Decoding Google Ranking Through the Google and Anthropic Leaks
First in the world to connect Google’s ContentWarehouse leak (May 2024)
AND the Anthropic claude-code repo within a single mathematical reading grid.
Where the global SEO community has analyzed both separately, we show how they answer each other.
14,014 attributes leaked in May 2024. 38 quality dimensions. 17 embedding modules. One thesis: Google penalties multiply, they don’t add up. Here’s the Hi-Commerce × EQOS Forecast study that rewrites what you thought you knew about SEO in 2026.
Reserve my early copy5 public discoveries that change the game — and much more in the full study
These 5 numbers have been public since May 2024 (Google ContentWarehouse leak, Anthropic claude-code repo).
Our study turns them into equations, cross-references a real case (HouseFresh, −95% then a 9% recovery),
and derives a bit-exact verifiable multiplicative model.
The full study goes far beyond these 5 public discoveries. The deliverable details
38 dimensions of the compressed_quality_signals vector modeled one by one, the
17 NSR embedding modules dissected, the formula
Q = baseline × ∏(1 − δ_k) applied to 4 documented real cases, the
Fact / Signal / Calculation protocol that separates what current SEO tools
(Ahrefs 60%, Moz 40%, Semrush 20%) blend together, the CRAPS saturation at 57% computed
via Shannon, and — a world first — the Google ContentWarehouse × Anthropic claude-code comparison matrix
across the 18 PARITY.md sub-modules, never published anywhere else.
In total: a unified mathematical reading grid for SEO ranking + LLM citation, actionable Monday morning, demonstrable in 90 seconds on any tested site.
14,014 attributes identified in Google’s ContentWarehouse
The largest SEO leak in history, the primary source for analyses that rewrite everything we thought we knew about ranking.
38 dimensions in the compressed_quality_signals vector
Google’s quality score is a multi-dimensional vector — each dimension is a distinct ranking factor.
Google penalties multiply, they don’t add up
HouseFresh case (−95% traffic after HCU, 9% recovery): the additive model predicts zero, only a multiplicative model fits the observation.
17 embedding modules inside NSR (Neural Site Ranking)
Google’s neural ranking goes far beyond BERT or MUM: 17 distinct embedding modules, including 2 named explicitly « embedding ». Semantics is everywhere.
18 sub-modules in PARITY.md of the Anthropic claude-code repo
The source code of an Anthropic agent reveals 18 internal parity modules — a strong signal on how LLMs structure their citation of web sources.
5 unpublished results — only in the full study. And more…
On top of the 5 public discoveries, here’s what our proprietary modeling work has proven, and what has been published nowhere else:
The multiplicative formula validated bit-exact on HouseFresh
Our study proves that Q = baseline × ∏(1−δ_k) reproduces the 0.09 recovery score observed on HouseFresh phase 2. The standard additive model predicts 0 (clipping). First critical test passed by a multiplicative model of Google ranking.
CRAPS saturation at 57% — info-theoretic limit of the leak
Shannon computation on the compressed_quality_signals vector: 18 useful bits out of 41.45 theoretical bits. The Google system already reaches 57% of its quality entropy capacity. A measurement never published before.
Google × Anthropic claude-code parity matrix (18 PARITY.md modules)
First bit-by-bit mapping between Google ContentWarehouse modules and the 18 sub-modules of PARITY.md in the Anthropic claude-code repo. Measured overlap OP_11 = 0.63. « Average fit » level, but the pattern is now traced.
Fact / Signal / Calculation protocol (F/S/C) — separates what SEO tools blend together
Data-provenance audit across Ahrefs (60% facts), Moz (40%), Semrush (20%). Our F/S/C method traces every audit line to its verifiable source. A product differentiator, usable Monday morning without changing tools.
4 real cases documented with recovery formula
HouseFresh + 3 additional cases (not publicly disclosed) prove that the dominant multiplicative factor can be identified and lifted within 30–90 days. Each case includes: penalty incurred, formula applied, pre/post measurement, recovery duration.
And more… the list continues in the full study
38 dimensions of the compressed_quality_signals vector detailed one by one, 17 NSR modules dissected, cross-system Google/Anthropic comparison measured at OP_11, blind calibration protocol (P2) on 20 URLs, proprietary Λ scoring for dominant diagnosis, actionable recommendations per identified Google penalty. Over 60 pages of ready-to-use material for an SEO/GEO leader or a consultant who wants to cut through guesswork.
You receive the study PDF exclusively 7 days before the public release. Registrations close at 100 reservations.
You join the e-commerce leaders, heads of SEO and consultants who take AI-driven ranking seriously, starting now.
While they still apply the standard additive model (which predicts zero), you will know how to identify the dominant multiplicative factor.
For the HCU 2023 mystery, the 2025 holiday shopping pullback, the 14,014 leaked attributes. Everything becomes readable with the right grid.
These 5 complete, measured, verifiable results are reserved for subscribers of the study.
→ Reserve my copyReserve your early copy
The full study (formulas, code, 4 documented real cases, F/S/C protocol) will be released soon. Subscribers below receive a PDF copy as soon as it goes live, exclusively 7 days before the public release.
Your data is used only to send the study and a possible interview. No sharing with third parties. For any question: stephane@hi-commerce.fr
The study is produced in partnership with EQOS Forecast
EQOS Forecast is the mathematical modeling tool (Codex Universalis, adaptive multi-scale equational system) that powers the critical analysis of the Google leak and the Anthropic claude-code repo. The EQOS Forecast engine generates vectorized forecasts injectable into any AI (Claude, GPT, Gemini) and makes it possible to test the multiplicative thesis bit-exact against the leak’s docstrings.
→ Reserve my copy of the study