Every change to the reference layer, with evidence and rationale
v1.21
Pillar pages: the reader’s question first, the machinery last
Foundation ·
Added[O] Official[C] Interpretation
All five pillar pages reordered: The question → What Google goes by → What this means for your page → What you can do → The systems → Deep Dives. The third block surfaces `konsequenz` from systems.json, which was written years ago, exists in both languages and had never appeared on a pillar page. The fourth needs a new `dims` field on the eight Helpful Content practices in bestpractices.json; the mapping is marked [C] on the page. New per-dimension reader questions in pillars.json; the /pillar/eeat hub shows all four as its entry point instead of system-count badges. Selection rules moved to src/lib/pillars.ts with a test.
Why: The pages opened with an intro paragraph that names system internals from its second sentence, then stacked cards of `wasMisst` — the mechanism. A reader who does not already know what chard is drops out there. The answers were in the repository the whole time, in the wrong order: mechanism before consequence. Two data-thin spots now say so rather than looking broken — exactly one of Google’s eight public practices addresses Experience directly, and the capped consequence list names how many systems it leaves out.
v1.20
Topic hubs and a Deep Dive menu grouped by function class
Foundation ·
Added
Ten new pages: /topic/<slug> and /de/thema/<slug> for the five function classes (behavior, quality-prediction, penalty-spam, index-infrastructure, freshness), each bundling that class’s systems, deep dives, leak fields and theses. The header carries the deep dives as a nested <details> menu — five topics with counts, expanding to their articles, no JavaScript, all 27 article links plus five hub links in the static HTML. /deep-dive and /de/deep-dive are grouped by the same five classes with anchors. The taxonomy moved out of the 108 KB homepage component into src/lib/klassen.ts with a test.
Why: 27 deep dives in one flat list is a catalogue, not navigation. The function classes were already in systems.json with order, colours and translations, and the homepage catalogue already filtered by them — the taxonomy existed, it just had no pages. Topic hubs are the second axis next to the E-E-A-T pillars: a pillar answers "which dimension", a topic answers "what does this system do". Two articles span two classes and are listed on both hubs, with a note; menu and index give every article exactly one home.
v1.19
contentEffort checklist: a preface crediting the texts that prompted it
New section 01 in /deep-dive/content-effort-checklist, in both languages: a plain summary of this article, of Cyrus Shepard’s "Content Effort: The Google Ranking Feature Nobody Talks About" on Zyppy Signal, and of the LinkedIn post by Anam Hassan that brought it back into view — with thanks to both. Two claims carry the links, so both sources sit in the reference list rather than only in the prose. Article version 1.1.
Why: Industry coverage of contentEffort reads the same leaked field from the practitioner’s side and ends in a pre-publication audit; this article says what is documented, what is official and where the reading starts. They answer different halves of the same question, and a reader is better served being pointed at both. Both sources are trade analysis, so they are coded [C], not [B] — the leak line itself is the only [B] here.
v1.18
ALL 25 SYSTEMS HAVE DEEP DIVES — Phase 04 content complete
24 Deep Dives covering ALL 25 systems. 327 claims with evidence mapping across E-E-A-T dimensions. Every system from NavBoost (S1) to Mustang/SuperRoot (S22) has analytical content with claim-level evidence codes. Build: 93 pages.
Why: Schicht 2 (Deep Dives) is now complete. Every system in the reference layer has corresponding analytical content. The 327 claims cover: behavioral signals (NavBoost, Glue, Chrome), quality calibration (IS-Calibration, chard, OriginalContentScore, contentEffort), spam detection (Anchor-Spam, SpamBrain), freshness (FreshnessTwiddler, Datums-Triangulation, hostAge), and infrastructure (Mustang, SegIndexer, Tangram).
v1.17
Deep Dives 11-13: OriginalContentScore, contentEffort, Topic-Embeddings
Three Deep Dives covering the content quality architecture: /deep-dive/original-content-score (13 claims, Experience primaer — measures content originality), /deep-dive/content-effort (11 claims, Expertise primaer — estimates editorial effort), /deep-dive/topic-embeddings (13 claims — focus vectors + radius for topic relevance). Together these complete the content quality picture: originality (what's new), effort (how much work), and relevance (what it's about).
Why: These three systems form the content quality triad: OriginalContentScore (originality), contentEffort (effort), and Topic-Embeddings (relevance). Together with chard (quality) and IS-Calibration (ground truth), they form the complete content evaluation pipeline from human raters to final ranking.
Why: These three systems form the quality calibration chain: IS-Calibration (ground truth from human raters) feeds chard (page-level quality model), which feeds NSR (site-level quality). SpamBrain operates independently as a terminal spam detection signal. Together with the existing 7 Deep Dives, this covers the complete quality architecture from human raters to final ranking.
v1.15
Sixth + Seventh Deep Dives — FreshnessTwiddler + Datums-Triangulation
Two Deep Dives sharing the same patent (US7346839B2): /deep-dive/freshness-twiddler (17 claims: QDF, Instant Glue, content update scoring, staleness detection) and /deep-dive/date-triangulation (17 claims: six inception date sources, multi-source triangulation). Both have Stufe 2 patent bridges via Cutts/Haahr.
Why: These two systems are architecturally linked: Datums-Triangulation (S25) feeds date information into FreshnessTwiddler (S24). Together with hostAge-Sandbox (S19), all three share the same foundational patent (US7346839B2) but through different claim sections — demonstrating how a single patent can cover multiple ranking systems.
v1.14
Fifth Deep Dive — Anchor-Spam (Penguin-Erbe)
Deep Dive ·
Added[P] Patent[B] Leak field[C] Interpretation[O] Official
Fifth Schicht 2 article: /deep-dive/anchor-spam. 5 sections covering two-layer detection (phraseAnchorSpamPenalty + IsAnchorBayesSpam), Google bombing era, patent mechanism (US8577893B1 reference context analysis), ranking architecture, and SEO implications. 19 claims with evidence mapping.
Why: Anchor-Spam has a Stufe 2 patent bridge via US8577893B1 (Patterson + Haahr). The patent describes an elegant context analysis approach: instead of looking at anchor text (which spammers control), analyze the surrounding words (which they don't). Paul Haahr appears in DOJ email chain exhibit.
Fourth Schicht 2 article: /deep-dive/hostage-sandbox. 5 sections covering passive damping mechanism, patent analysis (US7346839B2 Claims 38-40: domain expiration, DNS monitoring, name server quality), inception date formula (H = L / log(F+2)), ranking architecture, and SEO implications. 18 claims with evidence mapping.
Why: hostAge-Sandbox has a Stufe 2 patent bridge via US7346839B2 (inventors include Matt Cutts and Paul Haahr). The patent describes exactly the domain legitimacy signals that hostAge evaluates. The inception date formula provides a mathematical explanation for the 'hockey stick' growth pattern observed in new sites.
Bidirectional cross-links: System pages now link to related Deep Dives and E-E-A-T Pillars. Deep Dive pages link to related Pillars. Chatbot corpus expanded with 65 Deep Dive claims (evidence-mapped) and 5 Pillar hub pages.
Why: Cross-links complete the hub-and-spoke architecture: users can navigate System ↔ Deep Dive ↔ Pillar in any direction. The chatbot corpus expansion means the AI assistant can now answer analytical questions using Deep Dive content, not just raw system data.
5 pillar hub pages: /pillar/eeat (overview), /pillar/experience, /pillar/expertise, /pillar/authority, /pillar/trust. Each connects primary and indirect systems to Deep Dives. Hub-and-spoke architecture complete.
Why: Pillar Pages are the missing architectural link between Schicht 1 (system pages) and Schicht 2 (Deep Dives). They organize the 25 systems by E-E-A-T dimension and provide navigation hubs for users exploring the reference.
Third Schicht 2 article: /deep-dive/q-star. 6 sections covering the DOJ-revealed two-signal architecture (Q* quality + P* popularity), NSR chunk-based computation, Q*/P* interaction, hand-crafted signal design (PXR0356), and SEO implications. 18 claims with evidence mapping.
Why: Q* and P* are the architectural backbone: every other signal feeds into one of these two pillars. The DOJ trial revealed this bipartite structure, and the API leak confirmed it with field documentation (siteAuthority converted from quality_nsr, fedBy/feedsInto chains). Understanding this architecture is essential for any evidence-based SEO practice.
Second Schicht 2 article: /deep-dive/panda. 8 sections covering Panda's site-wide demotion mechanism, the content farm era, the Navneet Panda name bridge (inventor name = update name = Stufe 1), patent analysis (6 Stufe 1 patents, 1 Stufe 2 Ofitserov dwell-time), BabyPandaV2 evolution, and SEO implications. 24 claims with evidence mapping.
Why: Panda has a unique evidence chain: the inventor name (Navneet Panda) matches the update name, creating a Stufe 1 patent bridge without needing DOJ testimony. The DOJ connection is indirect (HJ Kim mentioned content farms in PXR0356). Combined with leak fields (Panda, BabyPandaV2) and official Google communication, this makes Panda one of the strongest-documented systems.
First Schicht 2 article: /deep-dive/navboost. 7 sections covering NavBoost mechanism, click signals (goodClicks/badClicks/lastLongestClicks), 13-month window, DOJ trial revelations, patent analysis, ranking architecture, and SEO implications. 23 claims with evidence mapping.
Why: NavBoost has the strongest evidence chain of any system: DOJ PXR0356 confirms Kim as developer (Stufe 1), 3 patents as [P], leak fields as [B], DOJ trial testimony as [A]. The Deep Dive demonstrates source discipline at claim level — every assertion traces to a source with an evidence code.
v1.7
Agent Rank patent review — entity-level gap documented
Patent ·
Reviewed
Agent Rank patent family (US7565358B2, US8296293B2, US9002856B2, US20070033168A1) reviewed. All Google LLC, inventors Minogue/Tucker. No system mapping — entity level missing from 2024 API leak.
Why: Google patented entity-level reputation ranking in 2005, but the 2024 API leak contains no entity-level system. Documented as a finding: either the entity level exists but wasn't captured in the leak, it was never implemented as described, or entity signals were absorbed into site-level systems (Q*/siteAuthority).
US8577893B1 'Ranking based on reference contexts' added as [P] to System 17 (Anchor-Spam / Penguin-Erbe). Inventors: Anna Patterson, Paul Haahr. Stufe 2.
Why: Patent explicitly combats anchor text spamming and Google bombing via context analysis: rare words in left/right text windows of links, context identifiers, distribution analysis. Suspicious patterns (one context with 10,000 count vs. 10/4/1) are discounted. Paul Haahr appears in DOJ email chain exhibit.
Discarded
US7953763B2, US8250114B2 (Google, Stufe 3): link farm detection via PageRank derivatives — link structure, not anchor text. US7533092B2 (Yahoo, not Google): TrustRank line. US7349901B2, US20090198673A1, US20110087648A1 (Microsoft, not Google).
Why: Assignee check was critical: 4 of 7 patents were NOT from Google. The Yahoo patent (Berkhin/Gyongyi/Pedersen) is the TrustRank line — related but not Google's implementation. The Microsoft patents cover email-spam crossover and forum mining.
US11354342B2 family (4 patents, Google LLC, inventors Carbune/Anders) reviewed and discarded. Stufe 3 — functional similarity only.
Why: Patent describes session-relative information gain (how much new info a document adds vs. what the user already saw in this session). OriginalContentScore measures corpus-absolute originality (how much original content vs. the entire web). Different scope, different mechanism, different purpose.
v1.4
hostAge/Freshness/Datums-Triangulation patents — one patent covers three systems
US7346839B2 added as [P] to System 19 (hostAge-Sandbox). Claims 38-40: domain legitimacy scoring (expiration date, DNS records, name server age). Stufe 2.
Why: Patent describes domain expiration as legitimacy factor ('valuable domains are often paid for several years in advance, while doorway domains rarely are used for more than a year'), DNS record monitoring, name server quality assessment. Matt Cutts as inventor = head of webspam team.
US7346839B2 added as [P] to System 25 (Datums-Triangulation). Claims 2-5: multi-source inception date. Stufe 2.
Why: Patent describes exactly the triangulation mechanism: inception date determined from (a) crawl/index date, (b) first link date, (c) domain registration date, (d) first news reference, (e) page count threshold, (f) server timestamp. Child patent US7840572B2 is more specific. Paul Haahr as inventor has DOJ email chain exhibit.
v1.3
Panda patent layer — 7 patents as [P], inventor = update name
7 patents added as [P] to System 16 (Panda): US8682892B1, US9195944B1, US9031929B1, US9760641B1, US9767157B2, US9684697B1, US10055467B1. 6x Stufe 1 (inventor Navneet Panda = update name), 1x Stufe 2 (Ofitserov, dwell-time).
Why: Navneet Panda is the inventor on 6 of 7 patents — his name matches the Google Panda update. The patents describe site quality scoring from query/click ratios and content farm combat. DOJ bridge: HJ Kim mentioned content farms in PXR0356, Panda was built for that purpose.
v1.2
NavBoost patent layer — 3 patents as [P], DOJ confirmed
3 patents added as [P] to System 1 (NavBoost): US10229166B1, US8661029B1, US8938463B1. All Stufe 1 — inventor Hyung-Jin Kim confirmed as NavBoost developer via DOJ Exhibit PXR0356.
Why: DOJ Exhibit PXR0356 (Feb 18, 2025 call with Google Engineer HJ Kim) states: 'Navboost. This was HJ's second signal project at Google. HJ has many patents related to Navboost.' Kim is inventor on all 3 patents. Stufe 1: inventor identity + DOJ testimony = strongest possible evidence.
Discarded
US9092510B1 discarded (Stufe 3): different inventors (Stets, Paskin), no Kim connection. Temporal component could relate to FreshnessTwiddler but functional similarity only.
Why: Stufe 3 — no inventor match, no DOJ connection, only functional similarity (temporal element of user feedback).
v1.1
Interactive tools — Myth or Proven lookup + corpus chatbot
Tool ·
Added
/myth-or-belegt: client-side fuzzy search against systems.json + fields.json with evidence-code badges. /ask: chatbot with full corpus in system prompt (~31k tokens, no RAG), direct fetch to Anthropic Messages API (claude-3-5-haiku-20241022), rate limiting (10 msgs/hour/IP).
Why: Myth or Proven is the viral lookup tool (gratis, no API key needed). Chatbot is Sensor 2 of the Living Answer Layer loop — it captures user questions for seed-query expansion and demonstrates corpus-grounded AI responses.
25 systems documented from Google API leak (May 2024) with [B] field references. 38 fields documented in fields.json. 8 best practices in bestpractices.json. 50 system pages (25 EN + 25 DE) with SSG via generateStaticParams. Interactive EEAT Mechanics map with searchParams-based filtering.
Why: Schicht 1 (Reference) is the foundation — belegpflichtig, never generated. All 25 systems are documented with evidence codes [B] (leak field), [O] (official Google communication), [A] (DOJ/eid material), [C] (interpretation). The interactive map allows filtering by pillar (E, E, A, T, NavBoost...) and evidence strength.