Good Recordable Rate, Collapsing Reporting Culture

Competing explanations:

1. Genuine improvement - real hazard reduction, fewer exposures, better engineering. Consistent with a falling recordable rate; inconsistent with flat workers' compensation frequency, because a real reduction in injuries shows up in claims too.
2. Reporting and classification effects - injuries recategorized as first aid, cases managed to return-to-work so they never reach the 300 log, or employees discouraged from reporting. This is the explanation that flat claim frequency plus falling recordables most strongly supports: the same injuries are occurring and being counted differently. Rising average claim cost fits too - the small claims are being suppressed out of the log while the serious ones still surface.
3. Culture collapse - near-miss reports down 64%, observations halved, two EHS managers in three years, and a supervisor-level incentive to keep the number clean. Falling near-miss reporting is the classic signature of a reporting system failing, not of a safer plant; safe plants usually report more near misses.
4. Exposure mix shift - work moved to contractors or temporary staff whose hours are not in the denominator. Check before believing any rate.

Two discriminating checks:

• Reconcile the 300 log against an independent injury source - the workers' compensation first reports and the clinic's billing for the same period. If claims exist with no matching log entry, the rate is a reporting artifact.
• Test reporting propensity directly - an anonymous survey asking whether workers would report a near miss and what happened to the last person who did, plus an audit of 10 recent cases traced from the clinic back to the log. This distinguishes 'fewer events' from 'fewer reports' in a way no rate can.

Also verify that the denominator includes all hours, including contractor and temporary labor, and break the recordable rate down by department and injury type - a sprain rate collapsing while strains persist elsewhere points straight at classification.

What the sample says. 16 of 25 permits (64%) lack the pre-entry atmospheric test signature. If that defect rate holds across 140 entries, roughly 90 entries occurred without documented verification of the atmosphere - a control failure at the point where the fatal hazard is actually decided.

Risk logic to present. Suppose the probability of an IDLH atmosphere on any given untested entry is only 1 in 1,000 and the probability of a fatality given exposure is 1 in 2. That is 0.0005 per entry; over 90 entries the annual probability of a fatality is about 4-5%, and over five years it exceeds 20%. Those inputs are arguable; the structure is not. A high-frequency control defect multiplies against a low-probability, high-consequence outcome, and the resulting number is not small. This is exactly the exposure that a recordable rate is blind to, because it produces a body, not a case on the log.

Reframing:

• The recordable rate measures outcomes with the injuries already realized. It has almost no sensitivity to confined-space, lockout, or other low-frequency/high-consequence exposures, so a 2.3 rate tells the board nothing about the site's largest risk. Presenting it as a safety summary is misleading even if every entry is accurate.
• The permit defect rate is a leading indicator and it is alarming. A 64% non-conformance is not drift; it indicates the permit system is not functioning or is being signed after the fact.
• Recommend: suspend entries until permit verification is restored and independently audited; investigate whether permits are being back-filled; and correct the VP's proposal - a near-miss program with falling reports is a failing program, and tying awards to a recordable number is a documented driver of underreporting.

Say plainly that the site cannot claim improving safety performance on the strength of a metric that cannot see its worst hazard, and that the annual report claim would not survive an incident investigation.

1. Control-verification defect rate from random audits (permits, lockout, and pre-task hazard assessments sampled, not self-reported). Defensible because it measures whether the critical controls actually happened, is sensitive to high-consequence exposures the recordable rate cannot see, and has a known denominator (items sampled). Target is a defect rate, not a count - you want the sample to stay large.

2. Reporting health index - near-miss reports per 100 employees, the ratio of near misses to recordable injuries, and the median days from report to visible corrective action. Defensible because a healthy reporting system produces near misses in multiples of injuries; a ratio falling toward zero means the system is dying. Measure the feedback loop, because workers stop reporting when nothing changes.

3. Severity-weighted potential-consequence metric - for every reported event and audit finding, record the credible worst outcome and the safeguard that failed, not just the actual outcome. Defensible because it keeps high-consequence signals visible even when nothing was injured, and it gives management a risk picture rather than a luck picture.

Each needs a defined data source, an owner, and a review cadence; a metric without those becomes a number people learn to manage rather than a control.