The Register has surfaced a compact engineering-history lesson from Microsoft veteran Raymond Chen: a Word 97 crash that appeared in the lab, then disappeared whenever engineers tried to observe it with a debugger. According to The Register’s account of Chen’s story, the defect emerged near the end of the Word 97 release cycle. A test script could trigger the crash fairly reliably, but the bug stopped showing itself as soon as the team tried to debug it. Chen framed the problem as the kind that unsettles release teams: if a sporadic lab crash ships, it can become a recurring field failure. The team’s problem was not simply that the crash was rare. It was that the act of investigating it changed the conditions enough to make it vanish. The Register says discussion turned to an in-circuit emulator, or ICE, which could stand in for the CPU and support breakpoints inside the emulated chip. Chen described such hardware as almost mythical to many software developers at the time. The path eventually led below the application layer. The Register reports that the team traced the issue to a specific manufacturer and then to a CPU erratum: a hardware-level defect triggered by a particular code sequence. The compiler being used for Word 97 had already been updated to avoid generating that sequence, but there was a release-management catch. By then, the toolset for Word 97 had been locked. Replacing the compiler at that stage might have removed the known risk while introducing new, unknown ones. The team therefore did not swap in the newer compiler, according to The Register’s account. Instead, engineers searched the binaries for code that could trip the affected CPUs. The Register says they found one instance. The fix was deliberately narrow: a binary patch inserted a NOP into the offending sequence, avoiding a broader rebuild or toolchain change late in the cycle. Word 97 is remembered for more visible additions, including Clippit, better known as Clippy, and Visual Basic for Applications. Chen’s anecdote is about the less visible release work underneath that debut: when a hardware erratum, a locked toolchain and a disappearing crash converged, the shippable fix was a single carefully placed instruction. Who benefits: Engineering teams benefit from the lesson: reproducibility, hardware errata awareness and disciplined release controls matter most when a bug resists ordinary debugging. The account also benefits anyone managing mature software where late changes can be riskier than targeted fixes. Who's exposed: Teams exposed to similar failures are those shipping close to hardware behavior while relying on locked toolchains. If debugging changes timing or execution enough to hide a defect, conventional inspection may not be enough.