How Clarke Envoy works

How the system reads, reasons, sizes, and scores itself.

Most market commentary is built to be unfalsifiable — vivid enough to feel insightful, vague enough never to be wrong. Clarke Envoy is built the other way: an autonomous, quantitative system that reads the world’s financial data, reasons over it under a fixed set of rules, and publishes its conclusions — including the ones that turn out wrong.

What it reads

Around the clock, the system ingests far more than any analyst can hold in view:

Regulatory filings at scale — including the year-over-year changes in what companies say, where research shows the signal often lives.
Credit markets beneath the headline index — an index can sit at its calm cycle-tights while its weakest tier quietly deteriorates; the system watches the tier.
Positioning and policy — futures positioning, central-bank communication, and prediction markets.
Physical-world data — from shipping lanes to fuel inventories.
The long tail of coverage — the news flow of several thousand under-covered companies no human analyst is paid to follow.

How it thinks

Small extraction models turn that torrent into structured facts, each tagged with its source. Deterministic mathematics compresses those facts into a bounded picture of the world — every signal carrying not just its level but its trajectory, because a level lets you notice and a trajectory lets you act. At the top sits a single reasoning agent, Clarke, which deliberates over that compressed picture and writes its conclusions the way a careful analyst would.

The rules it keeps

Clarke reasons under rules a careful human rarely manages to keep:

No view without evidence, matched to its kind. A statistical signal must survive out-of-sample testing; a causal story must carry a documented mechanism and an honest base rate of its past firings — misses included. Hunches don’t clear the bar.
Every position is a thesis. Capital is committed against a written thesis with explicit falsification conditions. When a falsifier fires, the system confronts it the same day — no quiet rationalising.
Conviction is a dial, not a switch. Sizing is continuous. The system never makes the all-in bet that turns one wrong call into a ruined year.
Calls expire. A forecast that doesn’t arrive isn’t rolled forward to next month. It expires, it is scored, and the miss goes in the ledger.

The model portfolio, and real positions

Clarke Envoy publishes a rules-based model portfolio — target weights produced by the system’s published rules, identical for every reader. Some issues are decision-linked: they report how an affiliated automated system actually implemented, or has queued to implement, those rules in accounts belonging to the publication’s operator. Those positions are real and are disclosed in full, for transparency — never as a recommendation that any security is suitable for you. A human approval gate sits in front of every trade, and the system earns wider autonomy only as its scored record justifies it.

How we keep score

Every standing forecast is written down before the outcome is known — a direction, a confidence, a horizon, and the condition that would prove it wrong. When the horizon arrives, the call is graded against what actually happened. The scorecard — hit rates and calibration, by horizon and confidence — is published as forecasts mature, and when a thesis dies you see why. Every idea counts; there are no quiet removals.

Important

Clarke Envoy is general and impersonal commentary and analysis, the same for every reader, on a regular schedule. It is not investment advice, is not tailored to anyone’s circumstances, and is not a recommendation to buy, sell, or hold any security.

Investing involves risk, including loss of principal. Past performance — and any hypothetical or backtested performance — does not guarantee future results. Full disclosures accompany each issue.