Leading Safety Indicators and Serious Injury/Fatality Prevention: What Employers Should Measure

An employer can finish a quarter with no recordable injuries and still have workers exposed to uncontrolled electrical energy, suspended loads, moving equipment, or unprotected falls. Conversely, an increase in reported close calls can reflect greater trust and better hazard detection rather than deteriorating safety. Neither observation proves what will happen next.

OSHA recommends combining lagging indicators (outcomes that have already occurred) with leading indicators (measures of preventive processes and safeguards). Its 2019 publication, Using Leading Indicators to Improve Safety and Health Outcomes (OSHA 3970), emphasizes choosing a meaningful indicator, assigning responsibility, collecting data, responding to findings, and reassessing whether the indicator remains useful. The publication explicitly says that it does not create new legal obligations. 1

For serious injury and fatality (SIF) prevention, the question is more specific than “How many safety activities did we complete?” It is: Where could a person be killed or permanently harmed, what should prevent it, and what evidence shows that protection is working while exposure exists? Construction research suggests that high-energy hazards, deficient planning, and missing direct controls distinguish many high-severity events from low-severity injuries, but those findings are not a universal prediction rule. 5

1. Leading and lagging indicators answer different questions

Lagging indicators describe outcomes that already happened: fatalities, serious injuries, OSHA-recordable cases, and days-away or restricted-work cases. Leading indicators describe whether prevention is being carried out and is functioning: high-risk task plans checked at the point of work, critical controls successfully verified, hazards corrected, workers' reports answered, and repeat failures detected. Measured exposure is context-dependent: a noise dosimetry or airborne-contaminant result documents exposure that occurred, but it can also be used proactively to detect inadequate controls before a worker develops an illness. Classify it according to the question being measured rather than assigning every exposure measurement to the lagging category.

The distinction is not absolute. A close call is a lagging event because it already happened; the quality and timeliness of its review, and the corrective action it triggers, can serve as leading measures of preventive work. A completed training session is a process measure, but it may tell little about competence until actual understanding and execution are checked.

An effective scorecard retains both types. A fall-protection inspection pass rate without injury outcomes can conceal failures in reporting. A low DART rate without evidence of effective fall, equipment, and energy controls can create false reassurance. OSHA's recommended program-evaluation practices expressly call for both kinds of measures. 2

What the Heinrich safety triangle can: and cannot, tell an employer

The historical Heinrich triangle is often rendered as 300 no-injury events, 29 minor injuries, and one major injury. That relationship is not a conversion factor: recording 300 near misses does not mean a workplace will have one serious injury, and reducing reports by a given percentage does not establish that fatalities will decline proportionately. Event definitions, exposure, the types of work performed, and reporting behavior all affect such counts.

The research is more nuanced than simply declaring the entire idea worthless. Yorio and Moore (2018) used records from 27,446 mining establishments during 2000–2012 and found that lower-severity events could be associated with the probability of subsequent fatalities within the same establishment. But the triangular pattern depended on how injury severity was classified; it did not hold uniformly across definitions. Their observational study does not establish a universal ratio, a transportable fatality forecast for a contractor, or proof that collecting more near-miss forms causes fewer deaths. 12

A useful distinction follows: examine recurring low-severity events for shared conditions such as deficient supervision, maintenance, or risk control, while also separately identifying events and exposures with credible catastrophic potential. A forklift narrowly missing a pedestrian and a superficial paper cut do not warrant the same severity classification simply because each contributes one report. The proper test is the hazard and failed protection, not a triangle count. 5

2. what TRIR and DART can: and cannot, show

The total recordable incident rate (TRIR, also called total case rate in some datasets) is commonly expressed as:

TRIR = qualifying recordable cases × 200,000 / hours actually worked

The days away, restricted work, or transfer rate (DART) is:

DART = (Form 300 column H cases + column I cases) × 200,000
       / hours actually worked

The constant 200,000 represents 100 full-time workers each working 40 hours a week for 50 weeks. Use actual hours worked, not paid leave or scheduled hours. OSHA's forms package and Injury Tracking Application guidance document these calculations. For comparisons with BLS nonfatal injury rates, use the corresponding nonfatal recordable-case count (Form 300/300A columns H, I, and J); OSHA recordable fatalities appear separately in column G. Some company or contractor uses of “TRIR” include fatal cases in the overall recordable count. State the convention and do not silently compare an all-case numerator against a nonfatal BLS benchmark. 3

Both rates are useful, but limited:

• They reflect injuries and illnesses that have already occurred and been reported correctly, not the condition of current safeguards.
• A DART numerator counts cases with days away, restriction, or transfer, not deaths in Form 300 column G. A fatality always needs to be tracked and investigated separately.
• Case counts can be volatile in small workforces. A single case in 100,000 hours changes an incidence rate by 2.0. OSHA advises caution when interpreting small establishments' rates. 4
• Different job mixes, workforce scopes, hours, contractor arrangements, and reporting practices make comparisons misleading if they are not reconciled.
• A zero in one period is an observation, not a statistically defensible prediction that fatality risk is zero.

Do not use a favorable rate to close an unresolved high-consequence hazard. Equally, do not discard recordable-case data: investigate it alongside high-risk exposure and control information.

3. Define SIF and SIF-potential before counting either

There is no single OSHA Part 1904 classification called “SIF-potential.” The term SIF is widely used in prevention practice for serious injury or fatality, generally emphasizing fatal, life-threatening, or life-altering outcomes. Organizations need a written, reproducible definition. For internal SIF analysis, classify an actual fatality separately from a serious nonfatal injury. The latter can be defined as an injury that threatens life or causes a substantial or permanent loss of function. Examples may include a traumatic major amputation, permanent major vision loss, severe brain injury, or paralysis. Case classification needs to rely on credible medical and incident evidence, not speculation about diagnosis. Some programs also include serious occupational illnesses; if so, document their inclusion and classification rules. Industry definitions differ. 7

SIF-potential (pSIF) means an event that did not produce a SIF but presented a credible pathway to one. It may be a near miss or a lower-severity injury. Examples:

• A suspended load swings through an occupied work area without striking a worker.
• An energized machine starts unexpectedly during servicing, but nobody is caught.
• A vehicle enters an occupied pedestrian exclusion zone without contact.

An unprotected fall edge with an exposed worker, even if nothing unusual has yet happened, is an urgent high-potential hazardous condition or exposure. It should be captured and controlled, but do not automatically relabel it as a pSIF incident if the employer's incident taxonomy calls for an unplanned event. Maintain distinct fields for pSIF events and high-potential conditions so trend numerators remain interpretable.

An event with an actual minor injury may simultaneously be OSHA-recordable and internally classified as pSIF because its potential consequence was much worse than the actual outcome. Conversely, a serious injury is not automatically in OSHA's separate severe-event reporting category; apply the exact Part 1904 criteria independently. Never substitute internal SIF labels for recordkeeping or legal reporting decisions.

Avoid defining pSIF as any event imaginable to cause death after a long chain of unlikely assumptions. Review the actual energy or hazard, the person's position, the failed or missing safeguards, and a reasonably foreseeable change that could have resulted in severe harm. Record both actual outcome and credible potential outcome in different fields. Use trained reviewers and a consistent classification guide; audit disagreement between reviewers.

Legal distinction: OSHA's recordkeeping criteria are in 29 CFR 1904.7. OSHA's separate severe-event reporting rule, 29 CFR 1904.39(a), generally calls for reporting a covered work-related fatality within eight hours and a covered in-patient hospitalization for care or treatment, amputation, or loss of an eye within 24 hours, subject to the regulation's timing and other qualifications. A pSIF with no actual injury is not automatically a reportable or recordable case. Certain hazard-specific standards can nevertheless call for review, training, or investigation after near-miss events. 8

4. Prioritize high-energy hazards and failure pathways

“High-energy” in SIF research refers to energy capable of causing severe harm if released or contacted. Think about gravity and falls, moving vehicles and machinery, pressure, stored mechanical energy, electrical energy, temperature, and unstable structures. Hazardous atmospheres, toxic exposures, and other catastrophic hazards also deserve SIF attention even when a mechanical-energy threshold cannot describe them.

Hallowell, Alexander, and Gambatese (2017) examined 505 retrospective injury reports involving gravity and motion. Researchers estimated energy magnitude and, where possible, energy intensity (energy relative to the area of contact) from incident descriptions, separated those estimates from injury-outcome classifications, and used statistical comparisons across severity groups. In the studied cases, greater magnitude and intensity were associated with more severe outcomes; intensity distinguished severity levels better because it considered contact area. The distributions still overlapped. An injury involving a sharp object striking a vulnerable body part could be fatal despite comparatively modest energy magnitude. The paper explored an approximate 2,000-joule magnitude criterion for certain gravity/motion hazards, but the authors did not identify a clean dividing line that made lower-energy work safe. These observations are not a prospective validation of a workplace fatality predictor. 6

Bayona, Hallowell, and Bhandari (2024) assessed 38 retrospective construction incidents: 13 SIF, 12 pSIF and 13 low-severity cases. Experts blinded to outcomes assessed seven factors. Separate Fisher tests compared SIF with low-severity cases and SIF with pSIF. Missing direct controls and absent or unfollowed work plans distinguished SIF from low-severity cases; none of the factors significantly distinguished SIF from pSIF. The analysis did not pool SIF and pSIF for comparison with low-severity cases. 5

Research definitions: That study selected pSIF incidents involving more than 1,500 joules without a direct control, and low-severity cases involving less than 500 joules with first aid or less. These are study criteria, not OSHA definitions or thresholds for safe work. An employer’s broader assessment needs to also cover toxic, atmospheric and other potentially fatal hazards. 5

The table describes selected cases, not the probability of a fatality given a missing control. Its practical use is to direct attention to safeguards and planning failures, then evaluate those controls during actual work. 5

The study used a small convenience sample and reconstructed incidents; selection, retrospective judgments and confounding limit generalization. Its 14 factor comparisons were exploratory. The 1,500-joule study definition and the approximately 2,000-joule criterion explored in 2017 answer different research questions; neither is a regulatory limit or assurance of protection. 5

For each high-consequence task, create a short exposure profile:

1. Task and energy: What could move, fall, energize, release pressure, or expose the worker?
2. Person at risk: Who could enter the line of fire, including contractors and nearby crews?
3. Credible severe outcome: What harm could realistically follow from loss of control?
4. Needed safeguard: What needs to be present and effective before work starts and remain effective during it?
5. Verification: Who can confirm the safeguard is working, by what approved method, and at what point?
6. Failure response: Who can stop or isolate the affected work and authorize its safe resumption?

Use these profiles to select the work that warrants more frequent independent observations. Do not substitute an arbitrary inspection quota for exposure-based sampling.

5. Critical controls: verify effectiveness, not paperwork

A critical control is a specific safeguard whose absence or failure could permit a high-consequence event. The term is a risk-management classification; it does not replace the requirements in applicable OSHA standards. For each selected control, specify its safety function, a measurable or directly observable pass condition, the verification method and trigger, the authorized person, and the needed response to failure. Controls should reflect the hierarchy of controls, elimination, substitution, engineering protection, administrative controls, then PPE where needed. NIOSH and OSHA favor controls that do not depend continuously on worker actions. 9

These are examples, not substitute compliance procedures. The correct standard depends on the work and jurisdiction; regulatory requirements and employer-selected performance checks are different things. Where a specific OSHA inspection, test, or training rule applies, follow that rule even if the internal dashboard would otherwise show a satisfactory percentage. See the exact source provisions in 8, 13, and 14.

Example: a critical-control verification card

Illustrative control: isolation of hazardous energy before applicable machine servicing under 29 CFR 1910.147.

The card's pass/fail field should describe what was actually checked. “Form signed” and “training completed” are not substitutes for the control's protective function. A control can pass in the morning and fail after a configuration or weather change; build rechecks around meaningful changes rather than assuming a previous pass lasts forever.

A signed checklist shows that someone signed a checklist. It does not establish that a guard works, an isolation point is correct, a barrier actually separates workers from traffic, or a protective system remains appropriate after conditions change. Verification should include appropriate field inspection, testing where authorized and safe, review of maintenance status, and confirmation that people can work within the intended protected condition. Never deliberately defeat an interlock or create exposure in order to test a scorecard.

OSHA's recommended hazard-prevention practices explicitly call for checking that engineering controls are installed, tested, used correctly, and remain effective. A critical-control failure is an operational issue to correct promptly, not just a red box awaiting the next monthly dashboard meeting. 10

6. Build the reporting and corrective-action loop

Near misses. A count of submitted reports is not a measure of risk by itself. A rising number can mean more hazards, better reporting trust, more workers, or broader observation. A falling number can mean prevention, or silence. Use the reports to ask: Was there credible SIF potential? What failed? Was the exposure removed? Did the control pass independent verification? Did the reporting worker receive a useful explanation of what changed?

Evaluate quality with a short checklist: specific event and location; task and involved energy; people who could have been exposed; actual outcome; potential outcome; safeguard status; independent review; assigned control action; and verification evidence. An incomplete report should trigger supportive follow-up, not punishment.

Hazard identification. Do not reward simply finding more minor hazards. Track planned coverage of high-risk tasks and whether serious hazards are appropriately prioritized and controlled. Record the scope of inspections: sites, shifts, job types, contractor work, and exposure categories. Spot-check whether independently observed hazards are already visible in the employer's reporting system.

Corrective action. Classify by potential severity and exposure, assign a responsible owner, specify an interim protective measure, retain the original due date, and record closure evidence. Distinguish administratively completed from physically implemented and verified effective. A calendar-based closure target is an employer program rule unless an applicable standard, contract, or order says otherwise. Serious uncontrolled hazards need to be addressed promptly regardless of KPI deadlines.

Worker involvement. Count meaningful participation rather than attendance signatures alone. Invite affected employees to identify energy sources, test whether proposed controls fit the work, and verify that fixes resolve the concern. Report back to workers in a language they understand. Include contractors and temporary workers where they face exposure.

OSHA's program-evaluation and worker-participation guidance recommends tracking participation, hazard and near-miss reporting, response times, corrective-action completion, and worker opinion. Under 29 CFR 1904.35, covered employers need to maintain reasonable injury/illness reporting procedures and may not retaliate for reporting. OSHA's 2018 interpretation permits incentive programs, including certain rate-based programs, when they are not implemented to deter reporting. 111

7. A practical eight-indicator framework

Start with a limited set, name the owner for each metric, and keep the raw exception list available. The suggested frequencies, quality tests, and any dates below are internal design choices, not OSHA-mandated intervals.

For every percentage, define the denominator before measuring. If no cases were eligible, show N/A (0 eligible), never 100% by default. Report small counts beside percentages. Keep stratified results for fatalities, pSIFs, task types, failed critical controls, and overdue high-severity actions rather than burying them in an average.

The eighth metric can be combined with the first for a smaller employer. OSHA has no universal set of eight safety KPIs.

An exact specification for the critical-control indicator

Unit of analysis: one needed control-verification opportunity, not one inspection tour, one signature, or one worker. A high-risk task with two separately needed critical controls can contribute two opportunities if each has a defined verification trigger. Record the task ID and control ID so the denominator can be audited.

Numerator: scheduled opportunities due during the period for which an authorized, appropriately timed, satisfactory verification was documented before the relevant exposure or decision point. A late pass is not an on-time pass. A failed check corrected and subsequently passed needs to retain the original failure and its timing; it cannot be silently reclassified as an initial pass.

Denominator: all verifications that became due during the period, including those triggered by unplanned work, changes in conditions, or a repeated check requirement. Exclude a canceled task only when it was canceled before the verification became due, no work or exposure occurred, and the cancellation reason and timestamp are documented. Keep checks that had already become due, including failed or missed checks, in the original denominator. A postponed task creates a new applicable trigger when it actually resumes. Missing results, failures, and unauthorized checks cannot count as passes.

Reporting: present the pass rate alongside the absolute number of failures, missed checks, active exposures, and unresolved critical deficiencies, broken down by control type and site. Review whether the underlying list of due checks matches actual high-risk work through independent observation. This is a practical employer metric design, not an OSHA-prescribed calculation.

8. A hypothetical monthly dashboard

Hypothetical contractor, one month. Each metric uses the task or record population specified in its definition; the sample sizes therefore differ.

Worked calculation 1, verified controls. Eighty scheduled checks were due. Sixty-four were satisfactorily verified on time; 10 found failures and six were missed. Therefore 64 ÷ 80 × 100 = 80.0%. A denominator limited to 74 attempted checks would show 64 ÷ 74 = 86.5%, falsely hiding six missed verifications. Retain separate failure and missed-check counts.

The 16 exceptions are not presumed to have resulted in actual worker exposure. The employer needs to determine, case by case, whether work was prevented or stopped, what interim protection was used, and how the safeguard was later verified. Reporting 16 checks as failed/missed is different from claiming 16 injuries or even 16 uncontrolled exposures.

Worked calculation 2, overdue corrective actions. Ten actions had original deadlines in the month; seven had been physically implemented and effectiveness-verified by those deadlines. 7 ÷ 10 × 100 = 70.0%. Moving a deadline or closing a ticket without checking the safeguard does not turn a late action into an on-time verified closure.

Worked calculation 3, TRIR and DART. Suppose the same hypothetical operation logged 100,000 employee hours worked, three nonfatal OSHA-recordable cases, of which two were days-away/restricted/transfer cases. TRIR (nonfatal comparison basis) = 3 × 200,000 ÷ 100,000 = 6.0. DART = 2 × 200,000 ÷ 100,000 = 4.0. These are rates per 100 full-time-equivalent workers, not 6% and 4% probabilities. No fatality was assumed in the example. If there had been one, it would need a separate fatal-event line and careful disclosure of any all-case TRIR convention used.

Worked calculation 4, a small number problem. In 100,000 hours, one additional recordable case adds 1 × 200,000 ÷ 100,000 = 2.0 rate points. Comparing 6.0 in one small period to 4.0 in another cannot by itself establish that the workplace became safer or less safe. Investigate the actual events and exposures, use longer periods where appropriate, and show denominators.

Worked calculation 5, uncertainty in the 3-of-50 task sample. Three affected tasks out of 50 observed yield 3/50 × 100=6.0%. Only if those 50 task observations could reasonably be treated as independent, randomly sampled opportunities from a defined population, an illustrative two-sided 95% Wilson score confidence interval for a binomial proportion would be approximately 2.1% to 16.2%. Using hat p=3/50=0.06, n=50, and z=1.96, the Wilson limits are:

Wilson limits = [p + z²/(2n) ± z × sqrt(p×(1-p)/n + z²/(4n²))]
                / [1 + z²/n]
For this example: 0.0206 to 0.1622

This calculation is an illustration of sampling variability, not an inferential claim about the hypothetical contractor: the dashboard never established that the 50 tasks were randomly selected or independent. Repeat observations from one crew, convenience sampling, unobserved night shifts, and contractors omitted from the sampling frame can invalidate this simple binomial interval. An apparently precise percentage, or confidence interval, cannot repair a biased sample. NIST describes the Wilson method and why symmetric normal-approximation limits can behave poorly with small proportions. 15

The dashboard is not a composite “safety score.” A control failure involving a worker under a suspended load cannot be offset by meeting a toolbox-talk attendance goal. The first management decision is whether current high-consequence exposure is adequately controlled; only then should the team discuss trends.

9. Data quality, denominators, and detection bias

Create a one-page metric dictionary containing the numerator, denominator, inclusion/exclusion rules, data source, owner, calculation period, treatment of missing data, review method, and change history. Questions that change the rate include:

• Are observations by task, worker, control, inspection, crew, or site? One observed task with three bad guards may not silently become three tasks.
• Are contractor hours and contractor incidents included consistently? Different reporting scopes defeat comparability.
• Were high-risk nights, weekends, emergency jobs, subcontractors, and short-duration tasks sampled, or only the easiest daytime activities?
• Was the sample defined in advance, with a method that does not systematically exclude difficult work?
• Are repeat visits to the same task marked, or mistakenly presented as independent exposure opportunities?
• Do revised classifications or duplicate near-miss reports change prior-period comparisons? Retain an auditable correction trail.
• Is an unverified inspection treated as missing/failing rather than presumed passed?
• Are zero-denominator cells explicitly N/A, and are rare outcomes presented as counts as well as rates?

Use task observations or high-risk task-hours for exposure-related measures only when the scope can be collected reliably. In low-frequency SIF outcomes, short-term apparent improvement can occur by chance; one month of zero fatalities is not evidence that a given control program caused a fatality reduction. When changes in metrics coincide with an intervention, describe the timing and plausible mechanism, look for confounders, and avoid implying causal proof from an uncontrolled trend.

Rules for difficult denominator cases

Sampling protocol: Before the month begins, choose the categories and shifts to observe; include a defensible mix of routine and nonroutine high-energy work; record who selected each sample; and reserve some reviews for independent, unannounced checks. Reconcile the list of observed tasks with permits, work orders, production logs, and contractor activity. A statistical random sample can help reduce selection bias, but convenience samples are not representative. Explain any sampling changes beside the trend chart.

When several sites are combined: show each site's raw numerator and denominator before calculating an organization-wide weighted rate. Averaging site percentages without weighting can mislead: a small shop and a large project should not automatically have equal statistical weight. Always expose a serious single-site control failure even when the combined average is favorable.

10. Gaming and underreporting: how the dashboard can lie

An increase in reported pSIFs after introducing anonymous reporting may reflect better detection. A decline in reported events accompanied by worker accounts of management retaliation raises a different concern: the dataset may no longer represent actual conditions. Safety managers should speak directly with workers, compare independent inspections with reports, and review incentive pressures. 11

11. Management review: a decision routine, not a presentation

At the workface: Before high-risk work begins, verify needed safeguards using the applicable procedure. If a critical control is missing, ineffective, or unverified, stop the affected exposure, make conditions safe, correct the failure, and obtain authorized verification before resuming.

Weekly, with operations and workers: Review unresolved high-consequence exposures, missed verifications, pSIF events, repeat failures, interim safeguards, and actions past their original deadlines. Choose specific changes with an owner, due date, and effectiveness test. If the same failure occurs repeatedly, examine design, staffing, scheduling, maintenance, interface between trades, and supervisory decisions, not just worker conduct.

Monthly: Compare the indicator trends with lagging outcomes; review sampling coverage, data completeness, complaint or reporting anomalies, and discrepancies between self-inspections and independent verification. Ask whether leading indicators predictively relate to useful prevention actions in this workplace, not merely whether their percentages improved.

Periodically and after major change: Update the hazard inventory and definitions when equipment, work methods, contractors, or exposures change. Review whether each metric still produces an actionable decision. OSHA's recommended practices call for initial and periodic program evaluation, with no less than annual assessment in the guidance; this is a recommended program practice, not a universally binding annual dashboard audit requirement. 2

An exception-to-decision protocol

The following sequence is a suggested management process, not a universal OSHA notification or closure timetable. Applicable standards, orders, permits, and incident-reporting rules take precedence.

1. Detect and contain: The person discovering a missing or failed critical control alerts the responsible supervisor and prevents the affected exposure using the site's approved stop-work and emergency procedures. Do not continue an unsafe task merely to obtain a complete sample.
2. Record the original result: Preserve the task, control ID, location, shift, time, verifier, finding, and whether anyone was actually exposed. Record an initial failure separately from any later satisfactory recheck.
3. Choose the protection: The operations owner identifies a compliant correction and any necessary interim measures. If effective protection cannot be established, the affected work does not resume. Where OSHA assigns a duty to an authorized employee, competent person, or qualified person, the company needs to use the person specified by the applicable rule, not an arbitrary dashboard owner.
4. Verify and authorize restart: An appropriate person confirms the applicable requirements and task-specific pass criteria have been met. A management sign-off without evidence that the safeguard works does not satisfy the control objective.
5. Learn and escalate: EHS and operations review repeated failures, pSIF events, missed verifications, and recurring conditions across shifts and contractors. An unresolved high-consequence exposure is escalated when discovered; it is not held for a weekly or monthly meeting.
6. Close the learning loop: Tell affected workers what changed, document effectiveness checks under the organization's defined procedure, and retain any records independently needed by the applicable OSHA standard.

What to do in the first 30 days

1. Inventory the tasks that expose workers to credible fatal or life-altering harm; involve the crews doing that work.
2. Define SIF, pSIF, critical control, task observation, and verified closure in plain language, with example cases.
3. Choose two or three high-consequence task categories and identify their needed controls and safe verification methods.
4. Start a small, independently reviewed observation sample and log failures/missed checks separately.
5. Test the entire near-miss-to-corrective-action pathway with workers, including acknowledgement and feedback.
6. Review the first dashboard with operations leadership as a list of decisions and unresolved exposures, not as a contest for the highest score.

Related HazComFast resources

• How to Calculate TRIR & DART, calculations and recordkeeping distinctions.
• TRIR Calculator, practical rate calculation.
• Is a Near Miss OSHA Recordable?, internal investigation versus OSHA reporting/recording.
• Near-Miss & Corrective Action Report, reporting and corrective-action workflow.
• Incident Investigation Form, investigation structure and action tracking.
• Job Hazard Analysis Builder, task-level hazards and controls.
• Near Miss Reporting Toolbox Talk, worker reporting discussion.
• Are Safety Incentive Programs Legal Under OSHA?, reporting and incentive distortions.

Sources and research

1. OSHA, Using Leading Indicators to Improve Safety and Health Outcomes, OSHA 3970 (2019), especially introductory definitions, SMART principles, and the action-plan method: https://www.osha.gov/sites/default/files/publications/OSHA_LEADING_INDICATORS.pdf . Advisory guidance.

2. OSHA, Recommended Practices for Safety and Health Programs: Program Evaluation and Improvement: https://www.osha.gov/safety-management/program-evaluation . Advisory guidance.

3. OSHA, Injury Tracking Application Data Users Guide and OSHA Forms for Recording Work-Related Injuries and Illnesses: https://www.osha.gov/sites/default/files/ITA_data_users_guide.pdf and https://www.osha.gov/sites/default/files/OSHA-RK-Forms-Package.pdf . Official calculation and form instructions.

4. OSHA, Clarification on how the formula is used by OSHA to calculate incident rates (August 23, 2016): https://www.osha.gov/laws-regs/standardinterpretations/2016-08-23 . Agency interpretation.

5. Bayona, A., Hallowell, M. R., and Bhandari, S. (2024). The Things That Hurt People Are Not the Same as the Things That Kill People: Key Differences in the Proximal Causes of Low- and High-Severity Construction Injuries. Journal of Construction Engineering and Management, 150(8). https://doi.org/10.1061/JCEMD4.COENG-14545 . Original construction research; findings bounded by study design and population.

6. Hallowell, M. R., Alexander, D., and Gambatese, J. A. (2017). Energy-based safety risk assessment: does magnitude and intensity of energy predict injury severity? Construction Management and Economics, 35(1–2), 64–77. https://doi.org/10.1080/01446193.2016.1274418 ; publisher-hosted original full-text PDF: https://www.tandfonline.com/doi/pdf/10.1080/01446193.2016.1274418 . Original retrospective research based on 505 gravity/motion injury reports.

7. National Safety Council, Serious Incident and Fatality Prevention Model: https://www.nsc.org/workplace/sif-prevention-model ; Campbell Institute (2018), Serious Injury and Fatality Prevention: Perspectives and Practices, original PDF now hosted by the National Safety Council: https://www.nsc.org/getmedia/42c2d272-8ff3-48c8-ab03-e7ccd4b47806/campbell-wp-sif-prevention-perspectives-practices.pdf . Industry frameworks, not OSHA rules.

8. OSHA, 29 CFR 1904.7 and 1904.39: https://www.osha.gov/laws-regs/regulations/standardnumber/1904/1904.7 and https://www.osha.gov/laws-regs/regulations/standardnumber/1904/1904.39 . Regulatory text.

9. NIOSH, Hierarchy of Controls: https://www.cdc.gov/niosh/hierarchy-of-controls/about/ . Official prevention guidance.

10. OSHA, Hazard Prevention and Control: https://www.osha.gov/safety-management/hazard-prevention . Advisory program guidance; individual task standards may impose legal duties.

11. OSHA, 29 CFR 1904.35 and Clarification of OSHA's Position on Workplace Safety Incentive Programs (October 11, 2018): https://www.osha.gov/laws-regs/regulations/standardnumber/1904/1904.35 and https://www.osha.gov/laws-regs/standardinterpretations/2018-10-11 . Binding rule plus agency interpretation.

12. Yorio, P. L., and Moore, S. M. (2018). Examining Factors that Influence the Existence of Heinrich's Safety Triangle Using Site-Specific H&S Data from More than 25,000 Establishments. Risk Analysis, 38(4), 839–852. https://doi.org/10.1111/risa.12869 ; NIOSH repository: https://stacks.cdc.gov/view/cdc/210656 ; NIOSH explanation: https://archive.cdc.gov/www_cdc_gov/niosh/research-rounds/resroundsv4n2.html . Original observational mining research.

13. OSHA, 29 CFR 1910.147(d)(5)–(6) (Stored energy; verification of isolation), 1910.178(n) (Traveling), and 1910.28 (Duty to have fall protection and falling object protection): https://www.osha.gov/laws-regs/regulations/standardnumber/1910/1910.147 ; https://www.osha.gov/laws-regs/regulations/standardnumber/1910/1910.178 ; https://www.osha.gov/laws-regs/regulations/standardnumber/1910/1910.28 . Binding standards when applicable; not optional KPI guidance.

14. OSHA construction standards: 1926.501 (Duty to have fall protection), 1926.502(d)(19), (d)(21) (Post-impact and pre-use personal fall-arrest inspection), 1926.503(a)(2) (Inspection training), 1926.1425 (Keeping clear of the load), 1926.651(k) (Excavation inspections), and 1926.652(a) (Protective-system scope and exceptions): https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.501 ; https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.502 ; https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.503 ; https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.1425 ; https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.651 ; https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.652 . See also OSHA's December 18, 2003 interpretation distinguishing pre-use inspection from post-impact inspection: https://www.osha.gov/laws-regs/standardinterpretations/2003-12-18 . Binding standards and explanatory agency interpretation; apply the exact task conditions.

15. National Institute of Standards and Technology (NIST), Engineering Statistics Handbook, §7.2.4.1, Confidence Intervals: https://itl.nist.gov/div898/handbook/prc/section2/prc241.htm . Official statistical-method reference for the Wilson score interval; its assumptions need to fit the observation design.