On the morning of July 3, 1988, the US Navy cruiser Vincennes was in a gunfight with Iranian small boats in the Strait of Hormuz when its radar picked up an aircraft rising from Bandar Abbas, an airport shared by civilian and military planes.
A few minutes later, one of the cruiser’s guns fouled, and the ship swung with full rudder at 30 knots to bring the other to bear. At about the same time, an operator’s console showed a momentary identification signal of a kind Iranian F-14 fighters were known to use. The contact was entered as an F-14. Radio warnings went unanswered. At 20 miles the captain held fire. The aircraft was climbing through 10,000 feet.
Between 15 and 12 miles, the tactical information coordinator began reporting that the altitude was decreasing. The investigation put the time from the first report of descent to the decision to fire at 20 to 30 seconds. At 06:54:05 GMT, Captain Rogers gave the order to fire. Two missiles struck the aircraft at 13,500 feet.
It was Iran Air Flight 655, an Airbus on a scheduled flight to Dubai with some 290 people from six nations aboard. The ship’s own combat system had recorded it “continuously ascending.”
The loop on the whiteboard
The usual drawing has four boxes in a circle: observe, orient, decide, act. The usual lesson is a race: whoever runs the circle faster wins, because the slower side keeps answering a situation that has already changed.
John Boyd, the US Air Force fighter pilot behind the idea, did write in his briefing Patterns of Conflict that to win “we should operate at a faster tempo or rhythm than our adversaries,” or better, get inside the adversary’s observation, orientation, decision, action loop.
By that measure, Vincennes was not slow. The Central Command endorsement of the Navy’s investigation counted seven minutes and five seconds from first detection to the decision to fire. The trouble sat in the step the whiteboard treats as a box like the others.
The second O
Boyd spent his retirement on a set of briefings, A Discourse on Winning and Losing. The loop as a drawing came in a four-slide summary dated June 28, 1995. In that sketch, orientation is the big box in the middle. Into it feed genetic heritage, cultural traditions, previous experiences, new information, analysis and synthesis. Out of it run two arrows labeled “implicit guidance and control,” one back to observation and one straight to action.
In the 1987 briefings he had called orientation the “Schwerpunkt,” the focal point: “orientation shapes the way we observe, the way we decide, the way we act.” Grant Hammond, who edited the Air University Press edition of the briefings, calls the familiar diagram of four circular arrows one “which Boyd never created and which grossly oversimplifies the concept.”
Observe, orient, decide, act, faster than rivals. How you read things steers all four.
The Vincennes investigating officer asked Navy specialists in combat stress to explain the gap between the ship’s tapes and what crew members believed they saw. In their opinion, after the F-14 report, the coordinator “appears to have distorted data flow in an unconscious attempt to make available evidence fit a preconceived scenario.” They called it “scenario fulfillment.” That is orientation reaching back into observation, the arrow the whiteboard leaves out.
The record doesn’t pin the shootdown on that one arrow. The Chairman of the Joint Chiefs judged the descent reports unlikely to have “settled the issue one way or another,” given the uncertainty and the time left, and the chain of command did not fault the captain. Boyd’s own evidence was air combat and military history, not controlled tests. And speed still matters in his scheme: his recipe for getting inside an opponent’s loop is to be “more subtle, more indistinct, more irregular, and quicker.” Quicker comes last.
The Chairman’s fix was not about speed either. Senior air warfare coordinators, he wrote, should “confirm or disaffirm important reports (such as descending altitude),” especially ones that change the picture right before the firing point.
The discount nobody needed
Imagine you run a sales team, and two large customers cancel in one week. You’ve just heard that a rival cut its prices. From that moment, everything looks like price. You call your next ten accounts, fast, and offer each a discount.
One of them mentions in passing that your new billing system charged them twice in March. So, when you check, did one of the two who left. The discounts answered a question nobody was asking, and they now come off every renewal.
Slowing down for its own sake wouldn’t have helped. One line written before the calls might have: if this isn’t about price, what would I expect to see? Give it to someone who doesn’t share your hunch, and ask them to look.
A chat that can’t let go
Researchers at Microsoft Research and Salesforce Research ran more than 200,000 simulated conversations with 15 language models on six kinds of generation task. Each task came either all at once or in pieces over several turns, fed by a model playing the user. In pieces, performance dropped by 39% on average. The models, they found, “often make assumptions in early turns and prematurely attempt to generate final solutions, on which they overly rely.” In a separate check, they also tested translation on two models, GPT-4o and GPT-4o-mini. There, each new piece didn’t change the earlier ones, and they saw no such drop.
Their advice to users: if a conversation has gone off track, start a new one, and first ask the model: “Please consolidate everything I’ve told you so far,” and take that summary with you.
Read that summary before you paste it. It is the model’s picture of your task, written where you can check it. If it says “the client wants a refund” where your notes say “the client wants a replacement,” fix that line first. Then run the fresh conversation two or three times, since runs differ, and judge each result against what the task actually required.