The Clusters tab is the most persuasive screen in the module and that is exactly the problem. A card puts a ticker at the top, tells you how many politicians are involved, gives you a date window and then lists the members in a neat table. Everything about the layout says these people did something together.
They might have. But the grouping is done by software joining rows on shared attributes, and shared attributes are not the same as shared intent. Before you act on one of these, you should know what the join is actually made of, and the fastest way to learn that is to take one card apart field by field.
Every field on one card
The tab opens with two buttons, All Clusters and Bipartisan Only, and then a list of cards. Here is the one I pulled, read left to right and top to bottom.
The ticker and company name sit at the top left, Microsoft in this case. A bare number sits at the top right, 45, with no label next to it, so I do not know what it scores and I am not going to guess. Under the ticker runs a summary line: three politicians, Sale, 56.50 K USD volume. Under that, a window with explicit dates, running from the 21st to the 30th of the month, so a ten day span. Then a table with four columns, Party, Politician, Type and Amount.
Start with the size. The whole cluster is 56,500 dollars of disclosed activity across three politicians, which is under 19,000 dollars each. The module's own header reports average trade size at $40K under the label Midpoint USD. So this cluster, in total, is smaller than two average congressional trades. That is worth sitting with for a second, because the visual weight of a card with a table in it suggests something much larger than it is.
What the join is visibly made of
Two things are unmistakable from the card. Every row is the same ticker, and every row falls inside the same date window. Those are the two attributes the grouping is built on as far as anything visible tells me, and they are the two weakest attributes available.
What I cannot see on the card is any evidence that committee membership, chamber or party enters the grouping itself. Party is clearly available, since there is a Party column and a Bipartisan Only view, but a filter you apply on top of a finished list is not the same as a rule used to build it. So my working assumption, and I would rather state it as an assumption than dress it up, is that a cluster means these filings involved the same ticker within the same short window, and nothing more.
If that is what it means, then two members in one cluster may share nothing at all except a calendar week. Not a committee, not a state, not a conversation, not a broker. One of them may have been rebalancing an account after a house sale and the other following an adviser's model portfolio, and the card will present them as a group either way.
Three ordinary reasons a cluster exists
The first is the window. Ten days is long in filing terms. Congressional filings arrive continuously, and the module header reports 74 filings in a seven day period. Any window that wide will collect unrelated trades on popular names simply because unrelated trades keep happening.
The second is the ticker itself. Microsoft is in almost every retirement account in the country. Members of Congress hold what everyone holds, and a company that sits in every index fund and every managed portfolio will generate co-occurrences forever without a single person having a view about it. The rarer the company, the more a cluster means. On a mega-cap it means very little, and this is the single most useful instinct to develop on this tab.
The third is repeat filing. Look again at the rows on my card. Three rows, and the same member name appears on two of them. So the group of three includes at least one person twice, and the number of distinct people in it is smaller than the headline count implies. A single member filing the same company twice in a fortnight is one decision with two pieces of paperwork, not two independent signals.
When the summary line and the rows disagree
Here is the detail I would put at the top of the tab if it were mine to design. The summary line on this card reads Sale. The rows underneath read Sell, Buy and Buy.
Whatever rule produces the header direction, the group is not doing one thing. Two members bought and one sold, or one member bought twice and another sold, depending on how you count the repeated name. Either way, if you read only the summary and concluded that Congress was selling Microsoft, you would have taken a card whose members were mostly buying and turned it into the opposite signal.
This is not a bug you need to report, it is a reading habit you need to build. The header is a summary of a mixed group. The rows are the data. Always open the table, and if the rows do not agree with each other, the card is not a signal about anything, it is a coincidence with a border around it.
The five minute check before you act on a card
Four questions, in order, and any one of them can end the exercise.
How many distinct people are actually here? Count names, not rows. If three rows resolve to two people, treat it as a two-person cluster and lose most of your interest.
Are they doing the same thing? Read the Type column on every row. Mixed directions means no group view. All buys is the version worth continuing with, since there are fewer housekeeping reasons to buy a specific company than to sell one.
How common is the company? If it is a name held in every index fund, the base rate of co-occurrence is high and you need many more members before the card means anything. If it is a company you have never heard of, a small cluster is far more interesting.
How wide is the window and how old is it? The card shows the window explicitly. A ten day span is loose. And the module reports average delay from trade to filing at 32.5 days, so the trades in this window happened roughly a month before the card appeared. Pull a chart of the name before doing anything. If it already moved, the information in the card was consumed while the paperwork was in transit.
If the card survives all four, you have earned the right to look at the company properly, which is where the real work starts. What you have not earned is a position. A cluster is a reason to open a research tab, and the moment you let it become a reason to buy, you are sizing a position on the strength of a layout choice.