Links That Count: Five Link Tactics Against the Leak and the Patents

01What the post rests on

A guide by Cyrus Shepard, published on 20 October 2026, claims to show how to build links that Google cannot ignore. It names five tactics, namely topic stability, anchor context (the words around a link), editorial diversity, top-tier links with "magical" power and links that carry new information. It supports them with Google patents, leaked attribute descriptions and trial evidence. Every tactic points at a real artefact. This article asks how far each artefact carries.

Patents and leaked attributes prove different things. A patent describes what Google could build. A filing proves neither that it was built nor that it still runs. A leaked attribute proves that something is stored, sometimes with a one-line comment, but never with a weight or a threshold. Where the post moves from a field to an effect on rankings, that step is interpretation. Some of those steps hold up well, and this article marks each of them.

02Top-tier links: a PageRank bucket on onsite anchors

The post's lead example is one attribute comment, and both the comment and the scale behind it exist. The comment says that anchors from pages with a PageRank above 51,000 are the best and anchors below 47,000 are all the same. PageRank is Google's link-based score for a page. On every anchor's source record the leak stores the linking page's PageRank on a 16-bit scale, from 0 to 65,535, and beside it the linking page's NSR, a site-quality value between 0 and 1,000. A link arrives at its target with its source's standing attached.

The comment describes only onsite anchors, not the external backlinks the guide applies it to. It belongs to topPrOnsiteAnchorCount, a counter in a document's anchor statistics, and the name says onsite. Its siblings for off-domain and on-domain anchors exist without any description. The comment documents a quality bucket with a sharp top band and a flat floor. It does not document how much a top-band link moves a ranking. The guide's reading, that the rule describes the external backlinks one builds, skips the one word that limits it.

03The trusted-anchor test: news, not .gov

The second "magical" claim is the post's most interesting find, and its direction matches the leak. The anchor spam penalizer, Google's system that demotes manipulative anchors, stores per trusted site a score comparing that site's anchor text with anchors classified as spam. A positive value means the trusted anchors resemble the spam anchors, which in the field's own words means the spammy anchors are legitimate. The guide reads this as one strong link rescuing a hundred weak ones. The arithmetic is the guide's.

The same records change the picture in two ways. First, "Trusted" is defined: the fraction of pages with newsy anchors on a site, above zero for trusted sites. That describes news-like linking, not the government or university domains the guide imagines. The .gov example comes from a 2003 Google patent, not from the leak. Second, the per-source comparison is filled only when a debug flag is set. What a document carries otherwise are counts of trusted sources and matching anchors, a spam probability and a combined penalty. It also carries a flag that records whether all anchors in a demotion period were demoted, not only the spammy ones.

04Anchor context: a hash, a patent, a guideline

Anchor context, the words around a link, is the best-supported tactic in the post. Patent US8577893B1 by Anna Patterson and Paul Haahr, filed in 2004 and granted in 2013, takes a window of text on each side of a link, five words in one implementation. It picks the rarest word in each window and hashes the pair into a context identifier. A target can then be ranked by how many different contexts point to it, and a context with a disproportionate count is discounted as possibly machine generated. The leak stores context2 on every anchor, described as a hash of terms near the anchor, together with the full left and right context.

Google confirms in its own documentation that the words before and after a link matter, but the link between patent and leak field stays open. The Reasonable Surfer patent lists the words around a link among the features that predict whether it gets clicked, next to its position in running text, a footer or a sidebar. The leak says context2 is a hash of nearby terms, not how it is built. The patent's measure is how many different contexts point at a page, not how descriptive each one is. Varied sentences across many linking pages fit that measure. A longer sentence on one page does not obviously add to it.

05Editorial diversity: spam phrases and spikes

Google's history-data patent family, with priority in 2003, treats suspicious anchor patterns as a sign of coordination. "Deliberately different anchors" is a phrase from it. The patents describe naturally grown link graphs as the result of independent decisions and synthetic ones as coordinated. That makes the growth of anchor words, bigrams and phrases spiky, either through identical anchors from many documents or through deliberately different anchors from many documents. The patent lets the engine cap the impact of suspect anchors.

The leak shows that the anchor spam penalizer measures timing directly, which supports the guide's warning about campaign timing. It records how many spam phrases appear among unique domains, their share of all anchors, the average daily rate at which spam anchors are discovered, and over how many days 80 percent of them appeared. An older structure, now deprecated, tagged anchors created during a spike. The fields do not show a comparison of anchors with each other: no similarity score, no vectors. They count phrases classified as spam. Treating "cheap flights" and "low-cost flights" as the same move is the guide's extrapolation.

06Repeated links: a cap, not a zero

The rule that unique links beat repeated ones has the clearest wording in the leak, and it is a cap. When many anchors arrive from one domain, all but 200 are thrown out, recorded per domain and anchor text. A separate counter tracks domains above a per-domain phrase cap. Every anchor also stores how many additional links the same source page sends to the same target domain.

Two qualifications limit the guide's version of the rule. First, the documented mechanism is a cap: 200 anchors per domain survive deduplication, so the guide's graphic of 20,000 sitewide links marked "capped/ignored" is right about the direction and loose about the number. Second, repetition is also a policy matter. The reference-context patent names standard frames, the same links on every page of a site, as a way to inflate rank. Google's spam policies list widely distributed footer and template links as link spam. Deduplication is bookkeeping. The spam policy is about intent.

07Topic drift: a 2003 patent and three counters

Topic stability, meaning that a site keeps the subject its links were given for, has the most explicit patent support of the five tactics. The 2003 history-data family addresses the case where a page's content changes until it no longer matches the anchor text of its backlinks, for example when an expired domain changes hands. The proposed fix is to estimate the date the focus changed and to ignore or discount all links and anchor text from before it. A significant change in a document's topics likewise makes earlier indicators such as score and anchor text unreliable. Google's spam policies treat expired domain abuse as a category of its own.

The leak shows the bookkeeping for topic drift, not the rule. A forwarded anchor keeps the docid of its original target and a record of how it was forwarded to the canonical. A document's anchor statistics hold counters named linkBeforeSitechangeTaggedAnchors, pageMismatchTaggedAnchors and pageFromExpiredTaggedAnchors. The first two have no description, and the third only the function that sets it. The names fit the patent's idea closely. What triggers the tags and what follows from them is not documented. The guide's reading that these fields directly target a site's link history says more than they do.

08Where E-E-A-T comes in

Most of the fields checked here decide which anchors count before anything is weighed. Anchors enter topicality, Google's measure of how well a page fits a subject, and PageRank feeds the Quality score. In between, the anchor statistics carry a page-level Penguin penalty and a flag for documents protected by the goodness of their early anchors. Penguin is Google's system against link spam. Link history therefore works in both directions.

The E-E-A-T connection runs through trust rather than authority. These systems test whether the link signal itself is authentic: independent decisions, distinct contexts, no spikes, news-type sources vouching for similar anchors. Authority is touched only indirectly, through PageRank's role in Q*, Google's page-quality score. The practical contrast is concrete. A regional weather service cited in twenty news reports on a storm, each in its own sentence, produces the pattern these fields treat as trustworthy. Twenty thousand footer links from a partner network with one anchor text produce a capped count and a policy problem.

09References