Every figure XENA publishes carries a grade, every probability has a stated basis, and our own predictions are scored against what actually happened. This page explains all three, because intelligence you cannot interrogate is just opinion with formatting.
Market forecasting is conventionally done top-down: take a national defence budget or a share of GDP, apply a growth rate, and apportion the result across categories. It produces clean numbers quickly, and it cannot be interrogated. Ask which programmes make up the figure and there is no answer, because the figure was never made of programmes.
XENA works the other way. The foundation is individual, sourced procurement programmes — currently 1,150 of them across 90 countries — each with its own value, quantity, timeline, supplier set and probability. Market sizes, regional totals and forward projections are aggregations of those rows. Any number we publish can be opened up and read back down to the programmes that produced it, each with its source URL.
This also avoids a specific analytical trap. A headline budget rise does not translate proportionally into platform procurement: money moves into personnel, readiness, infrastructure and munitions through funding routes that are opaque from outside. NATO members committing to 3.5% of GDP does not mean 3.5% more aircraft. Modelling procurement as a fixed share of a rising budget produces confident numbers that are wrong.
What this costs us, stated plainly. Bottom-up data is strongest in the two-to-ten year window, because that is the horizon on which governments actually plan and publish. Beyond ten years our entries are forecasts built on triggers rather than announced procurements, and they are labelled as such. A provider quoting you a confident 2045 market size is not working from evidence.
Every value, quantity and unit cost in XENA carries one of four grades. The grade travels with the number everywhere it appears — in the pipeline, in company profiles, in Insight reports and in exports.
| Grade | What it means |
|---|---|
| Actual | A signed contract, a statutory filing, or an official document stating the figure. The strongest grade — a legally accountable source published it. |
| Reported | An official announcement that is not yet a contract: a budget request line, a ministry press release, a US DSCA notification ceiling, a framework-agreement value. Real, dated, citable — but the figure can still change before signature. |
| Estimated | A credible third-party estimate: an audit-office projection, a programme-office lifecycle estimate, a well-evidenced analyst range. We name the estimator. |
| Modelled | XENA's own arithmetic — a unit-cost multiplication, a currency conversion of a historic figure, an inference from comparable programmes. We show the working in the value note. |
Two rules govern the grading. A figure is never graded up to make a row look more authoritative — a ministry announcement stays Reported even when we believe it. And where no defensible figure exists, the field is left empty rather than filled: Gulf states frequently do not publish contract values, and an honest gap outlasts a plausible invention.
Each of the 1,150 programmes in the procurement pipeline carries a probability of go-ahead: the likelihood the programme reaches a signed contract. This number is not sentiment. It is derived from a structural model of how each country actually decides to buy defence equipment.
PGO is banded on the NATO Probability Yardstick, so a number means the same thing every time it appears.
| PGO | Band | What it means |
|---|---|---|
| 0% | Remote | Cancelled, or dormant so long it is effectively dead. We keep it — a cancellation is intelligence. |
| 1–24% | Highly unlikely | Superseded, unfunded, or the requirement has changed without formal cancellation. |
| 25–49% | Unlikely | Stalled or at risk. Named in official documents but with little or no progress since. Options on programmes that have overrun. |
| 50–74% | Likely | May progress, conditions uncertain. Publicly discussed then gone quiet, or a real requirement that remains unfunded. |
| 75–94% | Highly likely | The typical band. Clear requirement or official statement of intent, budget allocated or known available, political support present. |
| 95%+ | Almost certain | Economic, political and requirement backing, in a country with a record of completing procurements. |
A contract award does not automatically mean 95%. The UK's Ajax programme was contracted and in serious jeopardy for years, and a database that had marked it almost certain throughout would have been wrong for the whole period. Award raises the floor; it does not end the assessment.
Every PGO carries a written rationale naming the factors behind it — military need and its priority against competing requirements, availability of finance, the economic condition of the purchasing country, programme history, political support and election risk, the country's demonstrated ability to procure effectively, signalled commitment through white papers and procurement plans, and the manufacturer's production capacity and delivery record. A probability with no stated reasoning cannot be challenged by a subscriber, reviewed by us, or learned from when it turns out wrong.
For 26 countries (and the US Foreign Military Sales process as a system in its own right) XENA maintains a gate model: the sequence of approvals a programme must clear — requirement validation, budget committee, parliamentary vote, contract negotiation — with, for each gate, whether passing it is publicly observable, how long the next stage typically takes, and what share of programmes historically survive it. A programme's PGO reflects the last gate it has demonstrably cleared, adjusted for programme-specific risk: financing constraints, export-control dependencies, political calendars.
Why this matters in practice. A German procurement reaching the Bundestag budget committee's €25m list converts to contract roughly 90% of the time — a strong, evidenced signal. A signed Polish framework agreement looks similar in a headline but is materially weaker, because Poland's binding constraint is financing capacity, not approval. An Indian Acceptance of Necessity is routinely reported as an order; the evidence supports roughly 55–65% eventual conversion, over three to six years. The same press release deserves three different probabilities in three different countries — and gets them.
Where a country's process is genuinely opaque, we model observed behaviour instead, grade the country's transparency honestly, and say plainly what cannot be known from open sources.
A programme that has not been formally announced only enters the database if we can name the specific evidence that prompted it — an out-of-service date creating a replacement need, a minister or service chief stating a requirement and a date, a published procurement plan, a testing programme, a budget line, a request for information. That trigger is recorded on the entry.
"This country will probably need these eventually" is not a trigger, and a programme resting on it is not entered. The discipline is what separates a forecast from a guess, and it is why the probability attached to a forecast programme can be defended rather than merely asserted.
Programme value, unit cost and quantity have to reconcile, and when they do not we explain the gap rather than leaving you to find it. The usual reason is support: through-life support, procurement overhead and, on Foreign Military Sales, the FMS charge. Where a total is known but its split is not, we assume 20% support and 80% procurement and say so on the entry, deriving unit cost from the procurement portion.
Unit cost is quoted flyaway for aircraft, helicopters, UAS and land vehicles — power and core mission systems only. Naval vessels are the exception and are quoted gross, including all subsystems and hull modifications, because a warship's combat system is not meaningfully separable from the hull it is built into. Uncrewed surface and underwater vessels use flyaway.
Value, quantity and unit cost each carry their own confidence grade. A programme can legitimately have an Actual value from a signed contract, an Estimated quantity because the split across variants was never published, and a Modelled unit cost derived from the two. Collapsing that into one grade would hide exactly the thing a buyer needs to see.
A programme value is a lump sum; a market model needs to know which years the money lands in. We hold an annual expenditure profile per programme, using actual contract periods and delivery schedules where they exist and benchmarking against comparable historic programmes where they do not.
Spending starts at contract award, not first delivery. Major defence contracts carry long lead times and governments pay progressively — progress payments against costs incurred, or performance payments against milestones. In the US these are Contract Financing Payments, with suppliers eligible for up to 90% of costs incurred. Some governments pay set-up costs upfront. A model that books the whole value in the delivery year will put the money in the wrong decade.
Short-run, COTS and military-off-the-shelf procurements are pro-rated across the contract period. Bespoke and long-run production is ramped, reflecting low-rate initial production before full rate — and where full-rate production is expected to bring unit cost down, as it did on the F-35, that is modelled rather than ignored. Each year's figure carries its own basis and confidence, and a modelled profile is labelled as modelled.
The customer country appears in square brackets at the end of every programme title; batches and increments appear in round brackets. Non-English programmes are titled in English with the original acronym retained, so Frégates de Défense et d'Intervention appears as Defence and Intervention Frigate (FDI) and remains findable under either.
Where parts of one procurement have genuinely different statuses — most often a firm order and an option on it — they are split into separate entries and linked, because an awarded contract and an unexercised option do not share a probability and averaging them serves nobody. The link is recorded so the split does not read as double-counting.
On military aid, the customer is the country paying. Germany buying UAS for Ukraine is a German programme with Ukraine recorded as the end user, not a Ukrainian one. Recording it the other way would inflate the recipient's apparent procurement spending and understate the donor's, which is the opposite of useful when the question is who is actually in the market.
Out-of-service dates are the clearest forward indicator of a replacement opportunity, which is what most of our subscribers are actually looking for. We record the earliest OSD for the main in-service variant, starting from designed service life and adjusting for whether the equipment was bought new or second hand, whether it has been upgraded or life-extended, annual flying or running hours against design assumptions, funding availability, political appetite, and whether adversary capability has moved. It is a judgement, not a specification, and it is presented as one.
Inventory counting follows fixed rules so that fleet numbers stay comparable. Where a UAS comprises several air vehicles we count air vehicles. For surface-to-air and rocket artillery systems we count launchers, holding the missiles as a separate linked entry. Upgrades never double-count — when aircraft move to a new standard with its own entry, the old entry's inventory falls by that number while its orders and deliveries stay put, and the new entry gains inventory with zero orders. Second-hand transfers move inventory only; the orders and deliveries stay with whoever originally bought them. Where a country is known to operate a type but the number cannot be established, we include it and mark the figure unverified — an honest unverified entry is worth more than an invented number and more than a silent omission.
XENA's forward claims — analysis predictions, stock-impact forecasts, and the PGO numbers themselves — are resolved against real outcomes on a weekly cycle. A prediction is marked correct, wrong, partially correct or lapsed; a prediction too vague to score is recorded as exactly that, which is itself a finding about how we write. The scoring is deliberately harsher than the predicting: a +0.4% drift against a predicted +5–8% move is wrong, not a near miss.
Resolved outcomes feed a calibration record — hit rate against stated confidence, broken down by country and by claim type — and that record adjusts the gate model and the forecast agents that produced the claims. When the resolved sample is large enough to be meaningful, we will publish the scorecard, including the misses. We are not aware of another provider in this market that scores its own record; we think that is precisely why it is worth doing.
XENA monitors 3,000+ sources: the newsroom, press and investor-relations estate of every tracked company on its own publication rhythm, plus a named registry of ministry and parliamentary channels, procurement portals, trade press, financial data and multilateral publications — weighted deliberately toward Arabic, Turkish, French and Hebrew sources that rarely reach English-language coverage. Every item is resolved against a single entity spine of companies, programmes and countries, so the same fact links the news item, the pipeline row, the company profile and the capability entry it concerns. The live source count and category breakdown are on the About page.
The established providers in this market run analyst desks split by domain, in-house journalist teams, networks of regional contributors, and attendance at twenty or more trade shows a year. XENA does not. It is a continuously running intelligence pipeline: agents that read the sources on a fixed daily cadence, extract structured data against the rules on this page, and file it for human review before anything carrying judgement is published.
That is a real difference and we would rather state it than imply a newsroom we do not have. What it costs is the corridor conversation at a trade show and the contributor who knows a programme office personally. Some things only surface that way, and where a competitor has that access on a given programme, they may well be ahead of us on it.
What it buys is cadence, breadth and consistency. The pipeline runs every day rather than when an analyst gets to that desk; it reads Arabic, Hebrew, Turkish, Persian and French sources alongside English ones as a matter of routine rather than as a special project; and it applies the same grading rules to every row, because the rules are enforced in code rather than remembered. Entries carry the date they were last reviewed, and the oldest are worked first, so nothing sits unexamined indefinitely.
Judgement is where the human sits. Reported fact is published as it is found; anything containing XENA's own inference goes to a review queue first. That line does not move regardless of how confident the inference reads.
We do not publish figures we cannot source, we do not launder single-source claims into fact, we do not treat a belligerent's announcement as a capability, and we do not delete cancelled programmes — a cancellation is intelligence, not an embarrassment. Where XENA runs model portfolios to test its own intelligence, they are nominal paper simulations; nothing on this platform is investment advice.
Questions about methodology: contact the team. Definitions of defence terms used across the platform are marked inline and available on any Insight report.