Start here
The newest research note, and the main guide for each of the two big topics.
How RC2’s π table was made
The 256 numbers at the heart of Ron Rivest’s RC2 cipher were said to come from π, but the method was never published. A card shuffle that uses π’s digits as dice reproduces both of RC2’s tables exactly.
October 2026Runs in your browser Field guide · AI searchAI Search Optimization (GEO), mapped
What GEO is, how AI answers get built, GEO vs SEO vs AEO, and a working glossary.
Updated June 202614 min read Field guide · MeasurementHow to measure marketing in 2026
MMM, attribution and incrementality each answer a different question. Give each one a job and let them check each other.
June 20269 min readResearch notes
Old, unexplained questions with answers anyone can check.
How RC2’s π table was made
A Fisher–Yates shuffle that reads π three digits at a time rebuilds both of RC2’s tables, byte for byte. Open since 1996.
October 2026Code on GitHub Software historyThe Office password tables are a CRC
The numbers behind Word and Excel’s old password box are a standard 16-bit checksum, worked out in advance. Copied into LibreOffice, Apache POI and others, without an explanation.
October 2026Code on GitHub Game historyHow Quake’s 162 normals were made
Every Quake model is lit from a list of 162 directions that shipped in 1996 with “no known formula”. An icosahedron from a 1993 OpenGL textbook rebuilds it, line for line.
October 2026Code on GitHub CryptographyThe recycled numbers inside Certicom’s curve generators
Bitcoin’s curve has siblings from the same late-1990s standards. Divided by a small number, their starting points reuse each other’s hidden values.
October 2026Code on GitHubAI search (GEO)
How ChatGPT, Perplexity and Google’s AI choose what to cite, measured on sites I run.
AI Search Optimization (GEO), mapped
What GEO is, how AI answers get built, GEO vs SEO vs AEO, and a working glossary, backed by GrantCompass’s 192,924 AI citations.
Updated June 202614 min read Original dataMost “GEO” advice is wrong. I ran 87 experiments to prove it.
Twelve weeks, 192,924 AI citations, one real website: what makes AI engines cite you, and which popular advice didn’t hold up.
June 202611 min read Case studyHow GrantCompass earned 192,924 AI citations
87 experiments, 25,000+ users in 4 months, no ads or PR, and the caveats about what the data does and doesn’t prove.
June 202610 min read GEO Field Log · Entry 02The citation-vacuum log: zero to cited, measured
A site cited 0 times in 125 runs got 4 of 5 new studies cited within weeks, 3 at position 1, with controls flat.
August 202611 min read GEO Field Log · Entry 01The missing vocabulary of GEO
Seven mechanics of AI search that had no names, from Retrieval Floor to Fan-out Farming, defined from a year of experiments.
August 20269 min read PlaybookHow to show up in AI search: an operator’s playbook
Nine moves that earn citations from ChatGPT, Perplexity and Google’s AI, with a checklist that scores your page.
June 20268 min + checklist Playbook · ChatGPTHow to get cited by ChatGPT
ChatGPT reads Bing, not Google. Six moves that earn citations, and how to check whether it cites you.
June 202610 min read FundamentalsGEO vs SEO: what actually changes
Same foundation, different unit of competition: SEO earns a ranking, GEO earns a citation. What transfers and what breaks.
June 202610 min read MechanicsWhat is query fan-out?
One prompt becomes 8–16 hidden sub-queries, so you get cited for searches nobody types.
June 20267 min read Tools comparedGEO tools in 2026: what’s worth paying for
The three jobs a GEO tool does, what each tier costs (Otterly, Peec, Profound, LumenGEO and the suites), and the $0 stack.
June 20269 min read ReferenceAI search statistics 2026
AI Overviews reach, citation patterns and what moves visibility, with every stat attributed and linked.
Updated June 2026Sourced statsMeasurement
Attribution, MMM, forecasting and testing, as they work inside a real growth team.
How to measure marketing in 2026
MMM, attribution and incrementality each answer a different question and carry a different bias. Includes an interactive measurement triangle.
June 20269 min read PillarMeasuring growth across a mobile + desktop app
One user, three identities (web, iOS, Android) and three systems that never agree. The 5-layer stack that holds it together.
June 202614 min read MethodScrappy MMM: a directional read between the outputs
How to read which way budget should move without a six-figure vendor or a quarterly refresh.
June 20268 min read MethodForecast a range, not a number
Three models that should disagree (funnel, trend and cohort/LTV), combined into a range you can defend.
June 20268 min read MethodYou can run six clean tests a year
Choose incrementality tests like a portfolio: decision value × uncertainty ÷ cost.
June 20268 min readGrowth & positioning
Early-stage growth, positioning, and how teams use AI.
Zero to your first 25,000 users
Finding the one channel that works with no budget, brand or team, with GrantCompass as the case and a first-90-days plan.
July 202611 min read Field guidePositioning: the part most founders skip
The upstream choice that makes every channel and message easy or impossible, plus a live positioning-statement builder.
July 202610 min read Field guideHow marketing teams actually use AI
Every team has the same models; the gap is whether they build with them. A 5-level maturity ladder.
July 202610 min readFree tools
Small tools that came out of the writing above.
AI Citation Checker
Enter a domain and a question your buyers ask AI. See whether you’re cited, how often across runs, and who is cited instead.
About 30 seconds Tool · 14-signal scanGEO Readiness Scanner
Paste any URL for a 0–100 score on how likely AI engines are to cite it, with the fixes ranked.
Instant Tool · Response curvesMarketing Budget Allocator
Set a saturation curve per channel and watch an optimizer split a fixed budget, the same logic an MMM uses.
Instant