We gave 3Dogs four of the most-taught decisions in military history — with none of the hindsight.

By Alan Finney — Founder, 3Dogs Nexus

Staff rides and war colleges have studied these decisions for decades: the Cuban Missile Crisis, the Inchon landing, Pickett's Charge at Gettysburg, and Task Force Smith in Korea. Each one is fed to 3Dogs using only the facts the actual commander had at the moment of decision — no outcome, no hindsight. Then we compare its call against what history's own doctrine teaches and what the commander actually chose.

How does AI decision analysis perform on decisions where we already know the outcome?

It is the cleanest way to test a decision process: strip the hindsight, give the panel only what the commander knew at the time, and compare its call against both the historical decision and the consensus of military scholarship. Four famous command decisions, no outcome leakage in the prompts.

The scoreboard

Same test every time: give 3Dogs the situation as it stood at the moment of decision, let its panel debate independently, then check the call against what actually happened and how the war colleges assess it today.

The Cuban Missile Crisis — blockade, airstrike, or invasion?

October 16, 1962: U-2 photography confirms Soviet medium-range missile sites under construction in Cuba, one to two weeks from operational. The Joint Chiefs unanimously want airstrikes followed by invasion. A blockade is slower and doesn't remove the missiles already there, but preserves room to negotiate and de-escalate.

Kennedy ordered a naval "quarantine" on October 22 — deliberately avoiding the word "blockade" for legal reasons — while secretly negotiating a Turkey-missile trade through back channels. The crisis resolved without war.

The canonical case in political-military decision-making curricula (Graham Allison's Essence of Decision). ExComm itself ran a proto-adversarial process — assigning members to argue for options, then cross-examining them — and the lesson taught ever since is that preserving off-ramps and avoiding irreversible action under nuclear uncertainty was the right call.

Required modifications, generated with no knowledge of the actual outcome: declare it a legal "quarantine" not a blockade; keep a Turkey-style trade confidential while running a visible diplomatic track in parallel; define explicit rules of engagement to prevent a local miscalculation at the quarantine line escalating to nuclear use.

The Inchon Landing — bold strike or the safer option?

Late July 1950: UN forces are pinned into the Pusan Perimeter. The theater commander proposes landing far behind enemy lines at Inchon — a port with a 30-foot tidal range, a landing window of hours per month, and no room for error. The Joint Chiefs' own inspectors prefer a safer site near Kunsan instead.

MacArthur personally overruled the Army and Navy's preference for Kunsan, ordered a preliminary airstrike on Wolmi-do Island the day before the main landing, and executed the assault on September 15, 1950. It broke the North Korean position within days.

Widely studied as a paragon of bold operational art and calculated risk — "brilliant generalship" in service literature — though later scholarship also treats it as a cautionary tale: the success bred an overconfidence that contributed to being caught off guard by Chinese intervention months later.

Required modifications, again with no knowledge of the real invasion plan: a pre-landing airstrike on Wolmi-do Island to neutralize the artillery overlooking the causeway — the same target actually struck the day before the real landing — plus a hard go/no-go gate tied to securing Kimpo Airfield, which was in fact a key first-days objective.

Pickett's Charge — attack the center, or maneuver?

July 2 evening, 1863: two days of attacks on both Union flanks at Gettysburg have failed to break the line. Longstreet, the army's senior corps commander, argues for disengaging and maneuvering around the Union left instead. Lee believes the Union center may be weakened and that one decisive, massed assault could break it.

Lee overruled Longstreet and ordered the assault. On July 3, roughly 12,500 men crossed nearly a mile of open ground under fire. The charge failed with casualties near 50-60%, and it is widely regarded as the turning point of the battle and the war.

The U.S. Army Center of Military History's own Gettysburg staff ride treats this as the textbook case of attacking a fortified position after two failed attempts, and the "one more push" overconfidence bias — Lee's own trusted subordinate's objection is the case's central teaching point.

This is the one case where 3Dogs' recommendation does not match what the commander actually chose. All three independently-composed panels reached REJECT on their own, converging on the same failure point: depleted artillery ammunition undercuts the entire premise that a bombardment can suppress the Union defense long enough for infantry to cross open ground.

Task Force Smith — commit the unready force, or wait?

Early July 1950: North Korean forces are advancing fast; the nearest available U.S. troops are under-strength, under-trained occupation units from Japan lacking effective anti-tank weapons. Waiting to properly prepare a force could take weeks — time the collapsing front may not have.

The task force was sent immediately as a delaying action. It was overrun on July 5, 1950; of roughly 440 soldiers, only 185 made it back to friendly lines. "No more Task Force Smiths" has been a standing readiness mantra in the Army ever since.

The canonical cautionary tale on hollow-force readiness, taught in professional military education whenever budget-driven readiness tradeoffs come up. The failure is attributed less to the decision to delay the enemy than to committing a force with no realistic chance and no clear withdrawal plan.

3Dogs matched the actual decision to commit — but attached exactly the safeguards history shows were missing: explicit withdrawal/abort triggers, designating the force as a screening element rather than the main effort, and a hard pre-commitment of budget and timeline for the real follow-on force before sending the first unit in. One panelist (Llama 4, 80% confidence) explicitly dissented against proceeding at all, citing "high likelihood of heavy casualties among the under-prepared force" — which is exactly what happened.

What this proves

Not that 3Dogs "gets history right" — it doesn't know the outcomes, and history is not a fair test set (hindsight bias runs both ways). What it does show: independently-composed panels reasoning from the same facts a real commander had, arriving at recommendations that line up with sound military judgment even where the panel didn't know what happened next — and in one case, disagreeing with the actual historical decision-maker while lining up with what military education has concluded about that decision for over a century.

Why this matters for high-stakes decisions

Every one of these cases came with genuine, named dissent on the record — not a single case ran to unanimous, uncontested approval. That's the point: a decision-support system that always agrees with itself isn't trustworthy on decisions like these. What you want is calibrated confidence, preserved disagreement, and named required modifications — exactly what a war college critique gives you, on demand, for your own decision.

Three times it agreed with the commander. Once it agreed with the professors instead.

That's the harder trick. Matching history is easy to fake with hindsight. Independently reproducing the war college's own century-old critique of a decision — using only the information available in the moment, with no idea how the battle actually ended — is a real signal about the quality of the reasoning underneath.

Try this on your own question.

Free, no card. Bring a real decision — ideally one where you already know the answer — and see what the panel does with it.

Questions this case answers

Why test on history rather than hypotheticals?

Because history has an answer key. A hypothetical can be argued forever; a historical decision has a documented outcome and decades of scholarship to score against.

Was hindsight leaked into the prompts?

No — the briefs were written to contain only what was knowable at the time of the decision.

Try this on your own question.

Free, no card. Bring a real decision — ideally one where you already know the answer — and see what the panel does with it.

People also ask

Questions this case answers directly

Plain answers on military decision-making under uncertainty and how war college case methods work. Every figure below comes from the delivered report for this case. These are clearly-labeled panel estimates from a multi-model adversarial analysis — not investment, legal or professional advice.

Real examples of military decisions under uncertainty

The instructive cases are the ones where the commander was genuinely missing information rather than merely wrong in hindsight — which is most of them. This study applies an adversarial multi-model panel to a historical military decision, reconstructing what was actually knowable at the moment of choice rather than judging it against outcomes nobody could see. The distinction is the whole point: a good decision with a bad outcome is still a good decision.

How do generals make decisions with incomplete information?

By reasoning about the distribution of possibilities rather than waiting for certainty that never arrives — assess the enemy's most likely and most dangerous courses of action, plan against both, and preserve the ability to change your mind as information arrives. Formal doctrine encodes this in the military decision-making process. The failure mode is the same everywhere: committing early, then reading incoming intelligence as confirmation.

What are historical counterfactuals in military history?

A counterfactual asks what would plausibly have followed from a different choice at a specific decision point. Done well it is a rigour exercise, disciplined by what was actually knowable at the time. Done badly it is fan fiction with maps. The discipline comes from constraining yourself to the information available at the moment of decision and stating explicitly what would have had to be true for the alternative to work.

What is the war college case method?

Officers are given a historical situation as the commander saw it — the same partial, contradictory information, without the outcome — and must decide. The learning sits in the discussion afterwards, where classmates attack the reasoning. That structure is the direct human analogue of what an adversarial model panel does: multiple independent judgments, forced argument between them, and the disagreement preserved rather than smoothed away.

Related decision case studies

• Should You Pay a Ransomware Demand?

  Two Las Vegas casino operators faced the same attacker in 2023. One paid, one refused. We put the decision to a 13-model panel and it s

• We Let ChatGPT Run a Decision Case and Review Us

  OpenAI's ChatGPT drove an entire 3Dogs Nexus engagement as the client, answered both clarification rounds, then wrote a review we publi

• What Does the Deepest Multi-Model AI Analysis Do?

  Our deepest analysis mode run on a real capital decision, with every step observable: two full ensembles, multiple research passes, and

• AI That Refuses to Fabricate: The Deloitte Re-Run

  Most AI invents something when the evidence is thin. We ran a well-documented public AI failure through a multi-model panel to see whet