At its Vencorex France inspection of 14 January 2026, the DREAL Auvergne-Rhône-Alpes recorded an allocated budget of EUR 17 million to EUR 20 million for making-safe and cleaning operations. Its report, reference 2026-Is004SPF, prints the range on page 6. That is an allocation, not expenditure, and it concerns stopped installations, not new production capacity. The distinction is the story: engineering delivery does not always mean producing more.
What counts as delivered engineering work when one site is securing stopped units, another reports commissioning, and a third is preparing a future start? Vencorex at Pont-de-Claix, Arkema at Jarrie and Michelin’s 5-HMF project at Roussillon offer three different answers. Their public documents are not a league table of successful companies. They are a way to examine what a milestone actually establishes before a candidate, employer or investment committee starts drawing conclusions from it.
Chemical engineering delivery needs a proof matched to the task: a bounded safety finding, an attributed commissioning milestone and a construction launch cannot be exchanged for one common completion score.
This third chemicals and process piece in the CareerOn Industry Atlas uses eight document families, including original French inspection reports, operator announcements and a funder’s release. Native English and French editions of an announcement are one family, not two witnesses. The record was checked on 11 October 2026. Our comparisons and practical exercises are analytical proposals; they do not describe advertised vacancies or claim access to company acceptance files. Where the public record stops, the argument stops too.
Making a stopped installation safe is work, not an empty project
The January 2026 DREAL report says Vencorex began making stopped installations safe on 5 January 2025 (page 6). The date is reproduced as the report prints it; it should not be silently shifted to fit a later legal milestone. The report’s inspection date and the date it gives for work beginning describe different events. A reader can acknowledge the work without pretending that the document is a comprehensive history of the site’s shutdown.
On page 8, the DREAL states that the ANITA, DNT and TDA workshops and the electrolysis installations of chains 2 and 3 had been emptied, disconnected and secured in accordance with regulatory requirements. Those are named installations, not a certificate covering every part of Vencorex. The visit excluded Dessalage and wells, and corrective work remained. “Secured” is therefore meaningful within a boundary. Removing the boundary would convert a useful finding into an unsupported whole-site conclusion.
The same page records the liquidation team’s estimate that making the site safe could be completed by 30 June 2026, subject to corrective-action deadlines and the necessary resources. By our reading date, that target had passed. No admitted follow-up document proves that it was achieved. Neither the calendar nor a new industrial proposal fills that gap. A conditional estimate belongs in the milestone record as a conditional estimate, even when the date has become historical.
An earlier inspection shows why scope matters. The DREAL report signed on 28 March 2025, following a visit on 18 March, says on printed page 4: “This inspection concerns only the HDI.2 sector as regards making installations safe” (our translation). It is not the same inspected set as the later report. Combining the two into a percentage of installations cleared would require an inventory and a common denominator which these credited passages do not supply.
On page 5, that earlier report asks the operator to trace installations it would be unable to make safe before judicial liquidation and, at minimum, identify and quantify waste still to be removed. This is a requested control, not evidence that the requested work was completed. The analytical lesson is not that Vencorex failed a new test devised by CareerOn. It is that residual actions are part of the evidence needed to understand a closure claim, alongside the work already documented.
Completed named work does not close the remaining scope
| Document | Bounded finding | Closure limit |
|---|---|---|
| Visit 18 March 2025; signed 28 March 2025, pp. 4–5 | HDI.2 only; traceability requested for unsecured installations and remaining waste | Request is not proof of fulfilment |
| Visit 14 January 2026, p. 6 | EUR 17m to EUR 20m allocated to making-safe and cleaning | Allocation is not expenditure |
| Visit 14 January 2026, p. 8 | ANITA, DNT, TDA and chain 2/3 electrolysis emptied, disconnected and secured | Not site-wide clearance; conditional 30 June 2026 target remains unverified |
Sources 1 DREAL Auvergne-Rhône-Alpes · 2 DREAL Auvergne-Rhône-Alpes
The budget range adds scale, not a spending verdict. Averaging it into a single number would create a figure the DREAL did not publish. Treating it as capital for new capacity would change the purpose of the allocation. For a reader comparing industrial projects, the sharper question is what the money is assigned to and what the document can establish about the result. Safety work can be substantial without creating a saleable tonne of additional product.
Jarrie: an input constraint and a reported commissioning are different records
Arkema’s 21 January 2025 refocusing announcement explains that substitute salt could support Jarrie’s hydrogen peroxide, chlorate and perchlorate workshops, but was not compatible in specifications, cost and quantity with maintaining all site activities. This is Arkema’s explanation in its original French release, not an independent measurement of those constraints. It shows why replacing an input is not merely finding another supplier: the announced choice was bounded by what the retained processes could use.
That announcement does not prove a separate achieved chlorine-cessation date, completed redundancies or the detailed design of a later hydrogen unit. The public record must not be made tidier by turning a proposed refocusing into proof of every subsequent event. Nor can a Vencorex inspection certify work at Arkema’s establishment. The two sites belong to the same industrial discussion; their evidence remains attached to the installation and publisher it actually names.
A different Arkema document, “Arkema inaugurates new hydrogen production unit at Jarrie”, published on 22 September 2026, reports: “It was commissioned in July 2026, on schedule, thanks to the strong commitment of the site teams and their partners.” July is the reported commissioning month; September is the publication and inauguration context. Moving the operating milestone to the press-release date would erase information rather than improve comparability. The native French edition reports the same milestone, not an independent second acceptance.
It is reasonable to call that an operator-reported commissioning. It is not reasonable, on this passage alone, to invent a witnessed acceptance test, measured sustained output or the unit’s precise technical configuration. These are different questions. The source supports a positive milestone while leaving the performance evidence outside this dossier. An evidence-led account should be capable of that combination without either promotional inflation or reflexive suspicion.
The comparison with shutdown work is useful precisely because the deliverables differ. A stopped workshop is being emptied and secured; a new input unit is reported commissioned. Our proposed decision discipline is to ask which object changed state, who reports the change and what remains outside the statement. It is not a claim that Arkema uses CareerOn’s checklist, nor a ranking of the engineering teams behind the two pieces of work.
Roussillon: a construction launch is not available production
Michelin’s 23 May 2025 announcement of a 5-HMF industrial demonstrator prints an initial horizon: “It will allow for the creation of approximately 30 direct jobs and should begin its activities during 2026.” These are prospective jobs and activities. They are not a hiring return or a commissioning certificate. The announcement matters because it provides the earlier public plan against which later publications can be read, without claiming to know a more precise internal schedule.
The same announcement describes an annual production capacity of 3,000 metric tons at the Osiris platform in Péage en Roussillon. Capacity is not achieved output. Its statement that the site would be the world’s largest for the molecule remains Michelin’s claim. The figure can help a reader understand the announced industrial scale; it cannot establish production, utilisation or a market position already attained. None of those outcomes follows from the capacity sentence.
GIE Osiris’s “Historique” page, read on 11 October 2026, records an October 2025 start of construction for Michelin’s 5-HMF unit. This is a living operator history: the month describes an event, not the page’s publication date. Its reading day is part of the credit. It adds a reported construction milestone, not an operating result, and should not be treated as a dated engineering acceptance document.
CBE JU supplies a dated account. On 24 October 2025, its release “Europe’s first industrial-scale production facility for a new bio-based molecule breaks ground in France” says: “Today, a 12-partner consortium led by Michelin officially launched the construction of Europe’s first industrial-scale production facility for the bio-based molecule 5-HMF in Roussillon, France.” The ranking is the funder’s wording. Launching construction does not establish its completion, much less commercial production.
Crucially, that release already says: “Operations are planned to start beginning of 2027, creating about 30 direct jobs and strengthening the region’s industrial base.” The early-2027 horizon was therefore public in October 2025. It was not first revealed by the July 2026 partnership announcement. CBE JU is financially involved in the project; its separate document deepens the chronology, but it is not independent technical certification, a verified disbursement or evidence of confirmed hiring.
The Michelin/Axens joint release of 8 July 2026 then prints: “Capacity available from early 2027, with the commissioning of a first 5-HMF production unit operated by Michelin ResiCare in Péage-de-Roussillon.” That is still a future horizon at publication. The native French version belongs to the same document family. The later release repeats an already public horizon; it does not prove achieved capacity or justify calculating a delay in months from a year-only original target.
Early 2027 was public before the later partnership release
23 May 2025
Michelin announcement
Activities expected during 2026; prospective capacity and jobs.
October 2025
Osiris history
Construction start, on a living page read on 11 October 2026.
24 October 2025
CBE JU construction-launch release
Operations planned at beginning of 2027; funder account, not acceptance certification.
8 July 2026
Michelin/Axens joint release
Capacity from early 2027 remains a plan, not achieved output.
Sources 5 Michelin · 6 GIE Osiris · 7 Circular Bio-based Europe Joint Undertaking (CBE JU) · 8 Michelin
There is a changed public horizon between the May and October publications. There is no measured delay duration in the admitted record, no confirmed output and no grounds here for concluding inactivity. For someone preparing a project discussion, this distinction is more useful than a dramatic delay headline: it identifies exactly which later evidence would answer the operating question, without pretending that a forecast answers it now.
The counter-case: incomplete evidence is not a failed project
These limitations must not become a negative verdict on the companies. The DREAL records completed work on named Vencorex installations. Arkema reports Jarrie commissioning in July 2026. CBE JU reports a construction launch at Roussillon. Different publishers document positive milestones, even though they do not answer the same acceptance question. An incomplete public record is not project failure, and refusing to invent a missing outcome is not evidence that the outcome never occurred.
Different engineering tasks establish different delivery states
| Task | Public evidence | What remains unproved |
|---|---|---|
| Secure stopped Vencorex units | DREAL: named installations secured; HDI.2 scope separately bounded | Whole-site clearance; fulfilment of residual actions |
| Commission Jarrie hydrogen unit | Arkema reports commissioning in July 2026 | Independently witnessed acceptance; sustained output |
| Build Roussillon 5-HMF unit | CBE JU reports construction launch on 24 October 2025 | Completed commissioning; achieved production |
Sources 1 DREAL Auvergne-Rhône-Alpes · 2 DREAL Auvergne-Rhône-Alpes · 4 Arkema · 7 Circular Bio-based Europe Joint Undertaking (CBE JU)
Our reading is narrower than an audit of engineering performance. We have neither plant acceptance files nor a complete inventory of corrective-action closure. We can compare the state each credited document describes, and we can show why those states are not interchangeable. That is the value of the comparison: a reader can retain both the reported achievement and the unresolved question, rather than being forced to choose between a success narrative and a failure narrative.
What this means for your career
For candidates, CareerOn proposes an evidence exercise, not an employer requirement: prepare a short milestone pack for one of the three tasks. Name the installation, the dated document, its publisher and the precise statement. Then write the question the statement leaves unanswered. For Vencorex, that might concern the residual scope; for Jarrie, performance after reported commissioning; for Roussillon, evidence of an operating start. Do not claim an answer that the pack does not contain.
For employers, our proposed interview discussion is to give a candidate the bounded passage and ask which decision it can support. Can they distinguish the allocated safety budget from money spent, and commissioning from inauguration? Can they preserve a conditional date without presenting it as achieved? This tests reasoning about the supplied evidence, not familiarity with confidential company procedures. A careful refusal can be more informative than a confident but unsupported completion claim.
For leaders, our analytical recommendation is to keep public milestone language separate from the evidence required for their own decisions. A funder’s construction announcement, an operator’s commissioning statement and a regulator’s scoped finding each answer a different question. If a committee wants to know sustained output, it must ask for output evidence; if it wants closure of remaining actions, it must ask for that scope. None of the three documents should answer by proxy.
CareerOn’s view is that the useful candidate is not the one who memorises the most announced capacity. It is the one who can explain what changed, who says so, which installation the statement covers and what remains unknown. These are proposed analytical exercises, not existing CareerOn simulations or asserted recruitment criteria. No practice link is supplied until a matching commissioned exercise exists and its destination is checked. The public record is already enough to practise disciplined judgement; it is not permission to invent the rest.