Methodology
A defensible route from an observable development to a decision.
What follows is the method itself — enough of it to judge whether it is rigorous, and enough for a client to challenge our work precisely. The internal models, weights and index specifications are not published; the reasoning that governs them is.
The Signal-to-Range Framework
Eight stages, kept apart on purpose
Each stage answers a different question and can be wrong on its own terms. Keeping them apart is what allows a disagreement to be located rather than merely felt.
Signal
What is changing · 2 stages
Range
What it could cost · 4 stages
Decision
What management does · 2 stages
Open any stage — arrow keys work too
Signal · Signal
What has changed, and is it material?
An observable development — a policy move, an escalation, a legal change, a macroeconomic shift — identified early enough to be useful, and filtered against exposure so that noise does not reach the executive.
A signal is an observation. It carries no consequence until it is attached to a scenario.
Scenario construction
A conditional, not a prediction
We build four variants of a situation — baseline, downside, severe and tail — as a single family rather than as four separate stories. Building them together is what keeps the probabilities coherent and makes the comparison between them meaningful.
Each variant is defined by its assumptions, not by its severity. “Severe” is not a mood; it is a specific set of conditions holding at the same time, written down before any modelling begins — so the number that comes out at the end traces back to a claim someone can dispute. Probabilities are given as intervals, and every material change to one is recorded with its reason.
Illustrative example — constructed to show the method. Uncertainty widens with the horizon; the output of a variant is an interval, never a point.
- Baseline55–65%
- Harassment of shipping continues at the current tempo. No sustained closure. War-risk premia stay elevated but stable, and rerouting is limited to individual operators.
- Downside20–30%
- Intermittent interdiction forces convoying. Insurers restrict cover for some flags and operators, and transit times lengthen materially.
- Severe6–12%
- Partial closure lasting several weeks. Rerouting where physically possible, emergency stock releases, and rationing of war-risk capacity.
- Tail1–4%
- Sustained closure beyond a quarter with regional escalation, capital-flow disruption in the most exposed importing markets and emergency policy response.
Transmission, vulnerability and exposure
Nothing affects a business. It travels, then it lands
A scenario reaches commercial reality through named channels, in an order that can be checked. This is the step most analysis skips, and it is the step that makes a later number arguable rather than asserted.
War-risk insurance
Hull and cargo premia for Gulf transits; availability of cover by flag and operator.
Freight & routing
Charter rates, rerouting distance, transit time and schedule reliability.
Energy
Crude and refined product prices, regional gas, and the cost of everything indexed to them.
Inflation
Headline and imported inflation in energy- and food-import-dependent markets.
Currency & policy
Import cover, currency pressure, policy tightening and, at the extreme, capital controls.
Demand & purchasing power
Discretionary spend, subscription retention, advertiser budgets and price sensitivity.
Channels are selected for the scenario and the client, not applied from a standard list. A business whose exposure sits in advertising revenue and one whose exposure sits in landed hardware cost are reached by the same event through entirely different routes, at different speeds. Where a channel does not apply, we say so rather than carrying it at zero.
Held apart, then combined in the open
External risk, market vulnerability, company exposure and mitigation are separate judgements. They interact, but they are never collapsed into a single unexplained score — because a reader who cannot open the number cannot challenge it.
Step through the four assessments
The market, assessed on its own terms
Political, regulatory and macroeconomic conditions in each market — an assessment of the market, entirely independent of who is operating in it.
These are four separate assessments, shown in sequence. They are combined explicitly and can be pulled apart again — which is why no single number is presented on its own.
How a range is built
A number you cannot take apart is not evidence
An impact range is the sum of stated assumptions, each with its own interval and its own confidence. It is not produced by a model that a client cannot open.
Mitigation already in place is modelled as a separate, negative contribution rather than netted off silently. Reporting gross risk alone overstates the problem; reporting residual risk alone hides the work the company has already done.
Switching an assumption off, as you can here, is exactly the exercise we run in a challenge session. It shows which single belief the conclusion is resting on — and in most engagements, one or two assumptions carry most of the range.
Switch an assumption off — the range recomputes
Resulting range
Indexed landed cost (plan = 100)
101.0 — 125.0
All assumptions included. This is the full stated position.
Illustrative example — constructed to show the method. Severe variant, indexed landed cost.
Trigger design
Decide in advance what would change your mind
A trigger is an observable tied to a named assumption, with a threshold agreed before the pressure arrives and a decision attached to its crossing. It proposes a review. It never rewrites an assessment on its own.
The tick sits at the same point on every track — the distance to it is the comparison
War-risk premium, Gulf transits
Above 0.7% of hull value, sustained for two weeks
Approaching
On crossing — Move to downside assumptions. Analyst review convened within five working days.
Carrier transit suspensions
Two or more major carriers suspend Gulf transits for more than seven days
Within tolerance
On crossing — Escalate to the severe branch. Activate the pre-agreed freight contingency.
Regional benchmark crude
Sustained above +25% versus the plan assumption for ten trading days
Within tolerance
On crossing — Re-run the pass-through model. Bring the pricing decision forward.
Most candidate indicators are rejected. An indicator earns its place only if it is observable without privileged access, moves before the outcome rather than with it, and has a decision attached. Frameworks fail far more often from having too many indicators than too few.
A crossing is interpreted by an analyst before it reaches you. Automated alerting without interpretation produces exactly the fatigue that causes real signals to be ignored — which is the failure mode monitoring exists to prevent.
Illustrative example — constructed to show the method. Standings shown are constructed.
Analytical distinctions
What we refuse to collapse
Most of the failures we are asked to correct come from two things being treated as one. These are the pairs we keep apart in every deliverable.
Scenario is not Forecast
A scenario says what follows if a set of assumptions holds. A forecast says what will happen. We publish the first and are explicit that we do not sell the second.
Probability is not Confidence
Probability is our estimate of how likely a scenario is. Confidence is how much weight our own judgement can bear given the evidence. A 40% scenario can be held with high confidence; a 90% one with low.
Evidence quality is not Source prestige
A well-known outlet reporting a single anonymous official is weaker evidence than an obscure but primary regulatory filing. We grade what the source can actually establish.
External risk is not Company exposure
The first is a property of the market. The second is a property of your business. Multiplying them without showing both is how a country score gets mistaken for a company assessment.
Impact is not Materiality
Impact is what a scenario does. Materiality is whether that is large enough, relative to your position, to change a decision. Many high-impact scenarios are immaterial to a given company.
Evidence, review and accountability
What a source can establish, and who signs it
Two separate fields, reported separately, on every assessment. Averaging them would destroy the information a reader most needs.
Evidence grade
- A
- Primary and verifiableOfficial filings, statute, published data, first-hand documentation.
- B
- Reliable and corroboratedEstablished source, independently corroborated, consistent with the record.
- C
- Plausible, uncorroboratedCredible reporting that has not been independently confirmed.
- D
- Weak or contestedSingle anonymous account, interested party, or contradicted by better evidence.
Grade reflects what the source can establish, not how well known it is. A primary filing outranks a prestigious outlet quoting an unnamed official.
Analytical confidence
- High
- Well-evidenced, mechanism understood, several independent sources, robust to any single one being wrong.
- Moderate
- Reasonable evidence with identified gaps. A specific piece of new information could move the judgement.
- Low
- Thin or contested evidence, or a mechanism we do not fully understand. Stated with what would raise it.
Where confidence is low we state what evidence would raise it. That turns a caveat into a research task instead of a disclaimer.
Extract from an assessment
Select any underlined claim
Since the beginning of the month war-risk premia for transits through the corridor have risen materially, and two operators are reported to have paused transits. On the current evidence we assess the probability of partial closure at 6 to 12 per cent over the next two quarters. Under that variant, landed cost would rise by 10 to 22 index points in the severe variant, before mitigation.
Every claim in a delivered assessment carries this record — not as an appendix, but attached to the sentence it supports.
- Classification
- Reported claim
- Source
- Two independent trade publications
- Evidence grade
- C — plausible, not independently confirmed
- Analytical confidence
- Moderate
- Last reviewed
- 14 July 2026
- Ownership
- Lead analyst, energy & maritime
- Contradiction log
- One operator has publicly denied a suspension. Logged rather than resolved; the assessment carries both readings.
Analysts sign the conclusion
We use structured research databases, data pipelines, statistical models, document extraction and language models internally. None of them decide a probability, set an assumption or publish an assessment.
- Evidence traceability
- Every conclusion retains the sources that support it, with corroboration recorded and contradictions logged rather than resolved silently.
- Source grading
- Sources are graded on what they can establish, not on their reputation. A primary filing outranks a well-known outlet quoting an unnamed official.
- Assumption registers
- Every quantitative output travels with the assumptions that produced it. A range without its register is not a deliverable.
- Confidence, stated separately
- Analytical confidence and evidence quality are two fields, always reported apart from the probability itself.
- Model versioning
- Methodology changes create a new version. Prior assessments remain readable under the method that produced them.
- Named analyst ownership
- Every assessment has an author. Technology accelerates research and monitoring; it does not sign the conclusion.
- Declared limitations
- Where the evidence does not support precision, we say so. Where a judgement rests on a single source, that is stated on the page it appears.
- Confidentiality
- Client information is held separately and is not reused across engagements. NDAs are available where appropriate, and no client work appears in public material without written permission.