<b>Natural anchor profiles obey a power law, and audits assume a normal one</b>

<b>Natural anchor profiles obey a power law, and audits assume a normal one</b>

There is a quiet statistical mismatch in how we evaluate profiles. Organic anchor distributions are heavy-tailed — a power-law shape where a handful of strings (usually brand and URL) dominate and a very long tail of near-unique phrasings trails off. Audit thinking, by contrast, tacitly assumes a tidier, more balanced spread.

This matters because power-law systems look 'lumpy' by nature. A real profile will have a few anchors with surprisingly high share — and a naive audit flags that concentration as manipulation, when it is exactly what organic accumulation produces.

— Organic: a steep head, a fat long tail, lots of one-off strings.
— Manufactured: a flatter, more uniform set of optimized strings — which is, ironically, the 'balanced' shape audits reward.

On one hand, this suggests the most natural profiles may trip diversity heuristics tuned to expect evenness. On the other, the engine itself was trained on organic distributions, so it presumably learned the heavy tail as normal even if our third-party tools didn't.

Limitation: I'm asserting a power-law shape from how citation networks generally behave; rigorous distributional fits on large anchor corpora are scarce in the public literature.

Open question: should anchor audits test for the right distributional shape — a heavy tail — rather than for arbitrary per-anchor percentage caps that penalize natural lumpiness?