01 · The problem
Conventional risk frameworks work until the moment they matter most.
Most professional risk analysis rests on assumptions that are rarely spoken aloud: that events can be understood one at a time, that the historical relationships between asset classes will hold, that volatility will revert, and that the consensus story about the world is broadly right about its shape. In stable periods those assumptions are good enough to produce useful analysis most of the time. They fail in the periods of structural change — when correlations that held for decades break in days and the story is wrong about the shape of the world rather than its details. The 1987 crash, the 1998 Russian default, 2008 and March 2020 were each a failure of this kind.
The clearest illustration comes from finance. The Black–Scholes model, the mathematics underneath most modern risk management, assumes smooth prices, constant volatility, liquid markets and stable correlations. In 1998 Long-Term Capital Management — a fund that counted two of the model's authors among its partners and had returned more than forty per cent a year — lost $4.6 billion in four months when the Russian crisis broke correlations the model treated as fixed. The model was not wrong in normal conditions. The partners trusted it beyond the evidence that it was breaking.
The lesson is not that models are useless; it is that models are tools with limits, and the discipline is knowing when the tool has stopped working. A method for reading the world therefore needs explicit machinery for detecting when the assumptions behind conventional analysis are starting to fail — and it needs the same machinery turned on itself, because an analyst who has formed a view will find every headline confirming it. This method is that machinery, written down so it can be checked.
02 · The four tests
Which events are structural, and which are noise.
Most events are noise: politically loud, structurally irrelevant. Each candidate event is put through four tests. Passing all four makes it high priority; passing none makes it noise; one or two means keep watching.
Reversibility
Could a change of government, a diplomatic agreement or a market correction undo it? Structural events are irreversible at the timescales that matter.
Multi-domain
Does it show up in several domains at once — markets, alliances, physical flows, institutions — or in only one?
Institutional
Does it require an institution to be created, dissolved or rewritten, or is it behaviour inside existing institutions? Behaviour is usually noise.
Reaction
How are sophisticated actors with something at stake responding? When conservative institutions change long-held positions, the signal is more reliable than the headline.
03 · The cluster rule
One signal flashing is normal. Several at once is the model cracking.
Twenty signals are watched across five categories. Each has an explicit activation threshold and is graded only on sustained movement, never on a single print. Individual signals are noisy; structural change shows up as several normal relationships breaking together — the pattern that broke LTCM and that conventional frameworks systematically underweight. The number of signals active, weighted for partial activations, sets the response.
Three examples of the kind of threshold used: the dollar's share of allocated reserves falling below 55%; a high-yield credit spread widening by more than 200 basis points in a short period; a major Gulf producer leaving OPEC. The full signal library and its thresholds are the working material of the service and are not published.
The highlighted tier is the current reading. The composition of the cluster — which categories are flashing — is reported alongside the count, because it is usually the finding.
04 · The verification pass
What gets checked before anything is graded.
Every claim that would change a signal's status or the cluster reading, and every claim of a kind that is costly to get wrong: a change of leader or government, a central-bank decision, a treaty or defence pact, an election result, a reserve, holdings or yield figure that moves a signal across its threshold.
Primary sources where they exist — the central bank, the statistics office, the treaty text, the official result, the filing — and otherwise two independent reputable reports. Where sources conflict, the conflict is recorded rather than resolved by preference.
An independent verification pass that has not seen the research strands, so the work is not grading itself; then the human owner, who reads the verification report before any grade is set and before the edition is signed.
A claim that fails verification is removed or marked unverified in the edition; a figure that cannot be confirmed is never used to move a status. Both outcomes are visible in the research notes behind each sweep.
05 · The human owner
One person is responsible for every edition.
TO FOLLOW
Philip Bennett
Founder, Nownet · author of the methodology
Author of Connecting the Dots: A Framework for Geopolitical Research Under Uncertainty (May 2026) and of the State of the World series, which has applied the method monthly since March 2026.
Founder of Nownet, a consultancy with a background in cybersecurity — the discipline of monitoring for the conditions under which systems fail — from which the signal framework was developed.
Sets the logic, verifies the high-impact claims, signs every edition and takes editorial responsibility for its contents.
06 · Sources policy
Every grade can be traced to a dated source.
- Primary sources preferred. Official statistics, central banks, treaty texts, filings and election authorities come before reporting; reputable reporting comes before commentary; forums and unattributed sources are not used.
- Every grade is traceable. The research notes behind each sweep record the source and the date for each figure that supports a status, so any judgement can be followed back to the evidence as it stood that day.
- Conflicts are flagged, not hidden. Where sources disagree or a figure could not be confirmed, the edition says so; the figure is not used to move a status.
- Nothing is backdated. Editions are appended, never rewritten; the ledger records what was expected when it was expected. A thesis that cannot be wrong is not a thesis.
- The thesis is held open. The method's own bias discipline — confirmation bias, recency, narrative overfitting, pattern-matching — is applied to the owner's view as to anyone else's. When new evidence confirms the thesis, the posture is more caution, not more confidence.