The most consequential number in workplace safety is the Total Recordable Incident Rate. Contractors win and lose bids on it. Executives carry it in bonus scorecards. Boards receive it quarterly and read its movements as the state of safety.
Then there is the research. Hallowell and colleagues, in The Statistical Invalidity of TRIR as a Measure of Safety Performance (Professional Safety, 2021), examined the metric's statistical behaviour and reached a conclusion the industry has still not absorbed: there is no statistical relationship between a company's TRIR and its risk of serious injury or fatality, and at the sizes of most real workforces, year-to-year TRIR movement is dominated by random variation. The decimal-point changes that drive celebration and alarm are, statistically, mostly noise.
The industry's primary instrument does not measure the thing the industry most needs to prevent.
The macro data says the same thing
Zoom out and the national statistics rhyme with the statistical critique. US Bureau of Labor Statistics injury and illness data shows recordable injury rates falling by nearly half since 2006. Across the same period the fatal injury rate fell by roughly a fifth. If recordables were a faithful proxy for catastrophic risk, the two curves should have moved together. They did not, and the divergence is the entire problem in one chart: we improved the measured thing dramatically while the outcome that matters most improved far more slowly.
The global picture is harsher still. The International Labour Organization estimates around 2.93 million work-related deaths per year, with the large majority caused by occupational disease, harms that accumulate over years and appear in no incident log at all. And in high-hazard industries the exposure concentrates where visibility is weakest: IOGP data on oil and gas fatalities shows 81% were contractors, the workforce segment least covered by the reporting systems producing the headline metric.
Why a real number can measure nothing
TRIR is not fabricated. Every recordable is a real event. The failure is statistical and structural at once:
- Serious events are too rare to count. Fatalities and life-altering injuries are, thankfully, rare enough that a rate built on total recordables is dominated by the frequent, minor events, which research shows arise substantially from different causal pathways than the catastrophic ones.
- The metric invites its own corruption. A number tied to contracts and bonuses attracts classification pressure: the case managed into first aid, the report discouraged. The metric can improve while nothing about the underlying hazard changes.
- It is a lagging count of harm. Even if TRIR were statistically sound, it reports after the fact. Prevention needs instruments that move before the event.
A metric that is unrelated to the worst outcomes, gameable under pressure, and only available after the harm is not a safety instrument. It is an accounting convention.
What a predictive safety instrument looks like
The data to do better usually already exists inside the safety function's own records, in observations, permits to work, near-misses, inspection findings, contractor onboarding, maintenance backlogs. Assembled, that history supports a different set of instruments:
- Precursor tracking: the presence of the specific high-energy conditions and failed controls that research links to serious events, counted directly rather than inferred from minor-injury frequency.
- Predicted incidents: site-level and activity-level probability estimates, backtested against history and honest about their calibration, so "this site, next quarter" carries tested odds rather than last year's rate.
- Decision-linked leading indicators: not dashboard decoration, but predictions wired to named owners with thresholds, so a rising probability triggers a scheduled intervention, and the outcome is recorded against the prediction that prompted it.
None of this retires the recordable count; regulators require it and trend context has value. The shift is in what the function steers by: recording harm remains the floor, predicting and preventing it becomes the instrument panel.
Count recordables because you must. Steer by predictions, because the count cannot see what kills people.
Predicted incidents with published calibration are what the Prophesee Compliance Suite's EHS module produces from the records a safety function already keeps. Built with and proven inside global enterprises. See your own sites' odds. Start here.