STRATUM.

What is the true strategic question behind the request as expressed?

No analysis is worth more than the question it claims to address. It is a banality which must nonetheless be recalled, because it is most often forgotten. Most mandates begin with a vague question that mixes several enquiries: some belonging to strategic analysis, some to operational advice, others still to a simple anxiety seeking reassurance through objectification. Our first task is to separate these layers, and to reformulate with the principal, in writing and explicitly, the precise question or questions that the analysis will answer.

This reformulation produces an artefact called the framing note. It states the intelligence requirements retained, the temporal and geographical perimeter of the analysis, the actors or objects at the heart of the question and those explicitly excluded from it, and the format of the expected deliverable. The note is validated by the principal before a single minute of collection is engaged. A neglected framing is a mandate that ends, six weeks later, with an embarrassed conversation about the fact that the note delivered brilliantly answers a question no one had really asked.

The methodological pitfall of this phase is known to every experienced analyst: the temptation to begin collection while the framing remains vague, on the belief that the material will progressively illuminate the question. It is an illusion. Material collected without clean framing never causes the analysis to converge on a clear question; it produces a disparate corpus that renders any subsequent ranking impossible.

What sources, what resources and what timeline are needed to answer this question?

No plan survives contact with the material. The formula was uttered by a nineteenth-century military strategist about operational plans, but it applies equally to plans of analysis. The documentary corpus that an analyst discovers once collection is engaged is never exactly the one anticipated. Sources reputed to be rich prove silent on the question posed. Sources he had set aside for their marginality turn out to document the very heart of the subject. Initial hypotheses deform on contact with the facts, and some must be abandoned even before the analysis phase formally begins.

One might conclude from this observation that planning is useless, since it is in any case condemned to be amended. That would be a false lesson. The plan does not exist to predict the future of the enquiry; it exists to render visible the gaps between what was foreseen and what is happening. Without a plan, the analyst who deviates from his trajectory does not know that he is deviating; he believes he is following a legitimate intuition when he is in fact only following his fatigue, his mood, or the first source that struck him as appealing. With a plan, the same analyst knows he is straying, can amend in full awareness, and keeps a record of that amendment.

Before moving through unknown terrain, one consults a map. Planning a mandate is precisely that: the preliminary cartography of the informational territory on which the analysis will move. This map is not meant to show everything — a map that shows everything is by definition unreadable. It hierarchises. It distinguishes principal axes from secondary paths. It identifies zones of documentary density and the deserts of information where one would exhaust oneself searching. It marks the points at which one will perhaps need to stop and verify. Without that map, the analyst advances through the material on instinct, which is precisely what the method claims to banish.

Drawing this map amounts to inventorying the documentary territories accessible to the enquiry. Some are dense plateaux, frequented and well surveyed: official publications and administrative documents, specialised press and sectoral publications, academic work and institutional reports. Others belong to deeper strata, less visible but often decisive: regulatory databases, economic registries, geospatial data, open archives. Still others fall under a human geography, where information is gathered through direct exchange with documentary sources, within a strictly defined ethical framework. For each of these territories, the analyst evaluates three things at the moment of drawing the map: relevance to the question posed, cost of access in analytical time, and the methodological risk associated with the use of the source.

The artefact produced is a collection plan that hierarchises these territories in order of priority, allocates the available analytical resources across the dimensions of the question, and sets an internal schedule of milestones. This plan is not fixed. It may be amended during the course of analysis if early collections reveal that certain promised seams are dry and that others, neglected at the outset, deserve to be explored. But the amendment is documented, dated, justified. An undocumented drift is no longer an enquiry adapting itself; it is an enquiry losing itself.

The typical pitfall of this phase is the rush into collection without preliminary architecture. It generally takes the form of an analyst who, after reading the framing, immediately opens fifteen press tabs to begin forming an idea. That idea, formed before any protocol of collection, then becomes the filter through which the rest of the material will be read. The diagnosis is corrupted before the method has even begun.

How to acquire information according to protocols that preserve its reliability?

Collection executes the plan. It obeys two fundamental rules. The first: privilege primary sources over derived sources. An official communiqué weighs more than a press article that summarises it; an administrative report weighs more than an editorial that comments on it; a legal document weighs more than a journalistic synthesis of the same document. This hierarchy is not an academic dogma, it is a constraint of reliability: at every relay, information loses precision and gains interpretation.

The second rule: maintain for each piece a provenance file. Date of access, URL or physical reference, context of production of the source, version retained when the source is revisable. This traceability conditions all subsequent analytical quality: without it, the phases of evaluation, analysis and synthesis lose their anchor and become exercises in approximate memory.

Concretely, the work of collection produces three artefacts: a corpus organised according to the nomenclature of the collection plan, a time-stamped journal recording each acquisition, and a provenance file for each piece. Before passing to the next phase, the analyst verifies that these three artefacts are complete. A piece without a provenance file is, in STRATUM, like an ungraded source: it cannot serve to ground a public assertion.

The pitfall of this phase is collection by capillarity. The analyst, set out to follow a precise source, follows a link that leads to another, then to yet another, and finds himself four hours later with a considerable corpus whose structure no longer follows any plan. Quantity has replaced quality, and the analyst himself no longer knows exactly why he retained one piece and not another. The discipline of the collection plan, and the traceability of the journal, are the safeguards against this silent drift.

What confidence can be placed in each piece of the corpus?

The evaluation phase is the one in which the raw corpus becomes an exploitable corpus. Each piece is rated along a two-dimensional grid, one bearing on the reliability of the producer of the information, the other on the plausibility of the information reported, in accordance with the professional conventions long established in Western analytical services. A highly reliable source may report a hardly plausible piece of information; an unreliable source may report information whose plausibility is high. The two dimensions must be evaluated separately, without one absorbing the other.

This double rating produces two artefacts. The first is a matrix of source evaluation that renders visible, at a glance, the composition of the corpus in terms of reliability. The second is a cartography of dependencies between sources, which answers an essential and often neglected question: two different articles citing the same primary source do not constitute two independent confirmations. They constitute a single confirmation, duplicated. The cartography identifies these redundancies and neutralises them methodologically.

This operation is more counter-intuitive than it appears. The analyst who sees an assertion repeated by five sources is spontaneously inclined to consider it well established. But if these five sources rest on the same initial article, which itself rested on a single declaration, the apparent robustness collapses as soon as the references are unravelled. STRATUM imposes that systematic unravelling. An assertion is never validated by the number of sources carrying it, but by the genuine diversity of the chains of transmission that document it.

The pitfall thus avoided is known, in the trade, as the confirmation bias by accumulation. It consists in multiplying the sources that say the same thing in the mistaken belief that this strengthens the conclusion. Yet the repetition of an assertion, even by multiple channels, does not increase its probability of being true; it only increases the probability that the analyst will end up believing it. The distinction is crucial.

What competing hypotheses do the facts permit, and which one best resists falsification?

The methodological heart of STRATUM lies here. Once the corpus is filtered and rated, the analysis explicitly formulates the competing hypotheses that can account for the available facts. All plausible hypotheses, including those the analyst or principal would not wish to confirm. This requirement is the only protection against the natural slope of the mind, which consists in formulating a hypothesis, finding it elegant, and then reading the rest of the facts through it.

Each hypothesis is then confronted with the entire corpus in an analysis matrix whose rows are the hypotheses, columns the key facts, and cells the assessment of consistency between each fact and each hypothesis. The hypothesis retained is not the most appealing, nor the one that matches the analyst's initial intuition, nor the one that will most please the principal. It is the one that best resists the test of facts: the one for which the fewest cells present inconsistencies, and for which the residual inconsistencies are the most minor.

This discipline is costly in time. It is also what distinguishes a strategic analysis from a structured opinion. Most analyses published in the financial press or by consulting firms never explicitly formulate competing hypotheses: they present a single reading, illustrating it with the facts that support it. This presentation has the advantage of elegance and narrative fluidity. It has the major disadvantage of masking the fragilities of the conclusion. STRATUM prefers the explicit trace of the path, even at the cost of a lesser literary virtuosity.

The analysis also produces an explicit formulation of the level of confidence retained for the principal conclusion. High, moderate, or low. An analyst who cannot formulate his level of confidence has not finished his work. An analyst who systematically announces a high level of confidence is not a good analyst — he is a militant.

The pitfall avoided is what is sometimes called narrative drift: progressively allowing an appealing hypothesis to orient the reading of the facts, until the facts themselves seem to confirm a thesis that none has truly established. The matrix of competing hypotheses, because it forces the explicit writing of inconsistencies, is the structural safeguard against that drift.

How to render the analysis in a format that the decision-maker can absorb in a few minutes?

Synthesis is the phase in which the analysis, which could fill dozens of pages, is reduced to a deliverable that its recipient can read in a few minutes without losing anything essential. This compression is among the most difficult exercises of the discipline. It supposes a constant renunciation: of all the nuances, all the details, all the secondary angles that the analysis has explored and that nonetheless deserved exploring. A good synthesiser knows the value of what he chooses not to write.

The STRATUM format imposes a structure. The principal conclusion is placed at the head, formulated in one or two sentences that can be read in isolation. There follow the key facts on which the conclusion rests. Then the ranking of the competing hypotheses examined, with mention of those that have been discarded and why. Then the level of confidence retained. Finally, when the mandate justifies it, a section of possible recommendations, clearly distinguished from the analysis itself. The final note never exceeds a few pages. It is accompanied by a technical annex containing the complete chain of sourcing, source by source, accessible to verification.

Before distribution, every deliverable passes through a review by a second analyst. This analyst has not participated in the drafting of the note. His function is neither to embellish it nor to validate it through collegial sympathy, but to look explicitly for what does not hold. An assertion insufficiently sourced. A competing hypothesis insufficiently treated. A conclusion stronger than what the facts permit. The second reviewer produces a written review note, archived with the deliverable, whose remarks must have received an answer before the note leaves the firm.

The pitfall of this phase is what is called defensive complexity: multiplying angles, nuances, conditions, not to better serve the analysis but to avoid having to conclude clearly. A note that presents five possible readings without privileging any is not a prudent analysis. It is an analysis that evades. The principal who pays for a STRATUM note buys, among other things, the analytical courage to conclude despite uncertainty, while assuming the level of confidence the facts allow. No more, no less.

How to deliver the analysis so that it actually serves the decision?

An analysis completed is not an analysis delivered. The distribution phase organises the delivery of the note in the channel and format agreed with the principal at the moment of framing. According to the case, this means an encrypted dispatch, a hand delivery, a confidential oral briefing, or a commented reading in the presence of recipients. The channel is never left to chance: it is integral to the confidentiality we guarantee.

Distribution also documents the exact list of recipients, archived with the deliverable. This traceability protects both the firm, in the event of subsequent leakage, and the principal, in case of question over the document's internal circulation.

But the phase does not stop at delivery. STRATUM imposes a post-distribution follow-up that distinguishes a living analysis from a dead report. Three weeks after delivery, the analyst makes contact again to verify that the note has been useful, gather the questions its reading has prompted, and identify the points that would deserve an update. If new elements modify the analysis, a dated amendment is produced, which does not rewrite the initial deliverable but consigns its explicit update. The journal of amendments is integral to the mandate file.

This final phase expresses a conviction underlying the whole method: an analysis note is not a definitive object. It is a living instrument, which must be able to accompany the decision it illuminates in the weeks and months following its drafting. The monument-report, delivered once and then forgotten, is an anti-model. The good deliverable is the one the decision-maker can take up six months later and which either still says something accurate, or has been honestly amended to take account of what has changed.