Most agencies sell you a report. We move the curve.

Methodology

• The Reasonable Surfer Patent

  Google’s Reasonable Surfer patent (US 7,716,225 — filed 2004, updated 2010) assigns link weight by the probability a link is clicked. A link inside a contextually relevant paragraph carries more weight than a footer or sidebar link. Our formula models this: context signals carry 100% of the base score (65% paragraph + 35% page); the anchor only adjusts it by a small signed δ.

  The patent formula

  r(A) = α/N + (1−α) · ( w1·r(B1)/|B1| + … + wn·r(Bn)/|Bn| )

  where w = click-probability weighting for each linking page. A relevant in-body link on a trafficked page produces a high w. A footer link on a zero-traffic domain produces a w near zero. Domain Rating assigns equal weight to both.

  What the leak confirms for the Reasonable Surfer

  In the 2024 Content Warehouse leak, the Reasonable Surfer maps most closely to GoogleApi.ContentWarehouse.V1.Model.AnchorsAnchor. Key fields:

  • pagerankWeight — “Weight to be stored in linkmaps for pageranker.”
  •  sourceType — quality/importance of the source page (HIGH/MEDIUM/LOW). This is the leak’s closest proxy for page-level traffic: a HIGH-tier page is one Google treats as important and regularly crawls — which correlates directly with real traffic. A zero-traffic page will not reach HIGH regardless of its domain’s DR.
  • weight — Google’s internal authority score for the anchor, on a 0 to 127 scale. Higher values indicate more PageRank-style authority transferred through the link
  • fontsize, context, context2, fullLeftContext, fullRightContext, offset, linkTags, locality, bucket — features around the link/anchor corresponding to the
    patent’s weighting by document and positional features.

• Topic-Sensitive PageRank and Site Focus Score

  WLDM scores links against two established ranking signals. The Reasonable Surfer handles page-level traffic weighting. Topic-Sensitive PageRank handles topical cluster relevance.

  Google does not crawl the whole internet for every query. For Nike shoes, the crawler enters the Nike topic cluster, travels a defined distance within that cluster, and ignores 99.9% of the
  web. A backlink from inside that cluster counts. A backlink from outside it barely registers. This is the math WLDM measures.

  The 2024 Google API leak confirmed that Google stores and uses multiple PageRank variants per document:

  PerDocData.pagerank0
  PerDocData.pagerank1
  PerDocData.pagerank2
  PerDocData.crawlPagerank
  PerDocData.toolbarPagerank
  PerDocData.homepagePagerankNs

  Six distinct PageRank scores stored per document. Topic variants (pagerank0/1/2) for bucketed authority weighting. A crawl priority score (crawlPagerank) that determines how
  often Google revisits a page — which correlates with real traffic. A site-level homepage score (homepagePagerankNs). And the old public toolbar score (toolbarPagerank), still tracked internally long after Google stopped publishing it. This is not a single global number. It is a multi-dimensional authority model — and none of it maps to Domain Rating.

  The leak’s direct field names for this signal:

  • siteFocusScore — how topically concentrated a domain is. A site where every page covers the same subject has a high score. A site covering unrelated topics has a low one.
  •  siteRadius — how far individual pages drift from the site’s topic centroid. Low radius means tight topical discipline.

  These two fields define whether a linking site sits inside or outside the topical cluster Google is weighting for your target keyword.

  

• Why cosine similarity matters for backlinks

  Cosine similarity (CS) is a 0 to 1 score for how topically matched a linking page is to your target page. It is an authority multiplier. From the same domain, a link scoring CS 0.6 passes approximately three times the authority of a link scoring CS 0.2. The placement costs the same. The authority transfer is triple.

  The score runs across five tiers:

  • 0.60+ — linking page is specifically about the same topic
  • 0.40 to 0.59 — clearly relevant, strong match
  • 0.20 to 0.39 — broadly related
  • 0.01 to 0.19 — weak connection
  • 0.00 or below — no topical match; zero authority value

  WLDM scores every candidate link before building it. Anything below CS 0.2 is not built. Most agencies do not measure this at all — they filter by DR and call it done.

  The math, precisely

  WLDM’s scoring model is built on the signals confirmed in the 2024 Google API leak. Each text x → a 384-d embedding v(x) (mean-pooled, L2-normalised, so ∥v∥=1). Cosine = dot
  product:

  cos(a,b) = (a·b) / (∥a∥∥b∥) = a·b (since ∥v∥=1)

  Let c = cos(context, target), r = cos(refPage, target), a = cos(anchor, target).

  base = 0.65·c + 0.35·r
  δ = −0.15 if a < 0.15 (mismatch → anchorMismatchDemotion)
  δ = min(0.10, 0.20·a) if c ≥ 0.40 (relevant anchor + strong context →adds)

  δ = −0.10 if a > 0.45 and c < 0.30 (exact-match in weak context → over-optimised)

  δ = 0 otherwise (or generic anchor)
  score = clamp(base + δ, −1, +1)

  Why context gates the anchor: a keyword-rich anchor like “buy gold online” always scores high a against a “buy gold” page — that overlap is circular, not evidence of a good link. The discriminator is whether the surrounding paragraph (c) supports it. Strong c → the match is a natural descriptive link (small bonus). Weak c → the same match is an over optimisation signal (penalty). The asymmetry (max penalty 0.15 > max bonus 0.10) reflects that Google’ link system is risk-averse: a bad anchor hurts more than a good one helps.

  

• Harmonic Centrality

  Stop optimising for fame. Optimise for network position.

  Frankfurt airport moves more passengers than Paris even though Paris is better known. Domain Rating is Paris. Harmonic Centrality is Frankfurt.

  HC measures where a page sits in the authority network — not how well-known its domain is. WLDM scores each target page by CS-weighted in-degree Harmonic Centrality: the sum of all incoming, relevance-weighted edges on a directed link graph. Direct links contribute most. Each extra hop roughly halves the contribution. Only links with a positive CS score count — irrelevant links are excluded from the graph entirely.

  The formula:

  HC(u) = Σ 1/d(v,u)

  where d(v,u) is the shortest path from linking page v to target page u, weighted by cosine similarity. A highly relevant link one hop away contributes far more than a dozen loosely related links two hops out.

  The multi-target publisher advantage

  A domain that places relevant links across several of your target pages accumulates HC multiplicatively — not additively. One publisher with six topically matched placements generates more total HC than twenty single-link domains at similar DR. This is what WLDM’s database is built to find.

  Example from the Stake.com engagement: filmthreat.com delivered 2.44 total HC from 6 placements at an average CS of 0.56. No single high-DR domain with one link came close.

  Head to head: WLDM vs the previous provider (Stake.com)

  On the same account, same pages, measured against the same keywords:

  Metric	WLDM	Previous provider
  Links analysed	603	30
  Average target-page HC	1.59	0.37
  Difference	+330% (4.3×)	—
  Zero or negative CS links	18.6%	30.0%

  The previous provider ran a brand and PR programme: F1 sponsorship coverage, cricket news, lifestyle press. 63% of their links sat in the zero, negative, or low relevance tiers. Brand impressions. Zero authority transferred to the commercial pages that needed to rank.

  WLDM took Stake from outside the top 30 to number one for “online casino” — KD 95, 85,000 monthly searches — in ten months.

• Database extraction

  Our link building database process started over four years ago. We’ve had a data science team for four years now. We spent hundreds of thousands of dollars on database improvement and years refining and developing our process. It currently involves over 600,000 domains x 1,000 passes for each domain that are all vectorised and customised for our link extraction method.

• Quality assurance

  We run the client’s topic through WLDM’s proprietary link database, the largest of its kind for any agency we know, and our machine-learning classifiers, which score and qualify backlink targets at 99%+ accuracy. Instead of filtering by raw Domain Rating, the system maps the client’s entity and topic graph and surfaces the domains and pages that are contextually closest, ranked by semantic proximity and harmonic centrality.

  This is the data-science half of the engine. It sifts thousands of candidate sources in minutes and hands outreach a pre-qualified target list, to get links that are ranking for the keywords you want to rank for. It would take a human 100,000 hours plus to do this manually for each keyword.

• Manual outreach

  Qualified targets go to a human team. Every pitch is written for the specific editor and the specific page: real relationships, real editorial placements, on sites that are ranking for the keyword you want to rank for.

• What the audit delivers — five outputs

  1. ML quality classification of your existing link profile Every domain currently linking to you is run through WLDM’s ML classifier — 5-model ensemble, 99%+ accuracy, trained on 18,000 manually-curated domains. Each linking domain comes back as quality tier or link-farm flagged.

  2. Cosine Similarity scoring against your target pages Every linking URL is scored using the three-signal model (65% paragraph context, 35% page body, anchor δ). Each link returns a CS score and tier — Strong, Moderate, Weak, or Irrelevant/Risk. Links below the build floor are identified as candidates for disavowal or deprioritisation.

  3. Page-level traffic verification Every linking page is independently verified for real traffic. Distinguishes links from trafficked pages — which carry Reasonable Surfer weight — from links on zero-traffic pages that contribute almost nothing regardless of DR.

  4. Competitor Harmonic Centrality gap analysis WLDM builds the HC profile for your top three competitors’ commercial target pages and maps it against yours. Output: which publishers carry the most HC weight in your category, which are linking to competitors but not to you, and which of your target pages have the largest centrality deficit.

  5. 90-day authority roadmap Priority-ordered build plan: publishers ranked by expected HC contribution × topical match × page traffic, target pages prioritised by commercial value and centrality gap, and a projected authority accumulation curve based on engagement arc data from our case studies.

• The Cosine Scorer — run it yourself

  The same scoring model WLDM uses in every audit is publicly available as a free browser tool. Input a prospective linking URL and your target keyword. It returns a live CS score, tier breakdown, paragraph context analysis, and anchor match. Available at cs.wldm.io.