Taipei · AI in investment · Free to read
Glacier Workspace Journal
The loudest promise in AI investing is that the machine will pick for you. The durable gain sits a step earlier — in how you collect, compress and check what you read before any decision. This journal documents that stage: research workflows for individual investors, tested against real documents, written for people who want AI as a reading tool rather than an oracle.
Prefer a straight question? Ask the desk directly — inquiries cost nothing, and nothing on this site is for sale.
Claim № 1, checked
“AI watches the market so you don’t have to.”
No chat window files an alert when a company quietly rewords its risk section, and no model carries the downside when a paraphrase smooths over a footnote. What today’s tools genuinely do well is the mechanical layer of research: pulling passages out of hundred-page filings, restating dense paragraphs, lining two figures up for comparison. The notes on this journal hold that layer to a strict rule — useful assistance, checked output, no verdicts.
See how a claim gets checked
Claim № 2, checked · “a chat reply is research”
A chat answer is not a research workflow
A single prompt returns a single paragraph, and research is never a paragraph. The desk works every subject along four stages — the same sequence an individual reader can copy at a kitchen table. Articles in this journal describe them because these stages are precisely where AI earns its keep, and where it quietly fails.
- Sourcing. List the primary material first: the annual report, the earnings-call transcript, the regulator’s page, the company’s own releases. Models index the pile; they never stand in for it.
- Extraction. The model pulls exact passages, tables and quotes out of long documents, each marked with its source page so a claim stays traceable to where it lives.
- Verification. Every load-bearing number is reopened against the original document. Anything the model produced that the source will not confirm is struck from the note, not phrased away.
- Filing. Checked fragments go into structured notes — quote, source, date read, one line of comment — so any line can be re-checked in minutes rather than evenings.
The workflow stage by stage, with the desk’s rules for each step →
Claim № 3, checked · “serious research is a paid product”
The journal is free. The desk still answers mail.
There is no paywall, no premium tier, no membership and no check-out anywhere on this site — the offer begins and ends with articles you can read for nothing. What the desk does run is a question queue: readers write in about research workflow problems, and recurring questions turn into published notes.
- How a specific note performed its verification, step by step
- What an AI tool can and cannot do for a reading task you name
- Requests for subjects the journal should take apart next
The desk does not recommend, rate, broker or time any investment, and no private analysis is sold — there is no paid channel to sell it through.
Received by the form handler
The handler took your question in. Replies leave the desk by email, and questions that arrive often turn into published notes — the queue is public that way.
The question was not sent
The inquiry endpoint did not confirm delivery on this attempt. Nothing was lost by trying; a retry reaches the same queue.
The office mailbox is the same desk under a different door: info@glacierworkspace.digital
Read this before subscribing to anything — including this journal, which has nothing to subscribe to
Who these notes are written for
The journal writes for one particular reader: the individual who does their own reading before committing money, keeps their own notes, and wants AI’s speed without adopting AI’s confidence. If that is you, the desk is at your service. If it is not, better to say so up front, before you spend an evening here.
Written for readers who…
- open the primary document themselves before forming a view on a company
- test AI tools against real filings, transcripts and regulatory pages
- want every claim traceable back to a page it came from
Not written for anyone who wants…
- personalised investment advice — this is an editorial publication, not an advisory service
- trading signals, price targets, or “my method never loses” promises of any kind
- the desk to buy, sell or hold anything on their behalf
Recent research notes
Twenty honest minutes with a filing beats a week of vibes
Invisible ink: what an AI summary leaves out of an annual report — what the compressor throws away, and how to spot it
Notes that survive a chatbot: file your reading so future-you can audit it — three layers of filing, one re-checkable page
The verification debt: checking what a model tells you about a company — a checklist for paying down unchecked claims
All research notes