chemXplore

What a company has committed to arrives as an announcement: a press release, a trade article, a line in a results call. Each one is findable. Together they are not a picture. There is no list of what is being built, no way to see which projects moved while you were busy, and no denominator — you cannot tell what you are missing.

An assistant will summarise any one of those announcements for you. It cannot tell you which ammonia projects in Europe passed their investment decision this year and have not started up, because that answer does not exist in any one source. It has to be kept.

chemXplore keeps it. Every project is followed from the first announcement through engineering, the investment decision, construction and start-up, with the companies on it and the role each one holds, what it will make and at what capacity, the plant it belongs to — and the source behind every one of those facts. Today:

• 2,209 projects, of which 804 have not started up yet
• 6,213 dated milestones on them
• 2,020 companies holding a named role

In depth across Europe since 2013, and a growing view of the rest of the world. How the record is made →

A fair question, and it has stopped meaning "ask a chatbot". You may already have a deep research run that does this weekly, or an agent of your own wired to search and scrape and write to a sheet. So do we: this record is filled by agents that read sources, extract what they say about a project and propose the change, with people ruling on it where accuracy decides the answer.

Which is why the answer is not that the technology falls short. It is that after the agent runs, the work starts — and that work is what a question like "which projects moved this quarter" actually rests on.

• A run gives you a snapshot; the value is in the diff. Run the same agent twice and you get two lists, assembled from scratch, with nothing to say what changed between them. Knowing that a project reached its investment decision in August means holding what it was in July. This record holds 6,213 dated milestones, each with when it happened and what reported it.
• An agent stops when it stops. That is not the same as coverage. It returns what it found, and cannot tell you what it missed — and in this industry what you missed is usually the point.
• The hard part is not extraction, it is identity. The same project reaches you five times under five names, and again next month with a new contractor attached. Deciding those are one project — and the same one you recorded in March — is a judgement made once and kept, not remade differently on every run.
• An answer has to come back while you are still thinking. Filter 2,209 projects by substance, region and phase and the list is there at once, ready to narrow again. A research run takes minutes, and you cannot refine a question ten times that way.
• And the second month costs what the first did. Keeping this current over the last 30 days took 1,365 sources read, 9,298 model calls and 590 proposed changes put in front of a person — every month, not once. Point your own agents at it and that bill is yours, along with the pipeline to maintain when a site changes its markup or a model changes its behaviour.

Build it yourself and within a month you own a maintenance job rather than an answer. What a subscription buys is the running already done — for thirteen years, and again this morning.