Checking the working¶

Two practices for not being wrong in the most expensive way. Neither is comfortable to maintain. Both run against the institutional grain.

Convenient conclusions¶

When analysis exactly confirms what the decision-maker already wants, that is a signal to re-check the working, not a reason to be pleased.

Convenient conclusions are sometimes correct. The problem is not that they are likely to be wrong; it is that the conditions for undetected error are at their strongest when nobody has an incentive to look. The structural reward for convenient conclusions can be high: they support the existing mandate position, they avoid disagreement among principals, they allow institutional leadership to proceed. The incentive for nobody to push back is built into the institution.

Structural dynamics¶

A mandate creates a position. An institution needs analysis that is consistent with that position, or at least does not contradict it. An advisor who provides confirming analysis is being helpful in the institutional sense of helpful. Nobody fabricated anything. The analysis was conducted by someone who believed the conclusion.

But the framing of the question, the selection of evidence, the confidence tier assigned: all of these can bend toward the institutional conclusion, and the bending tends to happen below the level at which it would be noticed as dishonest. The question was posed in a way that made the confirming answer probable. The evidence provided was the evidence that supported the position. The timeline allowed for a conclusion but not for scrutiny of counterevidence.

The SEM question applies here directly: which assumptions in this analysis, if wrong, would reverse the conclusion. A convenient conclusion is particularly likely to have been built on assumptions that nobody tested because nobody had the incentive to.

Re-checking¶

Which assumptions are load-bearing, the ones that, if wrong, would change the conclusion materially. Whether the evidence supports the confidence tier, or whether the tier was assigned because the conclusion felt secure. Whether the framing of the question excluded evidence that would have been relevant. Whether the counterevidence was examined with the same scrutiny as the supporting evidence.

This re-check is the work that nobody is paying for and that the institutional structure is not designed to encourage. It belongs in the advisor’s own process, applied to their own analysis, before the brief goes out.

Steel-manning a disliked option¶

Not an exercise in open-mindedness. An exercise in not being blindsided.

Some options are structurally suppressed. The institution has a position. Principals have agreed on a direction. Options that imply that position is wrong, or that the mandate itself is the wrong instrument for the situation at hand, are uncomfortable to voice and are not what the institution is paying the advisor to produce. They tend not to appear in briefing papers unless someone puts them there deliberately.

An advisor who has not built the reasoning for the disliked option under their own roof will not be able to construct it under pressure. When someone in the room picks it, or when events make it the only one available, the reasoning needs to already exist. Reasoning done at your desk, in advance, is cheaper than reasoning done in the room with everyone watching and no time to do it properly.

Steel-manning is not advocacy¶

The exercise does not require recommending the disliked option. It does not imply the disliked option is probably right. It is the construction of the strongest possible version of the argument for it, conducted honestly and in advance, so that the advisor understands the option on its own terms rather than on its weakest caricature.

An option that cannot survive its own steel-man is a weak option. One that survives it and is still disliked can be rejected for reasons that have been examined, not just asserted.

Composing both¶

Checking for convenient conclusions identifies the analysis most likely to be wrong. Steel-manning the disliked option builds the reasoning for the alternative most likely to be uncomfortable. They address opposite failure modes: confirmation of the expected conclusion, and inability to engage with the unexpected one.

Both run against the institutional grain. The convenient conclusion is what the institution wants. The disliked option is what the institution has suppressed. Maintaining both practices requires running against the direction of the institutional reward structure. That is what makes them worth treating as disciplines rather than as occasional intellectual acts.

Programme assessment¶

A mandate-driven institution has developed a programme to address a specific security threat. The programme has been running for three years. A renewal decision is approaching. The advisor is asked to assess the programme’s effectiveness.

The assessment comes out positive. The evidence supports continuation. Decision-makers are pleased.

A convenient conclusion check: which assumptions, if wrong, would reverse the finding.

The assessment rests on reported outcomes at face value. The programme measures success by indicators set at inception, which have not been revised as the threat evolved. The threat landscape has shifted in ways that are visible in other parts of the institution’s work but have not been incorporated into this assessment. The counterfactual, what would have occurred without the programme, is assumed rather than argued.

Examining these: the success indicators were chosen to demonstrate activity rather than impact. The threat has evolved beyond the programme’s original framing. Outcomes attributed to the programme include effects that may have other causes. None of this appeared in the assessment because none of it was examined.

A steel-man of the disliked option, constructed properly: the programme has not produced the outcomes attributed to it and the renewal case rests on indicators that do not measure what they claim to measure. The strongest version of this argument runs as follows. The institution measures programme success by output metrics: advisories issued, partner organisations engaged, incidents reported. None of these measure the underlying objective, which is reduced vulnerability across the sector. The threat actor has adapted to the programme’s methods, which are documented in the institution’s own published reports and therefore no longer provide the intelligence value they did in the first year. Resources committed to the programme would produce better outcomes redirected to real-time sharing infrastructure, which the programme was originally intended to enable but has not, because the programme absorbed the resources.

That argument exists. It has not been examined in the assessment. An advisor who has not constructed it will not be able to respond when it is raised; and if events later show the programme to have been ineffective, the assessment that recommended renewal will be part of the record.

The finding may still be positive. A re-checked assessment that has examined the counterevidence and still recommends renewal is a more durable recommendation than one that has not.