I wrote about this in Russian Nuclear Cruise Missile Accident Analysis; Reverse Engineering 9M730 Burevestnik.

Quoting (RFE) Did A Botched Bid To Recover A Sunken Missile Cause The Russian Radiation Blast?,

Documents, photographs, satellite imagery, and other open-source materials reviewed by RFE/RL point strongly to an accident — most likely underwater, or just near the surface — …

Quoting (CNBC) US intel report says mysterious Russian explosion was triggered by recovery mission of nuclear-powered missile, not a test,

“This was not a new launch of the weapon, instead it was a recovery mission to salvage a lost missile from a previous test,” said a person with direct knowledge of the U.S. intelligence assessment. “There was an explosion on one of the vessels involved in the recovery and that caused a reaction in the missile’s nuclear core which lead to the radiation leak,” said another person, who spoke to CNBC on the condition of anonymity.

This puts my analysis in the class of “brave efforts.” But the intelligence community analysis has a peculiarity, using low-to-normal risk elements to explain a lethal accident involving a high risk test article.

Recovery barges don’t need or carry explosive material, other than motor fuel. A gasoline explosion is possible, as are drunken crewmen, but with a high risk test, scenarios incorporating higher risk elements are preferable.

There is speculation that an explosion on the barge caused a control rod to drop out of the reactor, causing prompt criticality, resulting in explosion. In the West, the danger of a design that can go prompt critical as a result of withdrawal of one control rod was made apparent on January 3, 1961, with the explosion of SL-1. One might expect that since the Russians have had since 1961 to act, they would put a strong spring on that rod.

I can’t repair my analysis, so here’s a try at helping the IC analysis to base firmly on high risk elements.

• Try #1: The reactor had been operating under water continuously since the failed test. Contact with sea water prevented explosive disassembly until that contact was interrupted, by a lifting crane. When the hoist cleared the water, heat production was evident. It was then lowered into the water, resulting in a steam explosion.
• Try #2. The Russians were aware that the reactor was operating. They rigged a cooling system prior to lifting the reactor clear of the water. A hose broke. The rest follows Try #1.
• Try #3. The reactor design is over-damped. This means that intrusion of sea water slows fission. The air passages that support ramjet operation were filled with sea water.  When the water drained from the air passages, fission resumed at the rate allowed by the controls, without cooling by the swiftly flowing air of flight. When excess heat was noted, the reactor was dropped back into the water, resulting in a steam explosion.
• Try #4. A static test of the reactor. A static test involves a flight component in a non-flight configuration. As with the U.S. Pluto tests, the test reactor was cooled via tanks of compressed air. Pluto also had a water jacket. A barge was used for safety and ease of emergency disposal . The planned response to a reactor event was to drop it in the bay. The operators waited until it got too hot; steam explosion follows.

The above does not have the benefit of intelligence community resources. It does not exclude a low risk element. Someone could have been smoking next to a leaking tank of gasoline. In a confined space, gasoline fumes are explosive. But with test of a dangerous article, human factors tend to recede in comparison to  inherent danger.