The Top Relationships Project

I built a project to answer a simple question:

Who are the people I have actually had the deepest relationships with, as measured not by memory or sentiment in the moment, but by the long arc of contact over time?

The raw material was my email history. Not bodies, not attachments, not full-text search across private correspondence. Just the structure around the messages: who wrote whom, when, how often, how directly, how reciprocally, and over how many years.

That turns out to be enough to say a surprising amount.

If two people have been in repeated direct contact over a long span, if both initiate, if the exchange is not just one burst but a recurring pattern, if the thread structure suggests actual back-and-forth rather than ceremonial CC traffic, then you can start to detect something real. Not everything. But something.

So I built a relationship graph from my email archive and scored the connections.

The result is not “best friends” in the sentimental sense, and not “most important contacts” in the networking sense. It is something more mechanical and, for that reason, more revealing: the people who show up most strongly in the actual record of my life.

Some rankings are exactly who I would expect. Some are mildly shocking. Some people I think of as central turn out to have a thinner email trace than I imagined. Others, who were present across many years and many phases, emerge as stronger than memory alone would suggest.

That is part of what makes the project interesting.

Email is accidental autobiography. It stores duration, recurrence, reciprocity, responsiveness, intensity, silence, reactivation, drift, and return. If you analyze it carefully enough, it becomes a map of relationship structure.

The model I built uses signals like:

• total interaction volume
• direct versus group exchange
• mutuality and balance
• reply alternation over time
• longevity across months and years
• recency
• recurring subject patterns
• likelihood that a contact is actually human rather than automated noise

From those signals, I generated a ranked list of relationships and then built a report around it.

Here is the public writeup and report:

• Public post: https://amol.sarva.co/relationships-erdos-epstein-enron/
• Sanitized report: https://a.sarva.co/TopContacts/omnibus_contact_relationship_report.html

What I like about this project is that it turns a vague social intuition into something inspectable.

A life is not a flat address book.

It is a weighted graph.

Some edges are intense and brief. Some are low-grade but permanent. Some are deeply reciprocal. Some are mostly one-way. Some people disappear and then return years later. Some are present in every era. Some are artifacts of work. Some are artifacts of friendship. Some are both.

This project tries to make that visible.

It is imperfect, obviously. Email is only one slice of a relationship. Some of the closest people in life may not use email much. Some eras are overrepresented. Some professional relationships can look very strong in the data. Any ranking system can confuse intensity with importance.

But even with those limits, the output feels real enough to be interesting and wrong enough to be human.

That is the sweet spot.

The next step is the strange one: I may send people their rank.

Not as a pronouncement of emotional truth, but as a data artifact from a long personal archive. A kind of computational social mirror.

If you get one of those notes from me, it means the graph thinks you matter.

And in many cases, it is probably right.

Here’s the full sanitized report without names named: https://amolsarva.com/topcontacts/public.html