The Night I Was Too Tired to Trust My Own Judgment

It was the last night of a stretch that had been harder than it should have been — three port calls compressed into five days, a bunkering operation that ran long, and an unrelated repair on a cooling water pump that ate most of what would otherwise have been rest. I had slept in fragments rather than blocks for most of that week, the kind of sleep that technically counts on a rest-hours log but doesn’t actually restore anything. By the final night, I was operating on something closer to momentum than judgment.

Somewhere in the middle of that night, during a maneuvering period, No. 2 generator produced a brief, sharp vibration I hadn’t heard from it before, along with a knock lasting two or three seconds. It happened once, then again roughly forty minutes later, then not again for the rest of the watch. What actually caught my attention, beyond the sound, was a small spread in exhaust temperature across the unit’s cylinders that I noticed on the next routine round — a few degrees wider than the pattern I was used to seeing on that generator, not dramatic, not clearly abnormal on its own, but enough to make the sound and the reading feel like they belonged to the same story rather than two unrelated things. I suspected a fuel injector. I hadn’t confirmed it. By the time the watch was ending, the generator was running as smoothly as it had all week, temperature spread included, and I was tired enough to want that smoothness to mean the question had answered itself.

What I Almost Wrote

I sat down to write the handover the way I always do, one system at a time. When I got to the generators, I started typing something close to “No. 2 generator operating normally, no items outstanding,” the same phrase I have written honestly hundreds of times before. I want to be precise about what that sentence actually was, because the precision matters: it was true. At the moment I typed it, the generator genuinely was running normally. I was not fabricating a reading or inventing a clean bill of health for a machine that was visibly struggling. This was not the kind of lie that gets caught, because it wasn’t a lie at all in the ordinary sense.

What it was, instead, was an accurate description of the present moment standing in for a report of the whole watch. It left out two knocks, a temperature spread I’d noticed and not mentioned to anyone, and a suspicion I hadn’t confirmed. None of that made the sentence false. It made the sentence answer a narrower question than the one a handover is actually supposed to answer, and I don’t think that distinction — true but narrow, versus honest and complete — gets talked about nearly as much as outright fabrication does, even though I suspect it’s the far more common way a safety-critical record actually goes wrong.

The Rationalization That Almost Worked

What makes this worth writing about isn’t the sentence itself. It’s how reasonable every argument for leaving it that way felt in the moment. It had resolved on its own, twice, and hadn’t recurred in over three hours — a stable pattern is not the same as an active fault, and I have genuinely seen stranger transient noises turn out to be nothing. Writing it up properly meant staying an extra fifteen or twenty minutes past the end of an already brutal week, and I wanted, badly, to be relieved and sleep. And there was a thought I’m least proud of: that flagging something this vague might read as jumpy, or invite questions I didn’t currently have good answers to.

None of those thoughts felt like cutting a corner while I was having them. They felt like calibrated judgment — the kind a Chief Engineer with real years behind him is supposed to have earned the right to apply. I don’t think that feeling is unique to fatigue, exactly, but fatigue is what made the narrower, easier answer feel sufficient rather than merely convenient. I have no doubt that on a rested night, with the same set of facts in front of me, I would have written the fuller entry without a second thought, simply because writing it wouldn’t have felt like a cost worth negotiating with myself over in the first place.

I’ve tried, since, to work out what would have needed to be different for me to catch this as a problem in the moment rather than in hindsight. If someone had handed me a written instruction that said “record only what the instruments show right now, omit anything unconfirmed,” I would have refused it immediately and probably questioned why anyone would suggest it. Nothing that explicit happened. What happened was a sequence of small, private negotiations, each one defensible on its own terms, that arrived at the same destination that instruction would have produced. I think most reporting failures on a ship look closer to this version than to the dramatic one — not a decision to hide something, but a chain of individually reasonable shortcuts that only look like a pattern once you step back far enough to see where they were heading.

What Made Me Stop

What actually stopped me was a small, almost mechanical test, not a dramatic moment of conscience. I read the sentence back and asked myself what I would say if the relieving engineer walked in right then and asked, out loud, whether there was anything on my mind about the plant. I knew immediately I would tell him about the generator — the sound, the timing, the temperature spread, the unconfirmed suspicion. There was no version of that direct conversation where I would have said “nothing to report.”

That gap, between what I would have said out loud and what I had actually typed, was the entire problem in one place. A handover exists to transfer what I know to the person taking over, not just to confirm what the instruments currently show. Mine had quietly narrowed to the second thing. I deleted the sentence and started again.

What the Full Entry Said, and What Happened Next

The entry I actually submitted described both events in detail — approximate times, duration, the temperature spread, my suspicion about the injector, and an explicit line stating the cause was unconfirmed and the unit should be monitored on load. It took perhaps fifteen minutes longer to write than the version I deleted.

Three days later, the same knock returned during another maneuvering period, witnessed by the Second Engineer coming on watch. Because the earlier entry existed, he didn’t start from nothing — he already knew the sound had a history and what had been suspected, and escalated straight to an inspection rather than spending a watch deciding whether what he was hearing was worth mentioning. The inspection found a fuel injector on that unit with an early, developing fault, caught before it became the kind of failure that takes a generator out of service. I want to be honest about how much credit that outcome deserves to give me: not much. The Second Engineer did the diagnostic work, and the fault would probably have surfaced eventually regardless, in some form, on some later watch. What the fuller entry actually bought was three days of documented history instead of three days of a story living only in my head — a smaller, quieter kind of value than “prevented a disaster,” and I think it’s also the more accurate description of what good reporting usually does.

I’ve also asked myself, less comfortably, what would have happened if the knock hadn’t recurred at all — if the injector had simply failed quietly over a longer period, or the fault had turned out to be something else entirely that never announced itself with a second, corroborating event. In that version, the fuller entry might have sat in the log for weeks looking like nothing, an “unconfirmed, monitoring” note that never got followed up on because nothing forced anyone’s hand. I don’t have a tidy answer for that scenario. What I have is a preference: an honest, unresolved entry that might get overlooked is still a better artifact than a tidy one that was never true to begin with, because at least the first one gives the next person a chance to notice what the second one guarantees they won’t.

Why This Isn’t Really a Story About Lying

I’ve used the word “gap” more than “lie” on purpose throughout this, because I don’t think framing it as dishonesty is accurate, and I don’t think it’s what would have reached me on the night it mattered. I already believed, sincerely, that fabricating a reading was wrong, and that belief did nothing to stop what almost happened, because what almost happened wasn’t fabrication. The more useful lesson, for me, has been about what exhaustion specifically manufactures: not the impulse to falsify, which is rare and easy to recognize as wrong, but a steady supply of reasonable-sounding justifications for reporting only the part of the truth that’s cheapest to write down.

I don’t have data on this beyond my own experience and conversations with a handful of colleagues since, so I’ll say it carefully rather than as a settled fact: I suspect a fair number of near-miss and hazard-reporting programs are built around an assumption of alertness that the exact moments they’re designed to catch don’t always allow for, since unconfirmed anomalies disproportionately turn up at the end of long, tired stretches. That’s a suspicion, not a finding, but it matches what I’ve seen closely enough that I’ve stopped assuming a good reporting culture on paper automatically survives a bad week in practice.

There’s a version of this argument that would let me off the hook entirely — fatigue made me do it, so the responsibility sits with the schedule rather than with me. I don’t think that’s quite right either. I was capable of writing the honest entry. I did, in fact, write it, once I asked myself the right question. The fatigue didn’t remove my ability to report accurately; it lowered the price I was willing to pay to do it, right up until a specific, deliberate check reminded me what the actual price of not doing it might be. That’s a narrower and less comfortable claim than “I wasn’t myself that night,” and I think it’s the more honest one.

What I Changed on Every Vessel Since

I already had reasonable handover habits before that night — I wasn’t starting from nothing. What that week showed me was that those habits had a threshold, and I’d found it. The changes below are less about replacing bad practice and more about giving my ordinary habits less room to fail when fatigue is worse than usual.

• I keep a running note through the watch as things happen, rather than reconstructing the whole night from memory at the end of it, specifically because the end of a hard stretch is exactly when reconstruction gets unreliable.
• Any anomaly gets recorded as observed, whether or not it resolved on its own and whether or not I’ve confirmed a cause — “unconfirmed, monitoring” is a legitimate entry, not something to withhold until I have a tidier answer.
• Before submitting a handover, I run the same test that stopped me that night: would I say this exact thing out loud if asked directly. If the spoken answer is bigger than the written one, the written one isn’t finished.
• I tell junior engineers explicitly that “monitoring” or “not yet confirmed” is the correct entry for a genuinely unresolved situation, not a sign of weak diagnostics — I’d rather read ten of those than one falsely tidy “all normal.”
• When a stretch has genuinely been exhausting, I say so directly to whoever is relieving me, out loud, as its own piece of information, because a handover that doesn’t mention the state of the person giving it is missing something relevant to how much weight the rest of it deserves.