Compliance is proved, not asserted

There is a moment in the life of every weapons programme when somebody has to prove something; claiming it, modelling it, adding items to slides with a green tick are not the same. They must prove it, to a State that is deciding whether to adopt the system, with evidence that will stand behind that decision for the life of the capability.

In the first two pieces of this series I argued that legal input belongs at the design table, and that autonomous and AI-enabled systems have turned the legal review from a single event into a continuing relationship. This piece is about the part of the work that decides whether a review succeeds: the evidence, and specifically the trials and tests that generate that evidence.

A legal review is only as good as what it rests on. The reviewer must answer a set of questions about the weapon in its normal or intended circumstances of use: whether any rule of law prohibits it; whether it is of a nature to cause superfluous injury or unnecessary suffering; whether it is by nature indiscriminate, or whether it can be used discriminately; what its effects on the natural environment may be expected to be. Answering them is a matter of judgement, informed as much by an understanding of weapons, and how they are used as by the law, and judgement of that kind can only be as good as the evidence in front of it.

Here is the problem. Trials are designed by engineers to answer engineering questions. Does it work? Does it hit what it is aimed at? Does it survive the environment it will operate in? Those are necessary questions, and an expensive trials campaign will answer them thoroughly. But the legal questions are often different, or at least considered through a different lens, and a trial that was not designed to answer them usually does not, however much it cost. Accuracy data tells you the system hits the target; it does not, by itself, tell you how the system behaves at the edge of its engagement envelope, what its effects are beyond the intended point of impact, or how it degrades when the conditions are not the ones in the brochure. Those are the places the legal questions live, because that is where discrimination and control of effects are actually tested.

Take accuracy, which sounds like the most settled question in any trials campaign. One firing is a data point, not a conclusion; a reviewer needs enough of a distribution to know how the system behaves at the edges rather than at the centre. In cluttered surroundings the question stops being whether the system strikes what it was aimed at and becomes whether it can be relied on not to strike something else. Programmes that have not tested for that often find the gap in their own paperwork, in a caveat written by an engineer who was being honest about the limits of what had been shown. Found early, that caveat is an additional test point. Found late, it is a restriction on where, or whether, the capability can be used at all.

The consequence is one I have seen from both sides: a programme completes its trials, the review begins, and the reviewer asks for evidence the trials never captured. Now the choice is grim. Re-run trials that cost hundreds of thousands of pounds a time, or ask the reviewer to proceed on assertion rather than evidence, which is another way of asking the State to carry your risk. Neither is good, and both were avoidable.

The fix is straightforward and cheap by comparison: put the legal questions into the trial design before the trial is run. In practice that means a piece of work, early, that translates the review criteria into an evidence plan. What must be demonstrated, in what conditions, measured how, recorded in what form. Often the additional burden on the trial itself is modest; the same serials, instrumented slightly differently, or a handful of additional test points at the edges of the envelope, can carry the legal evidence alongside the engineering evidence. The cost of designing it in is a fraction of the cost of going back.

There is a practical version of this that costs very little. We start with the system requirements document (SRD). An SRD tells me a good deal about what a programme has committed to demonstrate, and rather more about what it has not. The legal questions are rarely written into it as requirements, but they must still be answered, and the obligation does not go away just because nobody captured it. Reading an SRD for what it does not say and setting out what will nonetheless need to be proved, is often a short piece of work that can focus a trials campaign.

Two further elements make the evidence work harder. The first is witnessing. Evidence carries more weight when the right people saw it gathered; there is a difference between a company reporting its own results and a demonstration observed by those with standing to speak to what happened, whether that is the customer’s own representatives or an independent presence. Which observers, at which trials, for which purposes, is itself a question worth thinking about in advance.

The second is the form of the record. A State conducting its review does not want a terabyte of telemetry and an invitation to draw its own conclusions. It wants a structured dossier: the legal questions, the evidence relevant to each, the conditions under which it was gathered, the chain from raw data to conclusion. Assembling that dossier deliberately, as the trials campaign proceeds rather than archaeologically afterwards, is the difference between a review that moves and a review that stalls. And for the systems I wrote about last time, the ones that change through life, the dossier is not a document at all in the ordinary sense. It is a living pack that grows with each update, carrying its own evidence with it, in the same structure, so that re-review is fast and efficient.

None of this is exotic. It is the same habit of mind that good test engineers already apply to safety cases, applied to a different set of questions. The only thing that is unusual is doing it early.

There is a wider point, and it presses harder this year than last. The United Kingdom needs to develop, procure and field capability at the pace of relevance, which means ready for the conflict that is already on the horizon rather than the last one. That pressure is real and it will not ease. It would be easy to treat evidence as one of the things that has to give in order to move faster, and that would be a serious error. A capability that cannot be reviewed cannot be adopted, and a system waiting at the final gate is not fast, whatever the programme plan says. The way to go quicker is to stop leaving the evidence until the end. Defence companies have a part to play in that, and in my view a commercial interest in playing it: the supplier who arrives with the proof already assembled is the supplier whose capability can be fielded in time to matter.

If your programme has trials ahead of it, the cheapest legal advice you will ever buy is the advice that shapes what those trials measure. After they have run, everything takes longer and costs more: re-running serials at full price, re-opening a campaign you had closed, explaining to a customer why the dossier does not answer the question they have just asked. If you would rather know now what your trials will need to prove, that is the conversation to have with me while the trial plan is still a draft.

This is the third in a short series for defence manufacturers, technology firms and those who bring military capability to market: what the law asks of a system, and how answering those questions early strips risk, cost and delay out of a programme. The first two made the case for legal input at the design table, and for treating the review as a continuing relationship rather than a single event. Next: why the legal exposure does not stop with the State that fields the system.