Skip to content

Link Intelligence & Curation · Essay № 04

How SignalScore Works.

A long-form essay on the formula behind the only link database that refuses to recycle the same DR-90 list your competitors are already pitching. What follows is the methodology, the math, and the editorial QA — written for the senior SEO who plans to audit every line of it.

Essay · 12 min read · Last revised · By the NDAGLinks research team, Tallinn

§ 01 — The Premise

A single comparable number, built from four honest signals.

SignalScore is a 0–100 composite that ranks every indexed URL by how useful it is as an outbound link target for editorial content. It collapses four signal categories — Domain Authority, Topical Relevance, Editorial Standards, and Outbound-Link Velocity — into one number that holds up across niches, languages, and content formats.

We did not invent SignalScore to chase a leaderboard. We invented it because the legacy link databases kept giving us the same DR-90 domains — half of which fail a five-minute editorial sniff test — and we needed a way to surface the long tail of genuinely linkable URLs that nobody else was surfacing. Today, the median SignalScore on our index is 41 and the 90th percentile is 68. Anything north of 80 is, in our experience, where the pitchable inventory lives.

The rest of this essay opens the formula, walks through each of the four signals in turn, shows the curation math that filters the index from 41 million pages down to the 1.4% of URLs we treat as link-worthy, and ends with the benchmark numbers a procurement lead will want before signing off on a trial.

§ 02 — The Four Signals

The ingredients, individually inspectable.

We expose every weighted input so a senior SEO can audit SignalScore rather than take it on faith. Click any chapter number to inspect the measurement.

  1. 01

    Domain Authority

    Weight: 30%

    A Moz-DA-style estimate computed on the live editorial link graph, refreshed every 15 minutes from 2.3 billion outbound links. We do not use a third-party DA score as a black-box input — we recompute the underlying link graph ourselves so that freshly-launched authoritative outlets (think a new Substack from a former NYT editor) earn authority within hours, not quarterly crawls.

    DA is necessary but not sufficient. A domain can score 92 and still be a junk farm — which is why DA is capped at 30% of the final SignalScore.

  2. 02

    Topical Relevance

    Weight: 30%

    A semantic vector match between the prospective linking page and the target URL, computed against a 12,000-label topic taxonomy. The signal rewards topical alignment over raw authority: a DA-55 niche site that publishes exclusively about climate-tech will outrank a DA-90 generalist when your article is about grid-scale battery storage.

    Measured with a fine-tuned embedding model trained on 4.1 million editorially-vetted (source, target) pairs.

  3. 03

    Editorial Standards

    Weight: 25%

    A binary editorial-quality filter applied to every URL in the index. We score 14 checks including: byline presence, corrections policy, fact-checking disclosure, ad-to-content ratio, syndication patterns, AI-generated content markers, and historical accuracy audits. URLs that fail any hard check are removed from the index entirely; URLs that pass are scored on a 0–100 softness scale.

    This is the signal that disqualifies 74% of "high-DR" domains surfaced by legacy link tools.

  4. 04

    Outbound-Link Velocity

    Weight: 15%

    The rate at which a domain acquires new inbound links from other editorial pages, normalized against its publishing cadence. High velocity on a publishing-light site is a strong proxy for "people are citing this because it earned the citation" — and is the single most predictive feature we have for linkability that traditional DA misses.

    Measured over a rolling 30-day window, refreshed every 15 minutes.

SignalScore = 0.30·DA + 0.30·Topical + 0.25·Editorial + 0.15·Velocity

§ 03 — The Curation Math

From 41 million pages down to the 1.4% worth linking to.

We get asked the same question in every procurement call: "If your index is so curated, why is it so big?" The answer is that size is upstream of curation — you cannot filter 2.3 billion outbound links down to a useful list unless you've already indexed everything.

Stage Filter Pages in Pages out Pass rate
Crawl Editorial pages discovered via 41M seed domains 41,000,000 41,000,000 100%
Editorial Standards 14 hard checks (byline, corrections, AI markers, ad ratio…) 41,000,000 10,620,000 25.9%
Domain Authority Floor DA ≥ 25 on the live graph 10,620,000 3,180,000 29.9%
Topical Coverage Mapped to at least one of 12,000 topic labels 3,180,000 2,140,000 67.3%
Velocity Sanity Outbound velocity within 4σ of domain baseline 2,140,000 574,000 26.8%
Pitchable index SignalScore ≥ 60, last refreshed ≤ 15 min ago 574,000 574,000 1.4% of crawl

Of the 41 million editorial pages we crawl, 574,000 survive every filter and become part of the pitchable index you query through the app. That is the 1.4%. Every page you see in a saved list has cleared all five stages in the last 15 minutes.

§ 04 — Freshness & Editorial QA

The infrastructure layer competitors can't easily copy.

15min

Refresh cadence

The full 2.3-billion-link graph is re-scored every fifteen minutes. New outbound links are absorbed within the same window — not on the next weekly crawl.

74%

"High-DR" domains that fail our editorial check

When we benchmarked our top-10,000 DR-90 domains against our own editorial-standards filter, 74% were disqualified — the same domains the legacy tools are still happily recommending.

3/week

Human editorial QA passes

Three times a week, a senior editor audits random samples of the index and the surfaced lists. Failed samples trigger a model retraining and a one-hour suppression window while we patch the underlying issue.

§ 05 — Independent Benchmark

Side by side, claim to claim.

We do not editorialise about competitors — that is not what this page is for. Below is the same benchmarking table we hand to procurement leads who ask "how is this different from the link tool we already pay for?" Numbers are from an independent benchmark conducted Q1 2025 on a 1,000-keyword SaaS link-prospecting workload.

Capability NDAGLinks Legacy DR tools
avg of 2 incumbents
Δ
Time-to-pitchable-list (1,000 prospects) 4 min 12 sec 3 hr 48 min −98.2%
Index refresh cadence 15 min 7 days ~670×
Built-in editorial-quality scoring Yes — 14 hard checks No
"High-DR" domains passing editorial standards 26% ~3% ~9×
Customer NPS (Q1 2025) 71 ~34 (industry avg) +37 pts
Average time saved per session 3 hr 12 min n/a

Benchmark conducted independently by the European Search Awards judging panel, Q1 2025, on a 1,000-keyword SaaS link-prospecting workload. Full PDF available on request to enterprise evaluators.

End of essay

Now audit it yourself.

The Pro trial gives you the full SignalScore index, the editorial-standards filter, and unlimited list exports for seven days. No card required for the first three days; cancel from settings in two clicks.