Limits
What it will not do.
These are the limits that could cost you money if you did not know about them.
It does not find an edge
Every filing it reads is published because the law requires it, on a public website, free to anyone. Professional investors read those feeds within seconds. By the time a filing reaches your inbox, whatever effect it was going to have on a price has already happened.
What the tool removes is the gap between a filing being published and a person seeing it, in dollars, next to everything else filed that day. The tool helps you read filings. It makes no claim about returns. There is no backtest in this project and no record of performance, so there is nothing in it that could support a claim about returns even if one were made.
It is not investment advice, it is not a trading system, and it does not tell you what to buy.
Being traceable is not the same as being right
The project guarantees that every figure it prints was copied out of a field in a filing. That guarantee says where a number came from. It does not say the field held what the provider claimed it held.
Those are two different promises, and the second one needs its own machinery. A vendor once sent a price field holding the total value of a trade instead of the price of a share, which produced a position valued at $1.6 quadrillion that passed every trace check cleanly. A separate set of checks handles this case. Those checks can also be wrong. The detail is here.
Fund holdings arrive late
US fund holdings reported on Form 13F are between 45 and 135 days old by the time they are published. A fund can have sold the entire position before you see it.
That data is also long only. It does not include short positions and it does not include derivatives, so a fund that looks heavily invested in a company by its 13F may be hedged in ways the filing never shows. Insider filings do not have this problem, and typically appear within days.
The weekly note carries this caveat in every issue rather than only here. It also notes that insider buys carry more information than insider sells, that the historical effect was concentrated in opportunistic rather than routine trades and in smaller companies, and that it has weakened since the research describing it was published. For Taiwan specifically, published work found no reliable abnormal returns from insider trading disclosures, and Taiwan is one of the markets covered here.
Not built yet
The current code has these gaps.
Currency conversion uses fixed rates
Non-US filings are converted using approximate rates that ignore the date of the filing. A figure converted from yen or Taiwan dollars carries an error that grows with how much that currency has moved since the filing. US figures are unaffected.
Storage is SQLite only
A Postgres connection string is read from the configuration and then not used. Everything is stored in a local SQLite file. That works for one person on one machine. It does not work for a shared setup.
You cannot choose a delivery channel
Slack delivery is implemented, and there is no way to select it. Delivery goes by email.
Several countries are not covered
There is no reader for South Korea, the United Kingdom, Hong Kong, Australia, India, or the European transparency filings. A disclosure in any of those markets will not appear in your email, and nothing on the page will tell you it is missing, because the tool does not know those filings exist. Covered today: the United States, Japan and Taiwan, plus optional vendor and on-chain feeds.
Filings without a dollar amount stay unpriced
Filings that report a share count or a percentage rather than an amount cannot be valued without a price provider, and the free price providers are not enough. Those filings are read and stored with no dollar figure. They are valued later if a paid price feed is turned on. Until then they sit under a size threshold they were never measured against.
What it does guarantee
Given the same filings and the same date, the daily table, the daily email, the weekly report and the ordering are identical every time. Written prose is the exception, and the figures inside it are checked against the filings before anything is sent.
A source that is broken or switched off becomes a line of text in your email naming it. The email does not just leave a gap. A source that failed and a source that had a quiet day are recorded differently.
A run that attempts to send and is refused reports a failure rather than success. If the pipeline stops delivering, you see it right away.