The Merge Tax

Estimated reading time: 6 minutes

When two hundred sources get pressed into one clean story, the sources stop being evidence.

A finished answer looks like progress. The trail that could overturn it is what you actually needed.

The Name That Looked Settled

Anyone who has used commercial genealogy tools has made this mistake, or watched someone make it.

You open another person’s tree. A spouse appears. A parent. A child with a surname you did not expect. The record looks complete. The dates sit still. The name is new to you, and that novelty feels like discovery. You copy the fact into your own tree.

In the best case, you learn that a relationship existed without learning what it brought to the family. You inherit a label without the tapestry: who argued, who migrated, who was left out of the story on purpose. In the worse case, the label was never earned. Someone upstream guessed, merged two people with similar names, or trusted a third tree the same way you are trusting theirs. Your clean row is now a vector. You share the tree. Someone else copies the same row. Bad data does not stay local. It propagates.

The excitement is the tell. We copy when we feel we have found something. We rarely pause to ask whether we found a source, or only a finished sentence about a source we never saw.

None of this requires bad faith. Most of the people who fill those trees are trying to help. That is why the pollution spreads. Helpful certainty travels farther than a cautious “not proved yet.”

What the Copy Erased

Provenance is the boring word for a practical question: how would I check this if I had to argue with someone who disagrees?

When you paste a “fact” from another tree, you usually paste the conclusion and leave the path behind. The marriage register, the passenger list, the conflicting census, the note marked “possible, not proved” do not travel with the click. Your tree now shows a clean relationship. The receipts that would let a later reader reopen the question are gone from your view, even if they still exist somewhere upstream.

That is the shape of the failure. Not ignorance. Compression. Many possible stories get pressed into one presentable line. The line is usable. It is no longer checkable from where you stand.

I have watched this cycle in my own family tree research and in other people’s. The first copy feels private. The second share makes it public. By the third hop, nobody remembers which tree invented the certainty. The clean row has become folklore with dates.

Family tree research makes the pattern visible because the product rewards completeness. Empty fields itch. A filled field feels like care. The same itch shows up anywhere an organization wants one coherent answer from a messy pile of sources: a research brief, a policy memo, a knowledge base article, an agent reply that summarizes “what we know.” The tools change. The fee does not.

Name the Fee

I call that fee The Merge Tax.

The Merge Tax is what you pay when cleanliness replaces the trail. You get a story you can share. You lose the path back to the sources that could challenge it. Historical Knowledge Fabric argued that readable pages are not the same as trustworthy knowledge until evidence, claims, identity, and interpretation stay distinct. This essay is the sequel about the celebrated next step: the moment those distinctions get flattened so the answer looks finished.

You see the tax in genealogy when a copied name spreads faster than any proof argument. You see it in teams when a polished summary becomes the document of record and nobody can find the conflicting notes underneath. You see it when an assistant produces a confident paragraph from many inputs and the paragraph becomes the only artifact anyone keeps.

The merge is not always a crime. Sometimes compression is honest: you keep the sources, name what you discarded, and leave a way to reopen the case. A tip that points at a register is still a tip. A tip that pretends to be a settled fact is already collecting interest. The tax comes due when the clean story ships without a path back.

Usable is not the same as checkable. A coherent answer is not a preserved trail.

Stake the Trail to the Domain

The portable lesson is not “never compress.” It is: decide how much the trail matters before you spend it.

There are domains where a clean story without provenance is fine. A lunch recommendation. A brainstorm that will be thrown away. An early sketch nobody will treat as settled fact. Demanding a full evidence path for every low-stakes answer is its own kind of drag. The Merge Tax is highest where a wrong clean story compounds: family history that other people will copy, institutional memory that will outlast the person who summarized it, agent answers that become the only artifact anyone keeps.

Having provenance can also be false comfort. A footnote nobody can open. A citation pile that points at the same secondary summary. A “sources” field that names a tool, not a record. The badge of rigor without a path a skeptic can walk. That is inventory dressed as care. It soothes the same itch as the empty field, and it fails the same test: can someone reopen the question without calling the person who pressed the story together?

So the habit is a domain question first. What breaks if this clean answer is wrong? Who will copy it? How long does it need to survive? If the stakes are low, compress and move on. If the stakes are high, refuse the free copy. Treat another genealogy tree as a tip, not a source. Follow the tip to a register, a passenger list, a conflicting census. Keep what you can defend. Leave the rest marked unresolved. In organizational work, refuse to let a summary retire the files it compressed unless the summary carries a real receipt: what went in, what was left unresolved, who is allowed to reopen it. In agent work, do not celebrate a fluent paragraph that cannot point back to the materials it used.

The COBOL Problem asked who sits between the contributor and the record. The Merge Tax asks what happens when the record looks finished because someone already did the sitting for you, silently, and threw the path away. Negative Knowledge is adjacent: keep the trail when a claim turns out wrong. Here the trail disappears earlier, at the moment of success, unless you decide the domain still needs it.

Two hundred sources. One clean story. Keep the receipts when the domain requires them. Otherwise admit you spent them on purpose.

How important is a checkable trail in the problem you are solving today?

Madam I’m Adam

This continues the thread from Historical Knowledge Fabric, where readable evidence still needed a path that disagreement could travel.