Unusable intelligence
A Socratic walk-through of unusable intelligence — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why would a commander deliberately not act on information they know to be correct?
It sounds like a contradiction in terms. Intelligence is gathered in order to be used; if a commander has a true report about the enemy and does nothing with it, we assume timidity, or stupidity, or a bureaucracy that lost the paper. But suppose the commander is competent and the report really is true. Then acting on it must carry a cost that its truth does not cancel. What could that cost be — and is there more than one kind?
Reasoning it through
REASONING #Begin by disputing the premise, gently. What does it mean to know a report is correct? Beforehand, a commander does not hold one true report. He holds a very large number of reports, most of them wrong, many contradicting each other, none stamped as the important one. Roberta Wohlstetter's study of Pearl Harbor made this the central point: the warnings were genuinely there in the record, and so were vastly more signals pointing elsewhere. Hindsight supplies the marking that the moment did not.
That has a consequence worth drawing out. The filter separating warning from noise has a threshold, and the threshold trades two kinds of error against each other. Lower it and you catch more real warnings — and far more false ones, since there is so much more noise than signal. Each false alarm costs money and readiness, and something less tangible: a system that has cried wolf several times finds the next warning met with a shrug. Collecting more does not fix this; it usually makes it worse, because collection increases noise faster than signal.
Now suppose the report does stand out and is believed. Is it usable? Only if it converts into an action worth its price. "An attack is coming" with no place and no date cannot be turned into a useful order. Dispersing a fleet, mobilising reserves, evacuating a city — these are expensive, cannot be sustained indefinitely, and in a tense standoff are themselves read as preparations for attack. So a commander may believe a warning entirely and still find that no available response beats waiting.
Then the reason that answers the question most directly: acting can destroy the source. Suppose the information came from reading the enemy's ciphered traffic. Intercept a convoy at a position that nothing but a decrypt could have supplied, and the enemy learns something more valuable than the convoy — that his cipher may be broken. He changes it, and you have traded one convoy for every message you would have read for the rest of the war. The calculation is not "is this true" but "is this worth more than everything the source will yield later".
Put that way, the resolution is usually not inaction but concealment. You arrange an innocent explanation: a reconnaissance aircraft is sent to be seen finding the convoy before it is attacked. British handling of decrypted German traffic in the Second World War worked on rules of this kind — material could be acted on when a plausible alternative source could be manufactured for it. Worth flagging that the most famous illustration of the harsher choice, the story that Churchill let Coventry be bombed to protect the secret, is rejected by most historians: the target was not identified in time. The practice of cover stories is well documented; the emblematic anecdote is not.
One more, and it is the least comfortable. Intelligence arrives into a decision process that often has a policy already. Reports supporting the chosen course are forwarded and amplified; reports contradicting it are questioned, sent back for corroboration, or discounted as thin. Nothing here requires anyone to lie. An analyst who has watched an unwelcome assessment sink learns what is worth writing next time, and that lesson propagates upward as silence.
The analogy
THE ANALOGY #Think of a police unit with an informant inside a burglary ring, who tells them exactly which house will be broken into on Friday. Preventing it is trivially easy — and instantly reveals that the ring has a leak, at which point everyone who knew the plan becomes a suspect and the informant is finished. So the choice is not between a good and a bad outcome but between one house and everything the informant would have told them for the next year, unless someone can invent a reason for a patrol car to be on that street anyway.
The informant's tip arrives already marked as important, whereas the commander's hardest problem is the one this picture cannot show — a warning indistinguishable from a thousand others until it is too late to matter.
Clarifying the model
THE MODEL #Four different things have been collapsed under one phrase, and they call for different remedies. Noise is an analysis and threshold problem, which more collection aggravates. Non-actionability is a problem on the response side. Source protection is a genuine exchange rate, best settled in advance by rules about what may be acted on and under what cover, rather than case by case under pressure. Politicisation is structural, and the usual answer is institutional distance between those who assess and those who advocate.
Two honest qualifications. Compartmentation, which exists to protect sources, separates the person holding a fragment from the person holding the context that would make it meaningful — so secrecy that preserves a source can disable the analysis it was meant to feed. And our view of the subject is warped by selection: we study cases with famous outcomes, and in any such case the warning will be found in the record, because we searched the record knowing what happened. Cases where a threshold correctly ignored a true report are not written up at all.
A picture of it
THE PICTURE #How to readEnter at the parallelogram, the raw stream, and read downward through four gates — each a separate way a true report fails to become an action. The first is noise, and its "no" branch runs to the cylinder: material correct all along and legible only afterwards. The second is specificity, and its "no" branch loops back up, because a raised alert followed by nothing quietly raises the bar for the next warning. The third is the sourcing trade, where the escape is a cover story rather than the action itself. The fourth is the one nobody puts in the manual: whether the report agrees with a decision already made. Only one path reaches the green terminal.
What became clearer
WHAT CLEARED #"Deliberately not acting on correct information" turns out to be four distinct problems wearing one description. Most of the time the information was not identifiable as correct in advance, because signal is buried in noise and hindsight does the sorting. Sometimes it was recognised and could not be converted into an affordable action. Sometimes acting would have burned a source worth more than the target — a real trade, usually resolved by manufacturing an innocent explanation rather than by doing nothing. And sometimes it was correct, recognised and actionable, and lost to a decision already taken. The first three are failures of the world. The fourth is not.
Where to go next
ONWARD #- Why the same evidence supports opposite conclusions in two competent commanders, taken up in the walk-through on the fog of war.
- How signal-detection thresholds behave when the base rate of the event is very low, which is the arithmetic behind the cry-wolf effect.
Key terms
TERMS #| Term | What it means |
|---|---|
| Signal-to-noise problem | the difficulty that true warnings arrive mixed into a much larger volume of irrelevant or contradictory reports, and are unmarked until afterwards. |
| Source protection | withholding or disguising action on intelligence so that the enemy does not learn how the information was obtained. |
| Cover story | a manufactured alternative explanation for how a target was found, allowing sensitive intelligence to be acted on safely. |
| Politicisation of intelligence | distortion of assessment by the policy preferences of its recipients, usually through selective attention rather than falsification. |
Every term the collection defines is gathered in the glossary.