Rana Plaza made factory inspection and buyer remediation a supply-chain safety accountability test

Summary

• On 23 April 2013, visible cracks were reported in Rana Plaza in Savar, near Dhaka. The next morning, workers in garment factories returned to the building and production resumed. Bangladesh Sangbad Sangstha's 2026 account of the pending prosecution says the building collapsed at about 9:45 a.m., shortly after generators were switched on. That sequence is part of the prosecution record summarized by the state news agency; because the murder trial was still unfinished in 2026, it is not a final judicial finding about individual guilt or the complete engineering mechanism.
• Public sources disagree on the final death count, using figures from 1,129 to 1,138, and they apply different injury totals. The defensible common statement is that more than 1,100 people died and thousands were injured, most of them garment workers. Numerical variation should not be solved by silently choosing the largest or most familiar figure.
• The triggering event was the return to production in a visibly distressed building. The physical failure arose from inadequate structural capacity in an altered building under industrial occupancy and load, but the public records reviewed here do not provide a complete, final component-by-component forensic report. Generator vibration, additional floors, unsuitable use, construction quality, and load are described in official or judicial records at different levels of certainty; none should be isolated as the sole proven cause.
• The deeper control failure was organizational. Building and land-use approval, structural alteration, factory licensing, production authority, worker protection, purchasing leverage, and inspection scope were divided among actors. No control reliably converted a credible hazard into evacuation, income protection, independent structural review, and permission-based reopening. Detection occurred. Escalation and enforcement failed.
• Pre-collapse social audits did not necessarily answer the structural question. In the Canadian Das litigation, the court record said Bureau Veritas social audits for Loblaw were not expressly commissioned to assess structural integrity and that social auditors were not structural engineers. The courts dismissed the proposed claims on applicable-law, limitation, and duty grounds. Those holdings do not prove that the building was safe, that every audit was adequate for its stated scope, or that purchasing power was operationally irrelevant.
• A separate Danish National Contact Point proceeding reached a different kind of conclusion. It found that PWT Group had not demonstrated due-diligence processes consistent with the OECD Guidelines for its Rana Plaza supplier and recommended stronger risk and decision systems. That was a nonjudicial responsible-business-conduct determination, not a damages judgment or a finding that PWT caused the collapse.
• The 2013 Accord on Fire and Building Safety in Bangladesh changed the buyer control architecture. It required independent inspections, public reports, time-bound corrective action, worker income protection during certain closures, a protected right to refuse unsafe work, complaints, supplier disclosure, commercial consequences for noncompliance, and buyer action to make remediation financially feasible. These obligations were more concrete than a general supplier code, but they applied through signatories and covered suppliers; they did not replace Bangladeshi law or public enforcement.
• The national and buyer-led inspection programs assessed 3,780 export-oriented garment factories by the end of 2015. The Department of Inspection for Factories and Establishments later reconciled that total as 1,505 Accord assessments, 890 Alliance assessments, 1,549 National Initiative assessments, and 164 factories jointly assessed by Accord and Alliance. Inspection was the opening inventory of hazards, not proof of repair.
• Compensation also required a new mechanism. The Rana Plaza Arrangement used an ILO-convention-based approach, received 2,895 claims relating to injured workers and dependants of deceased or missing workers, and reached the US$30 million estimated requirement for final awards in 2015; the ILO's later retrospective describes more than US$34 million disbursed by October 2015. A separate Primark route handled 630 claims connected to New Wave Bottoms. These payments provided material relief; they did not adjudicate each actor's legal fault or measure the full lifetime social cost.
• Reform remains measurable but unfinished. The International Accord's maintained Bangladesh page reported 243 global brands and retailers, 1,773 factories, 2.9 million workers, 74,535 inspections, 285,453 safety findings, and average remediation progress of 81 percent as of 30 June 2026. Those are program figures, not a national census or a certificate that 81 percent of all factories are safe. A completion claim requires factory-level closure of critical findings, stable coverage, worker confirmation, and public-enforcement capacity outside buyer programs.
• Legal accountability remains open. As of May 2026, the Rana Plaza murder case had been moved to another trial court, 150 of 594 listed witnesses had testified, and no verdict had resolved the charges. Charges and prosecution allegations must not be written as convictions. The long delay is itself an accountability fact, but it does not authorize a presumption of guilt.

The warning was detected before the collapse

Before 23 April: industrial production occupied a building-control gap

Rana Plaza was an eight-storey mixed-use building that contained several garment factories as well as other occupants. Sources vary between four and five when describing the factories relevant to particular proceedings. The Bangladesh Supreme Court's later account of the institutional response says five ready-made-garment factories were housed there; the 2026 prosecution update refers to workers of four factories allegedly being forced to return. The difference appears to reflect the scope of each record, not a basis for changing the scale of the event.

The building was not simply a neutral container rented by unrelated businesses. Garment production brought concentrated workers, sewing equipment, stored material, electrical demand, and backup power into a structure whose lawful configuration and capacity had to be established. A Bangladesh government trade study, while expressly describing its account as early analysis, reported code concerns, upper floors added beyond the original arrangement, industrial loads, and generators as relevant conditions.

The Bangladesh Diagnostic Trade Integration Study is useful for that policy record, but it is not substituted here for a final structural calculation.

Responsibility followed control. The building owner controlled alterations and leases. Factory owners and managers controlled production and immediate workplace entry. Planning, building, fire, and labour bodies controlled different permissions and enforcement powers. Buyers controlled purchase orders, supplier acceptance, commercial terms, and the audit requirements attached to sourcing. Auditors controlled only the work they were retained and qualified to perform. Workers had the earliest direct exposure to visible conditions but the least economic power to make closure endure.

That division did not make responsibility unknowable. It meant a safe system needed explicit handoffs: a building approval tied to actual industrial use; a factory licence tied to valid occupancy and structural records; a hazard report tied to an authority able to stop production; income protection during evacuation; an independent engineering decision; and a documented reopening approval communicated to workers and buyers. The collapse exposed the absence or ineffectiveness of those links.

23 April: a crack became a control test

The strongest common fact is that cracks were visible on 23 April. The Bangladesh state news agency's current case summary says visible cracks appeared that day. Worker and witness accounts collected after the event describe alarm, interruption, and pressure surrounding the next day's return. Human Rights Watch reported that the building had been evacuated and that factory managers then persuaded, cajoled, or directed workers back; its worker-rights investigation also found that none of the five factories had a union. Those accounts are evidence of worker experience and coercive conditions, not engineering measurements.

A crack is an observation, not by itself a diagnosis. It can be cosmetic, service-related, or a sign of serious distress. The correct response in a crowded industrial building was therefore not for an unqualified manager to declare the structure sound or for workers to decide the engineering question individually. It was to preserve evacuation, control access, reduce load, retain qualified structural expertise, examine drawings and alterations, and reopen only after a competent authority documented why occupancy was safe.

The accountability failure lies in what happened after detection. Shops and a bank were widely reported as remaining closed while garment production resumed. Even without using that contrast to infer anyone's state of mind, it shows that the warning produced different operational decisions inside one structure. A safety signal that closes one workplace but not another is not a completed control. The system lacked one authoritative status for the building and one enforceable consequence for all occupants.

Worker refusal also cannot be judged as a free consumer choice. A worker deciding whether to enter faced wages, attendance rules, supervisor power, production targets, and the possibility of dismissal. The ILO's post-collapse statement emphasized that workers who see hazards need a voice that does not put jobs, income, or safety at risk. The ILO's 2014 assessment correctly connects inspection with protected worker voice: either one without the other leaves a gap.

24 April: return to production became the immediate trigger

The public chronology supports a conservative statement. Workers returned on the morning of 24 April, production started, and the building collapsed during the morning work period. The 2026 prosecution account says the collapse followed the switching on of generators. That timing can make power equipment part of the immediate sequence without proving that generator vibration alone caused the building to fail. A sound structure approved for its actual use should not depend on workers recognizing which normal production action might be the last load increment.

The trigger was therefore operational: reoccupation and restart while the building's safety was unresolved. The physical failure was structural. The root accountability problem was the failure to make a detected warning binding across the people who could close, inspect, finance, and reopen the workplace.

More than 1,100 people died. The Bangladesh Supreme Court used 1,129 deaths and almost 2,000 injuries in Writ Petition No. 8716 of 2017. ILO publications commonly use more than 1,130 deaths and more than 2,500 injuries. BSS has used 1,134 or 1,136 in different updates. These records were produced at different times and for different purposes. This article uses the shared lower-resolution description unless a source-specific count is necessary.

Rescue and recovery continued after the structure fell, and survivors faced amputations, long rehabilitation, lost earning capacity, trauma, and uncertain employment. Those consequences matter because they define the performance required from compensation. A payment system that counts claims but cannot sustain medical care or replace lost income does not close the event merely because its fund is disbursed.

What is confirmed, alleged, inferred, and still unknown

Confirmed across authoritative records. The collapse occurred on 24 April 2013 after cracks had been seen the previous day. Garment workers had returned to the building. The event killed more than 1,100 people and injured thousands. It led to criminal proceedings, a coordinated compensation arrangement, the Accord, the Alliance, national inspections, and major changes in labour-inspection capacity.

Alleged in the pending criminal case. The prosecution alleges that workers were forced back despite known risk and attributes criminal responsibility to named defendants. Its case summary also places generator startup shortly before collapse. Those allegations are serious and evidentially specific, but the trial remained unfinished as of the article date. The accused are not described here as convicted of causing the deaths.

Supported institutional inference. A visibly distressed industrial building should have remained empty until a qualified structural review and lawful reopening. Production authority, employment power, purchasing leverage, and public enforcement were capable of preserving that state. Their collective failure to do so is supported by the sequence even where the record does not show every conversation or order.

Bounded physical inference. Alteration, occupancy, loads, construction, and dynamic effects are relevant to structural adequacy. The reviewed public record does not contain enough authenticated design drawings, material tests, load calculations, foundation analysis, alteration approvals, and collapse reconstruction to rank every physical contribution. A policy report's early analysis and a prosecutor's theory are not transformed into a final forensic report by repetition.

Unknown or unresolved. The record does not provide a single reconciled account of every permit, inspection, factory licence, audit finding, buyer order, subcontracting route, warning communication, attendance threat, and reopening decision. Nor does it resolve every victim's employment identity, exposure, medical prognosis, or lifetime loss. Those absences limit person-level claims but strengthen the case for traceable institutional records.

Failure classification: detection was not the missing step

Rana Plaza is often described as an inspection failure. That is true only if inspection is understood as a complete control cycle rather than a visit or checklist.

Physical failure. The building lost structural capacity under occupied industrial use. The exact progression through columns, slabs, foundations, connections, or other elements cannot be reconstructed from the public sources used here with component-level certainty.

Triggering event. Workers re-entered and production restarted before safety had been independently established. Starting machinery and backup power was part of the reported immediate sequence, but not a complete root cause.

Contributing conditions. Building alteration, industrial occupancy, equipment and stored loads, backup-power installation, fragmented permits, weak inspection capacity, factory production pressure, worker economic vulnerability, limited collective representation, narrow audit scopes, and buyer-supplier purchasing relationships all affected risk or the ability to control it. Presence in this list does not assign equal causal weight.

Detection failure. The crack itself was detected. The failure was in classification, escalation, and authority: the warning did not generate one controlled building status, qualified review, access restriction, and documented release. A system can detect a hazard and still fail at detection if it does not identify who must act or preserve the evidence needed for action.

Response failure. The evacuation did not hold. Workers returned while the status remained contested or unsupported. Income and job protection were limited public evidence to make refusal practical. A closure rule without wage continuity can push risk back onto the least powerful person.

Recovery failure. Rescue was followed by years of compensation, medical care, litigation, criminal proceedings, and sector repair. Recovery cannot be measured only by the end of rescue operations or final distribution from one fund. As of 2026, the principal criminal trial remained unresolved, and sector remediation data still recorded open findings.

Root cause / governance control failure. No integrated control bound actual building use, public permission, employer production decisions, worker voice, buyer leverage, engineering inspection, repair finance, and reopening proof. The root cause for accountability purposes was not the mere existence of a crack or the later presence of an inspection program; it was the absence of an enforceable chain that could convert a detected building warning into protected stop-work, competent structural review, financed repair, verified closure, and lawful reopening.

The post-collapse reforms are best evaluated by whether they created that chain, not by how many commitments were announced.

Why a social audit was not a structural safety case

The audit issue is clearest in the Canadian litigation brought by survivors and relatives against Loblaw entities and Bureau Veritas. The Ontario Superior Court's 2017 decision in Das v. George Weston Limited records that Bureau Veritas performed social audits at New Wave Style in February 2011 and April 2012. The first identified multiple noncompliances, including health-and-safety items. The agreement did not expressly require an investigation of Rana Plaza's structural integrity, and evidence before the court said social auditors were not trained as structural engineers.

Intertek was scheduled to conduct a later audit on 24 April 2013, the day of collapse.

Those facts establish a scope mismatch, not a universal excuse. An audit can accurately test wages, hours, documents, exits, sanitation, or other code items and still say nothing reliable about a building's load path. If a buyer communicates that a supplier has been audited without stating that boundary, the assurance can be misunderstood by managers, consumers, and workers. Conversely, asking a social auditor to improvise a structural opinion outside professional competence would create another unsafe control.

The correct design is layered. A social audit should identify the legal entity, site, occupancy, worker count, subcontracting status, obvious red flags, and required specialist records. It should have a mandatory escalation route when it sees cracks, alterations, heavy rooftop or floor equipment, missing permits, or conflicting addresses. A structural assessment should then be performed by qualified engineers against actual drawings, measured geometry, material evidence, loads, and use. Neither audit should be represented as the other.

The Das claims were dismissed. The motion judge held that Bangladeshi law applied, that most claims were time-barred, and that the pleaded duties could not succeed; the Ontario Court of Appeal affirmed. The courts also distinguished Loblaw's purchasing influence from ownership or workplace control. These are legal holdings in a particular claim. They do not establish that supplier codes create no expectations, that purchasing terms have no effect on safety, or that every future statute will allocate duty in the same way.

The record instead exposes three different meanings of control. A buyer may lack title to the building and power to hire or fire supplier workers. It may still control whether orders continue, which factories are approved, what evidence is required, and whether commercial terms make repairs feasible. Operational accountability follows that practical leverage even when a court does not recognize the tort duty pleaded by a particular class.

The OECD proceeding concerning PWT Group shows the distinction. The Danish National Contact Point case record says PWT had inspected New Wave Style in 2012, but the documents supplied did not show the checks performed, improvements required, or risk and decision systems used. The NCP found that PWT had not applied due-diligence processes consistent with the OECD Guidelines and recommended stronger risk assessment, supplier review, communication, and code implementation.

That conclusion was not a court judgment. It did not award compensation or determine that PWT's conduct caused structural failure. It did establish a documentary accountability standard: a buyer cannot demonstrate due diligence merely by proving that someone visited a supplier. It must show what risk universe was considered, what evidence was examined, what decision followed, who owned correction, and whether the result was verified.

The 2013 Accord converted buyer leverage into specified duties

The 2013 Accord on Fire and Building Safety in Bangladesh was signed within weeks of the collapse. Its importance was not the word "binding" alone. It assigned previously diffuse tasks to named actors and attached records and consequences to them.

First, it required a qualified safety inspector independent of companies, unions, and factories. Covered factories were to receive credible fire and building inspections by skilled personnel, with an initial inspection targeted within two years. A buyer's pre-existing inspection could count only if the chief inspector considered it sufficiently thorough and received full access to its findings. Every covered factory still had to receive at least one inspection under the program's direction.

Second, inspection results had to travel. Reports were to be prepared promptly, shared with factory management, safety committees, worker representatives, signatory companies, and the governing committee, then publicly disclosed with remediation plans. If the inspector identified severe and imminent danger, the agreement required immediate notice and a directed repair plan. This answered the Rana Plaza handoff problem: a hazard could not remain only in a consultant's private file or a buyer's supplier portal.

Third, remediation was time-bound. Signatory companies had to require suppliers to implement corrective actions. During qualifying renovation closures, the agreement required maintenance of workers' employment relationship and regular income for up to six months. It also protected the right to refuse work reasonably believed unsafe without discrimination or loss of pay. These provisions recognized that evacuation is ineffective if workers must personally finance it.

Fourth, the agreement gave workers an operating role. It required safety committees with at least half worker representation, regular training, and a confidential complaint route. The point was not to make workers unpaid inspectors. It was to create a protected signal path and give the people continuously present in a factory access to an authority outside line management.

Fifth, the Accord linked buyers' commercial leverage to repair. Signatories had to require supplier participation, escalate from notice and warning toward termination when cooperation failed, and negotiate terms that made safety upgrades financially feasible for Tier 1 and Tier 2 suppliers. Available methods included joint investment, loans, donor or government support, business incentives, or direct payment. The obligation was not a guarantee that every supplier would receive a grant, but it prevented a brand from demanding expensive structural work while treating price, order duration, and cash flow as unrelated.

Finally, the agreement created enforcement between its signatories through a dispute route that could end in binding arbitration. That made the buyer commitments more than a discretionary corporate statement. It still did not convert the Accord into a building department, criminal court, or universal compensation system.

Coverage was a central limit. The Accord reached factories supplying signatory companies according to disclosed tiers. A factory outside that network, an undeclared subcontractor, a supplier that lost all signatory orders, or a new establishment still depended on public enforcement and sector-wide rules. Buyer-based governance can concentrate resources where international brands source while leaving other workplaces with weaker oversight. Durable reform therefore required the national initiative and a capable state inspectorate.

Accord, Alliance, and the National Initiative produced one hazard inventory

The post-collapse inspection effort was divided among three programs. The ILO says the programs completed assessments of 3,780 export-oriented factories by 31 December 2015 and that 39 factories were closed for posing immediate danger. Its progress account also makes the essential qualification: no factory could be considered safe until remediation was successfully completed.

The Bangladesh Department of Inspection for Factories and Establishments later supplied a useful reconciliation. Its Labour Inspection Report 2021-2022 attributes 1,505 factories to the Accord, 890 to the Alliance, and 1,549 to the National Initiative, with 164 assessed jointly by Accord and Alliance. Subtracting the overlap produces 3,780. This is more defensible than adding program headlines that use overlapping populations.

The Alliance for Bangladesh Worker Safety was a separate buyer initiative concentrated among North American companies. Its final 2018 report reported 93 percent remediation of identified items and 428 factories completing all material items in their initial corrective action plans. Those are the Alliance's own program results. They demonstrate substantial activity, not independent proof that every listed factory remained safe after the initiative ended or that uncovered factories achieved the same result.

The National Initiative covered factories outside the two buyer programs. Inspection reports had to become corrective action plans, technical review, finance, follow-up visits, and enforcement. The government established the Remediation Coordination Cell in 2017 for that task. The ILO's RCC launch record says the cell initially worked with 1,293 factories after closures and movement between programs. The number was not fixed because factories opened, closed, or changed coverage.

This movement makes a permanent identifier essential. A factory can change name, owner, address, building, buyer, or initiative without eliminating a hazard. Closure in one commercial list must not become apparent remediation if the same production resumes under another identity. The evidence chain needs legal entity, site coordinates, building identifier, licence, buyer declarations, inspection reports, corrective actions, follow-up results, and disposition history.

Public authority could be assisted, but not outsourced away

Bangladesh's Supreme Court addressed this boundary in litigation over an Accord inspection. In Writ Petition No. 8716 of 2017, the court said responsibility for factory safety inspection vested in DIFE under the Labour Act. It also found that Accord, acting with DIFE's consent and assisting it after inspecting more than 1,500 factories, performed de facto functions connected with the affairs of the republic. The holding made the private program reviewable in that constitutional dispute.

That judgment is not a general declaration that every Accord decision is a government act or that the government transferred all statutory duties. It is evidence that the post-Rana Plaza system became institutionally hybrid. Public authorities retained legal closure and licensing powers; buyer programs supplied engineers, disclosure, finance leverage, and follow-up capacity; courts could review disputes about how those powers affected factories.

A related Supreme Court decision illustrates why handoffs mattered. In Writ Petition No. 10929 of 2015, a supplier disputed its treatment after structural concerns and buyer action. The record includes competing claims, engineering assessments, requests to stop production, and arguments over whether Accord had to inspect. The court directed an inspection and publication. It should not be used to generalize the disputed engineering facts to all factories. It does show that inspection status, commercial delisting, review rights, and public authority could collide unless procedures were explicit.

National capacity expanded. The ILO's ten-year review says DIFE inspector numbers rose from 92 to 393 and Fire Service inspectors from 55 to 365, and that an Industrial Safety Unit was established in 2022. The Labour Inspection Management Application, launched in 2018, was designed to connect factory records, inspection activity, accident notices, complaints, and National Initiative remediation; the ILO launch account describes the intended functions.

Staff, software, and organizational status are inputs. Their effectiveness requires evidence of risk-based visits, competent structural review, unannounced access, notice-to-correction time, sanctions, repeat findings, complaint protection, and accurate public records. A database can preserve accountability or merely digitize incomplete information.

An independent evaluation commissioned by the U.S. Department of Labor found both progress and limits. The fire and building safety project evaluation reported that economic consequences improved compliance in Accord factories, while worker representatives still described threats for raising hazards and DIFE capacity remained a challenge for National Initiative factories. Its findings came from a defined project and period; they are not a current audit of every regulator or factory.

They support a durable lesson: training and reporting work best when backed by protected voice, inspection capacity, and consequences that management cannot ignore.

Remediation finance is part of the safety control

Finding a deficient column, exit, alarm, electrical system, or fire door does not repair it. Structural retrofit can require design work, permits, temporary closure, equipment movement, construction, testing, and lost production. Small and medium suppliers may face costs that exceed available cash while buyers continue to demand price and delivery performance.

The Accord treated commercial terms as part of safety because buyers benefited from the production and possessed leverage over order volume, duration, prepayment, and supplier eligibility. Factory owners still owned the obligation to maintain a safe workplace and implement approved repairs. Buyer finance did not transfer that duty; it addressed the foreseeable claim that mandatory work was economically impossible under the purchasing relationship.

The later Accord structure retained that principle. Its 2023 annual report says finance planning accompanies corrective action plans and describes escalation when a supplier says remediation is not financially feasible. It also records that 90 percent of original findings had been addressed in the covered population in 2023 while only 32 percent of factories had completed their initial plans. Significant items, including safe exits, fire alarms, and structural integrity, remained unresolved in multiple factories.

Those two percentages are not contradictory. A factory can close many low-cost findings while one expensive life-safety item remains open. An average weighted by individual findings can look strong even when many factories lack complete closure. Safety reporting should therefore show both finding-level progress and factory-level completion, with critical items separately visible.

Finance claims also require proof. A supplier request should identify the approved repair, bids, cash need, existing buyer relationships, order forecasts, and proposed contribution. Buyers should disclose the mechanism selected and whether it preserved wages. The engineer should verify the finished work independently of the payment decision. A loan offered but unaffordable, an order promise later withdrawn, or equipment purchased but not installed is not completed remediation.

Compensation delivered awards without resolving every liability question

The collapse immediately created needs that ordinary litigation could not meet quickly: burial support, medical treatment, wage loss, disability, dependency, identification, and long-term care. The Rana Plaza Coordination Committee brought together the government, employers, workers' organizations, brands, unions, and civil society, with the ILO as neutral chair and trust-fund trustee.

The ILO technical account of the compensation scheme describes a claims process based on principles in ILO Convention No. 121 for employment injury. The later ILO program summary reports 2,895 claims: 720 concerning deceased workers, 2,027 injured workers, and 148 missing workers, relating in total to more than 5,171 injured workers and dependants of deceased or missing workers. A separate Primark process handled 630 claims associated with New Wave Bottoms.

In June 2015, the ILO announced that the fund had reached the US$30 million required for final awards. The ILO's later ten-year review says final payments were completed by October 2015 and that more than US$34 million had been disbursed. These figures answer different accounting questions: the first was the estimated funding target for awards, and the second was the later disbursement total. Neither amount should be confused with emergency aid, separate brand programs, medical services, litigation claims, or the total lifetime cost of death and disability.

Terminology itself was contested. The ILO later explained that the committee used "award" because Bangladesh labour law reserved "compensation" for payment by a direct employer. Some worker advocates and lawyers argued that donor-funded awards did not discharge employers' or other actors' compensation obligations. The ILO's eighth-anniversary review says the claims administration closed in 2015 and that the Trust for Injured Workers' Medical Care continued lifelong medical support.

Both propositions can be true: the Arrangement delivered a coordinated, rule-based benefit to thousands of people, and it did not adjudicate all legal liability or guarantee that every lifetime loss was fully met. Accountability reporting should state who contributed, under what formula, which population was covered, what deductions or offsets applied, what appeals were available, and which medical and dependency obligations continued after payment.

Brand statements are evidence of the issuer's actions, not independent settlement audits. Primark says New Wave Bottoms was its supplier and that it provided direct support in addition to a US$1.25 million trust-fund contribution. That record can establish Primark's stated commitment and payment categories. It cannot by itself verify every beneficiary outcome or establish the responsibility of other buyers.

The compensation architecture also revealed a missing national control. A mass-casualty workplace should not require an improvised international trust fund before injured workers and dependants receive predictable income protection. The ILO has repeatedly linked Rana Plaza to the need for a national employment-injury protection system. A durable system would specify contributions, eligibility, medical care, disability assessment, survivor benefits, appeals, funding security, and treatment of subcontracted or undocumented workers before an incident.

Buyer responsibility and legal liability are not synonyms

Rana Plaza produced several accountability forums with different questions.

The Bangladesh criminal proceedings ask whether charged persons committed offences under Bangladeshi law. The current record does not contain a verdict in the principal murder case. As of 6 May 2026, BSS reported that the case had been transferred to the Eighth Additional District Judge's Court in October 2025, 150 of 594 listed witnesses had testified, two accused had died, 13 were absent, 25 were on bail, and the building owner remained in custody. These are procedural facts and prosecution statements, not proof of guilt.

The Das courts asked whether the pleaded class claims against Canadian buyer entities and an audit company could proceed under the applicable law. Their dismissal did not approve the building, determine the criminal case, or review the later Accord's performance. It demonstrated how corporate separation, audit scope, choice of law, limitation periods, proximity, and control can block a civil remedy even where purchasing links are real.

The Danish NCP asked whether PWT's due diligence met the OECD Guidelines. Its conclusion carried reputational and procedural consequences but not the coercive remedy of a damages judgment. The Accord asked whether signatory companies and suppliers met contractual safety obligations. DIFE and other authorities asked whether factories met public law. The compensation committee calculated awards without waiting for all those disputes.

A careful article must not merge the forums. A company can fail a due-diligence standard without being adjudged to have caused a collapse. A buyer can meet a narrow audit contract while its broader sourcing controls remain inadequate. A factory can complete a corrective plan without resolving historic compensation. A criminal charge can remain pending while sector-wide prevention improves.

Operational accountability asks a different but necessary question: who could have changed the outcome? Building owners could preserve lawful structure and use. Factory owners could stop production and protect wages. Public authorities could control permits, licences, closure, and reopening. Buyers could select suppliers, demand specialist evidence, stabilize orders, finance repair, and withdraw business after noncompliance. Auditors could state scope and escalate red flags. Workers and unions could report hazards if protected. Each duty must be proportional to actual authority, not proximity to the public narrative.

The 2026 numbers show reach and unfinished work, not a final verdict

The International Accord's Bangladesh program page reported, as of 30 June 2026, 243 global brands and retailers, 2.9 million workers, 1,773 factories, 74,535 inspections, 285,453 safety findings, and 81 percent average remediation progress. The page also links factory-level inspection reports, corrective action plans, training status, and remediation progress.

These figures are valuable because they preserve a continuing buyer-linked safety system thirteen years after Rana Plaza. They also require four cautions.

First, the coverage is not all workplaces in Bangladesh. It is a defined supplier population linked to participating companies and program rules. National safety performance requires DIFE and other authorities to cover factories outside it, including domestic production, new entrants, and facilities no longer supplying signatories.

Second, a finding is not a factory. The average can change when new factories enter, extensions are added, inspections identify new findings, or definitions change. The 81 percent figure published in 2026 should not be treated as directly comparable with the 90 percent figure in the 2023 annual report without a reconciled denominator and method.

Third, incomplete does not have one severity. An open documentation item and an unresolved structural or fire-egress hazard should not carry the same public meaning. Reports need criticality, due date, interim protection, and closure evidence.

Fourth, completion at one date is not permanent. Buildings change, equipment moves, exits become obstructed, occupancy rises, extensions are built, and suppliers change names. Recurring inspection, complaints, change approval, and sanctions are required after the initial plan closes.

The RMG Sustainability Council took over Bangladesh-based Accord operations in 2020. Its public factory directory defines completed status as verification that all issues identified in initial structural, electrical, and fire inspections were successfully remediated. The directory provides unusually granular evidence, but even "initial CAP completed" is bounded to initial findings. New hazards and operational practices remain subject to follow-up.

The test is not whether post-2013 Bangladesh became safer in any respect. The scale of engineering inspections, closures, repairs, training, committees, complaints, and public data supports the conclusion that major controls improved. The harder test is whether those controls are durable, reach beyond export buyers, remain independent under commercial pressure, and can stop production before another visible warning is negotiated away.

Accountability matrix

Control area	Primary practical controller	Proof required	Failure signal
Lawful building use and alteration	Owner, planning and building authorities	Approved drawings, occupancy, alteration history, completion and use certificates	Extra floors, changed use, or heavy equipment without reviewed capacity
Factory operation	Factory owner and management	Valid licence, emergency authority, stop-work procedure, wage protection	Production continues after unresolved structural warning
Structural assurance	Qualified engineers and approving authority	Site measurements, calculations, material evidence, signed review, independent check	Generic audit substitutes for engineering assessment
Worker warning and refusal	Employer, unions, safety committee, complaint body	Protected reports, no retaliation, response time, paid evacuation	Workers fear lost wages or dismissal for refusing entry
Buyer sourcing	Brands and retailers	Complete supplier map, specialist-risk rules, order and exit records	Undeclared site, narrow audit presented as broad safety assurance
Remediation finance	Factory and responsible buyers	Costed plan, agreed commercial support, wage continuity, payment and installation proof	Repair delayed because price, cash flow, or order duration is unresolved
Remediation closure	Engineer, program, factory, public authority	Follow-up inspection, test results, critical-item closure, lawful reopening	High aggregate percentage with critical findings still open
Public enforcement	DIFE and relevant building and fire bodies	Risk-based inspections, sanctions, complaint protection, current registry	Coverage depends on buyer membership or records cannot be reconciled
Compensation and care	Employers, statutory system, fund administrators, medical providers	Beneficiary denominator, awards, appeals, long-term care and dependency tracking	Fund closes while medical or income loss remains unsupported
Legal resolution	Investigators, prosecutors, courts	Timely evidence preservation, fair trial, reasoned judgments	Charges remain unresolved for years or allegations are reported as verdicts

What durable closure would require

1. One factory and building identity

Every production site needs a persistent identifier linking legal owner, occupier, factory entity, brands, building address, coordinates, floors, extensions, worker count, licences, fire approval, utility loads, boilers, generators, and inspection history. Name changes and buyer exits must preserve the record. Public and buyer databases should reconcile rather than create parallel identities.

2. Change control tied to actual use

Approved capacity must be recalculated when floors, mezzanines, generators, water tanks, machinery, fabric storage, occupancy, or production processes change. The submission should include drawings, measured conditions, material assumptions, loads, foundation implications, and independent checking. A licence renewal should fail if the structure cannot be connected to its approved and current use.

3. A stop-work state that cannot be informally reversed

A credible structural warning should create one timestamped status visible to management, workers, buyers, DIFE, fire authorities, and the building regulator. Access should be physically controlled. Reopening should require a named competent authority, defined evidence, and written release. Workers should receive pay during the protected period and have a complaint route if pressured to return.

4. Inspection scope that is impossible to misunderstand

Every report should state whether it covers labour conditions, fire safety, electrical safety, structural capacity, boilers, or another subject; the qualifications of assessors; documents reviewed; areas inaccessible; assumptions; severity; and next action. A social-compliance visit should never be displayed as a structural certificate. Visible red flags should trigger specialist review even when they sit outside the original checklist.

5. Repair evidence, not only a corrective plan

For each finding, the record should show the hazard, interim control, responsible party, design, finance source, due date, permits, contractor, photographs, test evidence, follow-up inspection, verifier, and closure basis. Critical findings should have public status and escalation. Closure should be revoked when later changes invalidate it.

6. Buyer terms that support the required result

Brands should disclose how purchasing practices affect remediation: order continuity, lead times, price, prepayment, loans, direct support, and responsibility when several buyers share a factory. Exiting a supplier may be necessary when management refuses safety work, but an unplanned exit can remove the finance needed for repair and displace workers. The decision and worker protections should be documented.

7. Independent worker evidence

Safety committees, unions, and complaint handlers should be able to verify whether exits remain open, alarms work, cracks recur, equipment moves, and managers retaliate. Metrics should include complaint acknowledgement, investigation time, corrective action, repeat reports, retaliation allegations, and worker confirmation. Low complaint volume is not evidence of safety unless access and trust are independently tested.

8. A national denominator

Bangladesh needs a current count of operating factories and buildings, including export, domestic, subcontract, independent, and newly opened sites. The denominator should reconcile DIFE, associations, tax and customs records, utilities, buyers, Accord/RSC, and local authorities. Coverage percentages without that base cannot show who remains outside inspection.

9. Compensation linked to lifelong consequences

The record should distinguish emergency relief, wage replacement, death benefits, disability, medical treatment, rehabilitation, brand payments, employer obligations, court awards, and charitable donations. Beneficiaries need appeal rights and support when health or dependency changes. Fund closure should not erase unresolved medical care or pending liability.

10. Timely legal and administrative review

Evidence from permits, inspections, orders, attendance, communications, audits, and engineering work should be preserved in forms courts can use. Defendants are entitled to a fair determination, and victims are entitled to a process that does not lose practical meaning through delay. Published judgments should distinguish proven facts from allegations and explain how responsibility is allocated.

Source boundary

This analysis gives greatest weight to Bangladesh Supreme Court judgments, DIFE reports, ILO compensation and program records, the text of the Accord, official Accord/RSC data, the Canadian decisions, and the OECD NCP record. Company and buyer-program reports are used for what those organizations said they did and are not treated as independent certification.

The accessible public record does not contain one final engineering report that resolves every physical contribution to collapse. It also does not contain a completed judgment in the principal Bangladesh murder case as of 17 July 2026. Prosecution claims remain allegations; civil dismissal in Canada does not establish physical safety; OECD findings do not establish legal causation; and payment through the Rana Plaza Arrangement does not establish or extinguish every liability.

Counts use their source definitions. Death totals, injury totals, factory populations, findings, inspections, corrective-plan completion, claims, and beneficiaries are not interchangeable. Current web dashboards can change after publication. The dated figures in this article preserve the values displayed or reported by their issuers and should be reconciled before later trend claims are made.

Conclusion

Rana Plaza made supply-chain accountability concrete because the warning was visible and the control chain still failed. The crack did not need to identify a guilty person. It needed to stop occupancy, protect wages, summon qualified engineering review, and prevent reopening until evidence supported it. Instead, production resumed and workers bore the consequence.

The reforms that followed addressed real weaknesses. Independent structural, fire, and electrical inspections reached thousands of factories. The Accord made buyer leverage, transparency, worker refusal, commercial consequences, and remediation finance part of a binding safety program. National inspection capacity expanded. A coordinated arrangement delivered awards, and public factory records now expose findings and repair status that were previously difficult to see.

None of those achievements justifies declaring the accountability test complete. Aggregate progress can conceal one critical open item. Buyer coverage can leave other factories outside the program. A social audit can still be mistaken for engineering assurance. Compensation can deliver relief without resolving liability. A criminal case can remain pending long after evidence and witnesses begin to age.

The durable standard is a traceable decision chain: lawful building use, detected warning, protected stop-work, competent assessment, financed repair, verified closure, worker confirmation, public enforcement, and continuing review after buyers or owners change. Rana Plaza's central lesson is not that every buyer owned every local decision. It is that each actor with practical control must produce evidence that its part of the safety system worked, and the system must not leave the final risk with workers who cannot safely refuse it.