What has been measured, and what has not.

Executive MBA students reconnected with contacts they had not communicated with for at least three years. They then rated what came back against the advice they were currently getting from their active network.

On a seven-point scale:

Trust declined modestly (p = 0.037). Shared perspective declined by less, and not significantly (p = 0.374).

Then what the reconnections actually returned:

After three years of silence the relationship is largely intact. And the dormant contacts delivered more novel information than anyone currently active in these people's lives.

This does not show that dormant ties are flatly more useful than current strong ones. On useful knowledge it is 5.70 against 5.63, which is a hair, not a gap. Two claims survive the numbers: dormant ties clearly beat current weak ties, and dormant ties beat every current category on novelty.

Levin, Walter & Murnighan, Dormant Ties: The Value of Reconnecting, Organization Science 22(4), 2011

Same paper, second study. 116 executives were invited and 95 completed it. Each listed their top ten dormant contacts and ranked them by how much they wanted to reconnect. Each then reconnected with their number one, and with one contact drawn at random from ranks two to ten.

Correlation between preference rank and how valuable the reconnection actually was:

r = −0.074 p = 0.333

Flat. The tenth choice was worth as much as the first. In the authors' phrase, the pool of valuable dormant ties is deep.

A later paper by the same group says why the ranking comes out the way it does.

Read that carefully. The discomfort of reaching out does not merely reduce how often people do it. It steers them, systematically, toward the reconnections carrying the least new information.

That is why this is a search problem and not a discipline problem. Trying harder makes it worse, not better. Effort applied to your own ranking pushes you further down the list of people who can tell you something you do not already know.

Levin, Walter & Murnighan, Organization Science 22(4), 2011, Study 2 · Walter, Levin & Murnighan, 2015

More than 20 million people, five years, 2 billion new ties and 600,000 job transmissions, using randomised variation in LinkedIn's People You May Know algorithm. It is the largest causal test of weak ties that exists.

It is a correction as much as a confirmation.

The finding is an inverted U. Weaker ties do transmit more opportunity, but with diminishing returns, and past a point weaker is worse. Weakness is not a quantity you want to maximise.

Both results are worth reporting, not only the convenient one:

• Measured by mutual connections, moderately weak ties performed best.
• Measured by interaction intensity, the weakest ties did.

The effect is also conditional. Weak ties helped more in more-digital industries. Strong ties helped more in less-digital ones.

This measures what happens when an algorithm recommends weaker connections, with job transmission observed on the platform. That is a narrower claim than the headlines made of it.

Rajkumar, Saint-Jacques, Bojinov, Brynjolfsson & Aral, A causal test of the strength of weak ties, Science 377(6612), 1304–1310, 2022 · DOI 10.1126/science.abl4476

Every study above has a boundary. Stating them here rather than in a footnote, because the boundaries are the part most often dropped when this research is quoted.

Granovetter is smaller than it is quoted as being

The famous frequency table rests on 54 people in one affluent Boston suburb around 1969, in professional, technical and managerial occupations only. Not the 282 usually quoted, which was the survey sample.

The dormant-tie studies measured a self-report, not an outcome

They measured self-reported usefulness roughly a month after a reconnection the participant chose to make, in a study whose purpose they knew. No job offers, no revenue, no promotions. There was no no-reconnection control group.

Participants were mid-career executives, mean age 38.3, 79.1% male, with unusually valuable dormant networks. Whether this generalises to a 24-year-old, or outside professional labour markets, is untested.

Burt's structural holes are associations, not causes

673 managers, one firm, one year. The findings are correlational. Burt's own framing is that brokers are “at higher risk of having good ideas” — exposure to variance, not superior judgement.

In that study 32% of all ideas submitted were dismissed by both senior judges.

Dunbar's number is not a constant

The 1992 estimate was 148, with a 95% confidence interval of 100 to 230. A 2021 re-analysis produced estimates of 69–109 and 16–42, with intervals spanning 4–520 and 2–336.

We use it as a rough ceiling on how many relationships a person can actively hold. It will not carry more weight than that.

The evidence above is about people in general. It is not evidence that Rungs works. These are the observations that would defeat the premise, written down before we have the data rather than after.

Each of these needs a fixed window and a held-out group to be a real test. We will publish both before the measurement starts, not after seeing it.