How the desk reads, in full.

This page sets out the discipline the desk applies, in writing, to every brief that ships. It is published openly so that a reader can audit the practice before they trust a single read. The method is the product, and the method is open.

The unit of work

The unit of work is the brief. A brief is a written position on a specific posture in a specific jurisdiction at a specific moment, attached to specific structures on the subscriber list. It is not a feed. It is not an alert. It is a paragraph, sometimes several, that names what moved, what it implies, and what to consider. The brief is read in an inbox by a principal who has structures across borders, on their own time.

Source ingestion

The desk reads primary documents. A primary document is one issued by the body whose posture is being read. A central bank circular is a primary document. A reporter's account of the circular is reference, not source. An aggregator's headline is an alert that something has been issued, not the issued thing itself. Where a primary document is in a language other than English, the desk reads the original. The translation choice is footnoted in the editorial draft.

Standing source classes the desk reads each working day:

• Sanctions packages. OFAC SDN deltas, OFSI consolidated updates, EU sanctions packages, UN Security Council designations, and national equivalents. Read at publication. Named-vehicle and beneficial-owner traces extracted same day.
• Central bank circulars. Read in the original language. The change in language between consecutive circulars is often the signal, not the new circular itself.
• Regulatory consultations. Consultation papers, named-respondent submissions, and the regulator's response paper. Cadence and the identity of named respondents matter as much as the text.
• Court filings. Originating court documents, in PACER, the UK Courts and Tribunals service, and equivalents across covered jurisdictions. First-instance filings carry posture earlier than appellate judgments.
• Companies and beneficial-ownership registers. Filings, not summaries. Where access is gated, the gating itself is recorded as a posture signal.
• Audited financial filings. Of banks, insurers, asset managers, and listed corporates. Read for what counsel removed between drafts as much as for what remained.

Confidence scoring

Every claim in a published brief carries a confidence score from one to five, recorded at the time of writing. The score is fixed at publication and is not retroactively adjusted. The scale is below.

Falsification paths

Every claim ships with the conditions under which the claim would be wrong. This is not a hedge. It is a written test, recorded at the time of writing. If the brief says a banking corridor is tightening, the falsification path names the specific data the desk would expect to see if the brief were wrong: a counter-cadence in correspondent filings, a softening in supervisory language, a renewed bilateral protocol on the calendar. Quarterly review applies the test against published reality, and the result feeds the miss-rate.

A read that cannot be wrong is not a read. It is an opinion with citations.

Calibration discipline

Calibration is the willingness to be measured. The desk publishes a quarterly calibration record that names every claim from the prior quarter, the prior confidence, the falsification path that triggered, and what the desk did not see in time. The current quarter's record sits at mebusan.ai/calibration-q1-2026.

The miss-rate is reported separately at each confidence level. A read that was wrong at confidence five is a more serious miss than a read that was wrong at confidence two, because the desk asked the reader to act on the higher-confidence read with less hedge. The public record reflects that.

Three measures travel each quarter. Banking corridor precision and recall against the working set of corridors the desk follows. Pipeline reads, counted as the proportion of regulatory-pipeline calls that landed inside their published forecast window. Sanctions routing latency, measured as the median minutes between OFAC SDN delta publication and a structure-specific note reaching subscribers.

Routing windows by tier

Once a brief is filed, it routes to subscribers on a tier-defined window. The Letter, the public quarterly read, ships to all addresses on the same day. Subscriber tiers route earlier, configured to the principal's actual situation: the banking corridor they are reading, the residency regime they are about to enter, the exchange exposure they hold. The window is a service commitment, not a marketing claim, and is published tier by tier on the desk.

Editorial standards

A brief is written by a single named operator and reviewed by a second before it ships. The review is for falsification path, for confidence honesty, and for the absence of language the desk does not use. The desk does not use sanctioning verbs about subjects it has not named in writing. The desk does not predict outcomes it cannot falsify. The desk does not write in the voice of certainty when the underlying read is interpretive.

What this is not

This is not a model. It is not a scoring engine. It is not a feed of signals dressed as analysis. It is a discipline applied by a small desk reading primary documents, recording confidence honestly, accepting public correction when the read was wrong, and writing for a small group of subscribers who have structures across borders.

The corporate record sits at mebusan.com. The live desk and subscriptions sit at mebusan.ai. The discipline sits here.