An independent editorial desk in Taipei
When a model reads a market, how much of what it sees is real? We follow the alternative-data pipelines behind AI investment signals — the quality checks, the honest labels, and the places where bad data hides.
Port traffic is one of the oldest alternative datasets in investing — and one of the easiest to misread.
The craft
Artificial intelligence has moved investing’s hardest problem from finding numbers to trusting them. Every AI investment signal is only as honest as the alternative data feeding it, so this desk reports on the feeds, the cleaning and the failure modes — never on tips.
An investment signal cooked up by a model has a supply chain: something was captured, cleaned, mapped, scored and packaged. We trace that chain for the datasets we cover — who collects the source, what gets corrected or dropped along the way, and which steps a vendor would rather not discuss.
When the trail goes cold, the article says so plainly. An unanswered lineage question is itself a finding worth publishing.
See how each article is built →Crowding, vendor drift and regime shifts quietly hollow out signals that once worked. We walk through how decay shows up, and what to ask a data vendor about it.
Read the explainer →What a well-documented signal should disclose — universe rules, feature lineage, training window, turnover — and what to suspect when a dossier stays quiet.
Read the explainer →Method
Each piece starts with a lineage question: where do these numbers originate, who reshapes them on the way in, and what could quietly distort them before a model ever sees them?
We ask for point-in-time discipline, entity mapping that survives mergers, and vendors’ revision histories. If the documentation cannot answer, that gap is written into the piece — because computing a signal is the easy part, and knowing the data deserves it is the hard one.
Standards
Where a dataset comes from, who licenses it, and where the paper trail goes cold.
Whether each historical value reflects only what was knowable at the time — and how that was checked.
Why a signal that once worked may fade, and which kind of decay is at work.
Mechanisms explained, positions never recommended. The articles stay free to read.
Objections
No. Nothing on this site recommends any security, fund or strategy, and nothing is personalized. Articles explain how data and models behave, so readers can ask sharper questions elsewhere.
No paid tiers, no data products, no courses and no consulting. The essays are free editorial pieces; inquiries are welcome, and nothing is charged, booked or executed for them.
A one-desk editorial publication based in Taipei. The registered name, address and business number sit at the foot of every page, and inquiries reach the desk directly.
Inquiries
If a feed, a signal or a confident claim about AI in markets deserves a closer look, bring it to the desk. Inquiries are read and answered by email from Taipei — ordinary questions are welcome too.