In a 2010 Society of Petroleum Engineers (SPE) conference paper, Anderson and Denkl (Schlumberger) drew on 2.3 million risk reports from an oilfield-services HSE (health, safety, and environment) database to argue that the Heinrich accident triangle's fixed severity ratios hold only in aggregate over many combined hazards and break down for specific activities, so that injury-frequency-driven management (total recordable injury rate, TRIR) systematically neglects the rare activities that carry fatal potential (fatal accident rate, FAR). They proposed instead to manage reported incidents and observations by the severity of what could plausibly have happened (potential severity) rather than by actual outcome. The paper's acknowledged further-work list bundles four unresolved practical and methodological issues: standardizing and validating assessor-assigned potential-severity classifications across assessors, locations, and time; determining when pooled cross-industry incident data suffices for high-risk activities versus when bespoke expert intervention is required; an empirically validated economic or decision-theoretic justification for proactive, long-term control investment; and the design of incentive and compliance-assurance schemes that raise meaningful high-potential reporting while securing full compliance with top-tier controls. This problem poses the four issues as one linked open problem with a shared objective. The full text of the source paper is paywalled, and the framework's constructs are internal to the source company rather than publicly documented.

Background

The accident triangle (or accident pyramid) proposed by H. W. Heinrich in the 1930s postulates an approximately fixed numerical ratio between workplace incident tiers — for every major injury roughly 29 minor injuries and 300 no-injury incidents or near misses (events that could but did not cause harm). For over seventy years this ratio underpinned occupational safety management: driving down the broad base of minor incidents and unsafe acts was expected to also drive down the rare apex of fatal injuries. Modern practice operationalizes the base through recordable-injury statistics such as the total recordable injury rate (TRIR, the number of recordable work-related injuries per 200,000 hours worked), while the apex is tracked by the fatal accident rate (FAR). From the 2000s onward, oil and gas and other heavy-industry companies repeatedly observed a divergence: TRIR keeps declining year on year while FAR stagnates, suggesting the triangle's tiers are decoupled.

Anderson and Denkl (SPE 126661, 2010) confronted this divergence analytically. Using the 2.3-million-report HSE database of an oilfield-services company, they argued that Heinrich's numerical relationship holds only when averaged over a large, pooled set of heterogeneous hazards and activities, and fails when applied to specific activities. Fixed-ratio thinking therefore pushes organizations to over-attend to frequent, easily identified nonfatal risks at the cost of activities with fatal potential. As a replacement they advocated potential-based management: classify each reported incident or risk observation by the severity of what could plausibly have happened (its potential severity), rather than by the actual outcome, and direct management attention and controls at the high-potential tier. Their corporate framework zones risk reports by potential severity (Zones A–C, Zone A denoting the most serious potential), designates activities with high fatal potential as high-risk activities (HRAs), and subjects the top tier of mandatory risk controls (Category 1 controls) to compliance assurance. The same concerns reappear in the wider safety-science literature as the 'serious injuries and fatalities (SIF) plateau': large-scale record-based studies confirm that all-injury rates decline while serious and fatal injury rates plateau, contrary to the triangle's implication; related work finds heavy-tailed (power-law-like) accident severity distributions consistent with self-organized criticality (a statistical-physics regime in which a system self-tunes to a state with events of all sizes, with no characteristic scale), and develops exposure- or energy-based severity metrics and machine-learning predictors of SIF exposure as alternatives to injury-frequency metrics.

The specific unresolved issues recorded here are the ones the authors themselves acknowledge in closing. Potential-severity classification rests on assessor judgment with unquantified inter-assessor agreement and unknown predictive validity against subsequent serious-injury and fatality outcomes. Cross-industry pooling presupposes that a high-risk activity in one industry is sufficiently similar to its namesake in another for transferred lessons to be valid, but no formal criterion exists for when pooled data suffices versus when bespoke expert intervention is required; incident counts without common exposure denominators (hours or tasks at risk) cannot be compared across industries, and proprietary incident taxonomies hinder independent replication. Incentive and compliance schemes for hazard reporting are known to be vulnerable to gaming, where rewards for report volume degrade the quality mix of reports and can crowd out reports of high-potential conditions in favor of trivial ones. Economic evaluation of proactive controls faces the rare-event problem: fatal outcomes are sufficiently infrequent that naive before-after comparisons lack statistical power, and the counterfactual (how many fatalities would have occurred without the control) is unobservable, requiring quasi-experimental identification, reasoned bounds, or prospective rollout designs drawn from program-evaluation methodology. The source framework articulated these strategies qualitatively; the quantitative, testable versions remain undeveloped, and because the framework's constructs (the Zone A/B/C taxonomy, the Category 1 control inventory, the HRA catalogue) are internal to the company, they are not fully pinned down in any public source.

Problem Statement

The source paper closes by acknowledging four unresolved practical and methodological issues in its potential-based fatality-prevention framework. This record poses them as one linked open problem — establishing an empirically validated evidence base for potential-based proactive management of high-consequence occupational risks — with four sub-problems, each stated for real high-risk industrial settings:

  1. Classification validity. Each incident or risk report's potential severity (the severity of what could plausibly have happened, rather than what did happen) and likelihood is assigned by assessor judgment. How can potential-severity assignments be standardized and validated — shown to be reproducible across independent assessors, locations, and time, and shown to predict the serious-injury and fatality outcomes they are used to prioritize — with exposure and reporting bias controlled?

  2. Limits of pooled learning. Even when company data and cross-industry data are combined, some high-risk activities (the paper's examples are aviation-related and security activities) lack sufficient incident experience to develop robust control measures and still require expert advice. By what systematic, stated-in-advance criterion can one decide, for a given high-risk activity, when pooled cross-industry incident data (with common exposure denominators) suffices to derive control measures and when bespoke expert intervention is required?

  3. Investment justification. Sustained management commitment and resources for proactive, long-term controls on high-risk activities are hard to secure because benefits are diffuse and not demonstrable in the short term. What empirically validated economic or decision-theoretic framework — with a credible identification strategy for the causal effect of proactive controls on rare fatal outcomes, including where causal estimation is infeasible — can justify or halt investment in such controls?

  4. Implementation fidelity. How should incentive and recognition programs be designed so that they increase meaningful reporting of high-potential (Zone A/B) conditions rather than trivial (Zone C) reports, and how can complete compliance with the top tier of mandatory controls (Category 1 controls) be measured and assured under real-world operational constraints?

The four sub-problems are scientifically linked — classification validity gates the incentive and economic sub-problems, and all four serve the stated shared objective. They call respectively for: a standardized and validated potential-severity classification that is reproducible across independent assessors, locations and time and that predicts the serious-injury and fatality outcomes it is used to prioritize, with exposure and reporting bias controlled (sub-problem 1); a systematic, stated-in-advance criterion for when pooled cross-industry data suffices and when bespoke expert intervention is required (sub-problem 2); an empirically validated economic or decision-theoretic framework for proactive control investment (sub-problem 3); and an incentive-and-recognition design that raises meaningful high-potential reporting together with a scheme that measures and assures complete compliance with Category 1 controls (sub-problem 4). The linked problem is resolved only when all four of these standards are met together. The open core is the absence of answers meeting these standards for each of the four.

References

  1. Martin Anderson and Michael Denkl, 'The Heinrich Accident Triangle—Too Simplistic A Model for HSE Management in the 21st Century?', SPE International Conference on Health, Safety and Environment in Oil and Gas Exploration and Production, 2010, DOI 10.2118/126661-ms, https://doi.org/10.2118/126661-ms
  2. I. M. Threadgold, 'Reducing the Risk of Low-Probability High-Consequence Events', SPE Americas E&P Health, Safety, Security, and Environmental Conference, 2011, DOI 10.2118/141763-ms, https://doi.org/10.2118/141763-ms
  3. Linda J. Bellamy, 'Exploring the relationship between major hazard, fatal and non-fatal accidents through outcomes and causes', Safety Science 71:93–103, 2015, DOI 10.1016/j.ssci.2014.02.009, https://doi.org/10.1016/j.ssci.2014.02.009
  4. Alberto Pasquini, Simone Pozzi and Luca Save, 'A critical view of severity classification in risk assessment methods', Reliability Engineering & System Safety 96:53–63, 2011, DOI 10.1016/j.ress.2010.06.029, https://doi.org/10.1016/j.ress.2010.06.029
  5. John C. Mauro, Brett Diehl, Richard F. Marcellin and Daniel J. Vaughn, 'Workplace accidents and self-organized criticality', Physica A: Statistical Mechanics and its Applications 506:284–289, 2018, DOI 10.1016/j.physa.2018.04.064, https://doi.org/10.1016/j.physa.2018.04.064
  6. Deodete N. Ndahondyapo Pitra and Jorge Lima, 'Preventing Injury and Fatalities — Applying a Preventing Serious Injury and Fatalities Field Guide', SPE African Health, Safety, Security, Environment, and Social Responsibility Conference and Exhibition, 2016, DOI 10.2118/183606-ms, https://doi.org/10.2118/183606-ms
  7. David Bibby, 'Serious Injuries and Fatalities: A Study of Data and Prevention Strategies', SPE Health, Safety, and Environment Conference (SPE-206356-MS), 2021, DOI 10.2118/206356-ms, https://doi.org/10.2118/206356-ms
  8. Heather Jackson, 'Pathways to single fatality and serious injury incidents in coal and metalliferous mining in NSW, Australia: Can we learn from multiple fatality incidents to prevent serious injury?', Safety Science 165:106194, 2023, DOI 10.1016/j.ssci.2023.106194, https://doi.org/10.1016/j.ssci.2023.106194
  9. David Rowell, David McMillan and James Carroll, 'Offshore wind H&S: A review and analysis', Renewable and Sustainable Energy Reviews 189:113928, 2024, DOI 10.1016/j.rser.2023.113928, https://doi.org/10.1016/j.rser.2023.113928
  10. Elif Deniz Oguz Erkal, Matthew R. Hallowell, Ayoub Ghriss and Siddharth Bhandari, 'Predicting Serious Injury and Fatality Exposure Using Machine Learning in Construction Projects', Journal of Construction Engineering and Management 150, 2024, DOI 10.1061/jcemd4.coeng-13741, https://doi.org/10.1061/jcemd4.coeng-13741
  11. Matthew R. Hallowell, 'Energy-Based Safety' (monograph, CRC Press), 2025, DOI 10.1201/9781003641483, https://doi.org/10.1201/9781003641483
  12. Kailai Sun, Tianxiang Lan, Yang Miang Goh and Yueng-Hsiang Huang, 'Overcoming imbalanced safety data using extended accident triangle', Knowledge-Based Systems 327:114120, 2025 (preprint arXiv:2408.07094), DOI 10.1016/j.knosys.2025.114120, https://doi.org/10.1016/j.knosys.2025.114120
  13. Hendrik Lourens, 'Breaching the Mining Fatalities Plateau: Financial Pressure, Operational Instability, and the Case for Flow-Based Safety', SSRN working paper, 2026, DOI 10.2139/ssrn.6565978, https://doi.org/10.2139/ssrn.6565978