-weighted sources, working from different analytical traditions, have arrived at the same directional read on the same asset within a similar time window.
The Conviction Score is the publication's central output. Scores below 70 are not actionable. Scores between 70 and 85 are flagged for monitoring. Scores above 85 are signal candidates. Above 90 is rare; above 95 is the highest tier we publish. Some weeks produce several signals. Most days, we publish nothing.
Every signal we publish is logged at issuance — entry, target, stop, conviction, time horizon, source set. The log is append-only. We do not edit signals after the fact. We do not delete losing trades. We publish wins, losses, and pending side-by-side in The Record.
We backtest the methodology continuously, on the historical data of the source network. Signals that the system would have published in the past — with conviction at issuance, on the data that was available at the time — are tracked against actual outcomes. The receipts are the entire brand. If the methodology stops working, we will tell you, transparently.
We are a publication, not a model. The model is in service of the publication.
Entry, target, stop, outcome, return, and conviction at issuance.
View the record → Today's BriefingCross-asset summary, signal of the day, what we're watching.
Read the brief →