How the data was produced
Two exports from Majestic on 2 August 2026: the Historic index and the Fresh index for billhartzer.com, at the domain level, with all available columns. The Historic file is 809 MB and 1,301,839 rows.
Every figure was computed by streaming the full export rather than sampling. The survival analysis uses one observation per referring domain: the longest-lived link from that domain, or a still-live link where one exists. Live links are censored at their last-seen date.
Two practical notes for anyone reproducing this. The export contains NUL bytes, which will crash a naive CSV reader — strip them on read. And the Historic export returned 1,301,839 rows against the 1,382,639 links the Summary screen reported, or 94.2%; every figure here is computed on what the export actually contained rather than on the headline.
What this study cannot show
One profile. A twenty-year English-language blog about SEO and domain names is not a general sample. The concentration figures in particular are shaped by two sitewide links that a different site would not have. Nothing here should be read as a law.
Trust Flow is measured at export time, not at the moment each link was acquired. A domain that was strong in 2011 and is Trust Flow 0 today is banded as Trust Flow 0. That will bias the low bands' survival downward by an unknown amount, and it is the most important limitation of the Trust Flow survival finding. It does not explain away a fourfold difference, but it could be inflating it.
“Lost” means no longer observed in the index, which is not identical to “deliberately removed.” Crawl scheduling and index policy contribute to both the timing and the classification.
Survival is measured from first-indexed, not from link placement. A link may have existed for some time before the crawler saw it, which compresses measured lifespans slightly.
The crawler-blocking hypothesis is unconfirmed. It is labelled as a hypothesis on the audit page and it stays that way until server logs say otherwise.
What would make it stronger
A second profile. This is the big one. One profile is an observation; two would let the study say something about whether the pattern travels. A different niche, a different language and a different age would each test a different assumption.
A profile that did buy links, for contrast. The unbought characteristics here are only meaningful against something to compare them to, and nobody publishes a bought profile for obvious reasons.
Trust Flow at acquisition time, which would remove the largest caveat above. That needs historic metric snapshots rather than a single export.
Segmentation by link type and placement. Do in-content links outlive footer links? The data supports asking; I have not run it.
Reproducing this on your own profile
If you have a backlink index subscription with historic data and a domain with a few years of history, the survival analysis is reproducible in an afternoon.
You need four columns: referring domain, first-indexed date, last-seen date, and a lost flag. For each referring domain, take one observation — time from first-indexed to lost date for domains that stopped linking, or first-indexed to last-seen for domains still linking, flagged as censored. That is the input to a standard Kaplan–Meier estimator.
The trap to avoid is the one described on the survival page: do not compute this per link. If a handful of domains sent tens of thousands of sitewide links each, a per-link analysis measures those few sites re-templating rather than link durability, and it can invert your conclusion entirely.
If you run it, I would like to see the result — particularly whether the Trust Flow relationship holds. One profile is an observation. Two would be a finding.
Citing this study
Everything here is free to quote, chart, or argue with. No permission needed and no attribution required, though a link is appreciated and is, given the subject matter, somewhat to the point.
If you are quoting a specific figure, quote the median as 1,080 days rather than 3.1 years, and please carry the caveat that it is one profile. The 3.1-year figure that appeared in an earlier draft of this work was computed naively and is superseded — the survival page explains why.
Questions
Can I get the raw data?
The raw Majestic export is not mine to redistribute under their terms. Every aggregate figure, and the method behind each one, is published across these pages in enough detail to reproduce the analysis on any profile you have access to.
What software did you use?
Nothing exotic. The 809 MB export was streamed with a standard CSV reader and the Kaplan-Meier estimator was implemented directly — it is about thirty lines. Any statistical package will do it in one call.
Why does an earlier version say 3.1 years?
Because the first pass computed the median over links that had died, which ignores every link still working and answers a different question. Correcting it with proper survival analysis and weighting by domain rather than by link gives 1,080 days. Both numbers are on the survival page with an explanation of what each one measures.