Williams Geismar: The Action Was Closed. Was the Hazard Controlled?

An incident-based review of why action close-out evidence needs to demonstrate protection across operating, standby and transition states.

A closed action can become an assumption in the next hazard study. That is what makes action close-out more than a document-control task: later reviewers may treat the recorded outcome as an established feature of the plant. The useful question is therefore not simply whether work was completed, but what the evidence allows the next person to conclude.

The Williams Geismar incident offers a sobering example. This article separates a short account of the published investigation from our engineering review prompts. It is educational guidance, not an operating instruction, relief-system design or approval to accept a particular facility configuration.

What the investigation established

On 13 June 2013, an explosion and fire at the Williams Geismar olefins plant killed two workers; 167 people reported injuries. The US Chemical Safety Board concluded that heat introduced to an offline reboiler caused overpressure while it was isolated from its pressure-relief device. Its October 2016 case study describes weaknesses extending across hazard analysis, change management and pre-startup review.

Section 5.2.2 records that a 2006 PHA recommendation was marked complete in January 2010, but was not implemented as intended. Only the operating reboiler’s shell-side outlet valve was car sealed open. The standby reboiler remained isolated from the relief device. Section 5.2.3 describes how the later PHA relied on the action tracking record. These are investigation findings, not a claim about any Pro Engineers project.

Our analysis: the missing link is the operating state

A safeguard description can sound convincing while leaving its operating envelope unstated. “Connected to relief” is a relationship between equipment, a flow path and a configuration. It is not established merely by the presence of a relief valve elsewhere on a drawing. The review question should identify when the claimed relationship exists and what can interrupt it.

For an action affecting standby equipment, begin by distinguishing the states that people actually use. Running, ready for service, isolated for maintenance and awaiting reinstatement can mean different physical arrangements. A label in a register should not substitute for the marked-up drawing, approved procedure and field evidence that define the state. Ask operations and engineering to resolve differences in terminology before accepting a protection claim.

This is our proposed analytical lens, rather than a new finding about Geismar. It shifts the close-out discussion from “Was the specified item installed?” to “Which hazardous scenario does the installed arrangement control, in which states, and on what evidence?” Both questions matter, but they require different evidence.

Three claims that need different evidence

An action response often contains three separate claims. First, the proposed solution is technically suitable. Second, the agreed solution has been implemented. Third, the protection remains available under the conditions in which it is credited. A single attachment rarely demonstrates all three.

A calculation may support technical suitability within its assumptions. An installation record may establish what was fitted. A configuration check may show the arrangement at a particular time. The reviewer needs to connect these records and identify their limits, including which modes were excluded. None should silently inherit a broader meaning than its author intended.

Consider a fictional action concerning equipment that can be placed on standby. A response might identify an approved assessment and a completed modification. Before accepting it, the reviewer would still ask whether the assessment included standby, whether the implemented arrangement matches that case, and how future changes to its dependencies are controlled. This example does not prescribe a particular safeguard or valve position.

Build a state-and-evidence review

Use the following matrix as a discussion aid. It is an original review framework, not a reconstruction of the Williams plant or an acceptance standard. Complete it with the responsible site specialists; unresolved cells are questions to resolve through the facility’s governance.

For each relevant state, identify the equipment boundary, credible energy and material inputs, the protection being claimed, the evidence supporting it, and who maintains that evidence. Include transitions between states where appropriate. The act of moving equipment into or out of service may involve conditions that are absent from the steady-state description.

Avoid treating every blank cell as the same kind of problem. A missing document reference, an unverified installed arrangement and an unresolved engineering question have different implications. Describe the exact uncertainty and the decision it prevents. That gives the responsible authority something concrete to review instead of another broad instruction to “check safety”.

Original review matrix: protection claims by equipment state
StateClaim to examineEvidence question
RunningProtection available for assessed dutyDoes the accepted basis match the installed arrangement?
StandbyHazards remain controlled while offlineWhich inputs and protection dependencies were assessed?
TransitionProtection remains suitable during change of stateAre intermediate conditions covered?
Discussion aid only. No valve positions or operating sequence are prescribed.

Close the loop into the next study

The close-out package should be useful to someone who did not attend the original workshop. Preserve the original concern, explain the accepted response and state the limits of the conclusion. Link superseded and replacement actions so a transfer between registers does not erase the unresolved question.

When a later HAZOP or Bowtie review credits the resulting protection, retrieve the supporting basis instead of relying only on the status colour. Sampling selected closed actions can test whether the register leads to meaningful evidence. Select samples for their risk significance and complexity, not simply because their attachments are easy to find.

Our existing HAZOP action close-out checklist covers the structure of the record. The additional question here is whether that record demonstrates the claimed protection across the relevant equipment states. A tidy register and a sound technical conclusion are related, but neither guarantees the other.

Questions for the next review or walkdown

Ask which operating state was assumed when the action was written, and whether the accepted response uses the same state. Ask which physical or procedural dependency makes the protection available, and where that dependency is recorded. Ask what evidence demonstrates implementation, who accepted its adequacy, and what would trigger reassessment.

For a walkdown, agree the scope and evidence to be collected through the site’s authorised process. Do not manipulate equipment to demonstrate a claim. Where drawings and field observations disagree, record the discrepancy and refer it to the responsible authority. A photograph can identify a component, but its suitability, internal condition and performance may require other evidence.

Where the response changes a safeguard or its operating assumptions, connect the review to management of change and hazard-study review. Where overpressure protection is involved, the pressure-relief evidence register provides a complementary way to organise the design basis. Neither checklist replaces competent site-specific engineering.

A more useful completion statement

A strong completion statement explains the bounded result: the equipment and modes assessed, the solution accepted, the implementation evidence reviewed and any continuing dependencies. It should also make unresolved matters visible. This gives subsequent reviewers a basis they can challenge and maintain, rather than an unexplained green status.

For organisations planning a study, Pro Engineers’ HAZOP services can help define the review scope and required inputs. The practical lesson is to make the protection claim reviewable before marking the action complete, and to preserve the evidence needed to revisit that claim when the plant changes.

Source and scope

Primary source: CSB, Williams Geismar Olefins Plant: Reboiler Rupture and Fire, October 2016, particularly sections 1, 4.1 and 5.2.2–5.2.3. Source checked 19 November 2026. The review framework and questions above are our analysis. US investigation findings are not presented as Australian legal requirements.