We have gone four years without a lost-time incident across 6.4 million man-hours. I want to explain how, and then explain why I do not fully trust the number.
The boring practices
There is no innovation in this list. That is the point.
Pre-task planning at the crew level. Not a generic method statement filed in an office — a five-minute conversation at the workface, led by the foreman, about what this crew is doing in this location today and what could hurt them. Written on a card, signed, kept on the wall.
Stop-work authority that has actually been used. Anyone can stop work. The test of whether this is real is not the policy, it is whether anyone has ever done it and what happened to them afterwards. Ours has been used 41 times in four years. Nobody was disciplined. Two of those stops found genuine structural issues.
Plant and pedestrian separation. The single highest-consequence risk on our sites is a person meeting a machine. Physical segregation, banksmen, and reversing restrictions do more than any amount of training.
Scaffold inspected by someone who did not build it. Independence matters more than frequency.
Heat management as a safety system, not a comfort one. Mandatory shaded rest, electrolyte availability, and a work-rest ratio that changes with WBGT readings rather than the calendar.
Why I distrust the metric
Lost-time incidents are a lagging indicator, and a rare one. Four years of zero LTI across 6.4 million hours is genuinely good — and it is also a small enough sample that luck plays a real role.
Worse, LTI is gameable in ways that are invisible from the outside. A company can report zero LTI by putting injured workers on light duties, by classifying aggressively, or simply by having a workforce that fears reporting. I have seen all three, and none of them show up in the headline number.
So we watch leading indicators harder than the lagging one:
- Near-miss reporting rate. A rising near-miss count is good news. It means people are reporting. When near-miss reports drop, I get worried, not pleased.
- Percentage of pre-task plans audited and found accurate — not just completed.
- Time to close corrective actions. An open action from three months ago is a system that does not close its loops.
- Subcontractor induction coverage before first workface entry.
The thing that would break it
Growth. Every one of these practices depends on supervisors who have absorbed them, and supervisors take years to develop. Scale the workforce faster than you can develop supervision and the safety system thins out invisibly until something happens.
That is the risk we manage most deliberately: we turn work down when we cannot staff supervision properly. It is the least popular sentence in this article and the most important one.