The insight was true, the theory too neat
An AI caught one correct local judgment and completed it into a whole theory. How a devil's-advocate prompt took the pretty version apart, and what survived.
All content
An AI caught one correct local judgment and completed it into a whole theory. How a devil's-advocate prompt took the pretty version apart, and what survived.
16GB with WSL2 and VMware: chasing WSL, nonpaged pool, WebView2 and Widgets — and Apply/Restore made no difference. Chrome was always the answer.
If models get easier to swap, what is actually worth accumulating in an agent? The model supplies capability; the harness makes it engineerable.
From in-platform search to the AI answer layer: why being indexed no longer means being clicked, and why SEO is shifting toward becoming a citable source.
A de-identified Search Console case: why a strong average position can be an artifact of one impression, and which signals show a ranking is actually stable.
Turning one soft rule into a verifiable gate: where the criteria come from, which lifecycle point, what it can see, and four layers of verification.
A method never run on a client: one question that filters vanity metrics, a three-layer report structure, signal-based staging, and a worked example.
Not a guarantee of correctness, and not "I tried hard." Four levels of evidence behind a claim, and the hard boundaries you should not cross alone.
With a dozen English sites, the conflict is not duplicate keywords — it is that no domain, page or person owns the demand. A framework from GSC to 301.
The execution kit for multi-site SEO governance: who arbitrates conflicts, how the 30 days are scheduled, and three copy-ready templates.
Claude Code, agent swarms and AI bug reports all raise the supply of candidates. The scarce input is verifying results, ranking them, earning trust.
Cloudflare's World Cup traffic data shows a major event rearranges attention by time zone, channel and match phase — not a uniform traffic windfall.
I want neither a fully autonomous agent nor one that confirms everything. The real problem is permission boundaries and interrupt density.
A de-identified snapshot from an SEO delivery group's 50+ B2B export site portfolio, by industry code — and a plain statement of what it does not claim.
Version one should not be an application bot. What actually shipped: a Markdown kanban where the column is the only source of truth.
Everyone can bake bread, and bakeries have lasted five thousand years. On Sakana Fugu and why buyers pay for convenience and accountability.
One iPhone screenshot, one afternoon, a dozen commits: desktop TOC, mobile pill, a reversed test verdict, footer rebuild. A fix counts when behavior changes.
WebRTC can expose your real IP without a normal request, invisible in DevTools. The mechanism, test pages, and a console script you can paste and run.
A four-layer diagnosis — qualified traffic, page capacity, conversion tracking, sales feedback — and the minimum fields an agency monthly report must contain.
From yesyes.qzz.io to eigentime.org across two Cloudflare accounts: an honest log of every trap — deploy landmine, silent www redirect, shadowing secret.
An article genuinely behind a paywall: section 01 is a free preview, the rest unlocks over x402 with testnet USDC. No real funds involved.
A local time logger that does one thing: show where the hours went. Zero dependencies, no build step, and 14 root-cause postmortems.
GEO is not keyword stuffing for AI search. It is making content easy for answer engines to understand, cite and verify — with 2026 data and two corrections.
A practical framework from an SEO delivery group's 50+ site portfolio and my planning work on about 20 projects, with clear ownership and result boundaries.
A de-identified SEO retrospective of a new B2B UAV export site: what I did in the planning phase, and an honest split between my work and a news cycle.
A technical SEO teardown of a B2B export site: the sitemap declares 254 URLs while search engines know 5,794. The problem lives outside the sitemap.
AI never changes code at will: verifiable boundaries first — an AGENTS.md, a content schema, a publishing gate, and a CMS that can only propose.
Why I started this bilingual personal blog: what I'll write, what I'll skip, and how I want it to grow over time.
A reference list for studying information architecture, navigation, archives, interactions, and long-term content structure.
A template article for checking prose typography, links, lists, tables, code, quotes, and image zoom behavior.
Turning the blog idea into a publishable structure with bilingual routes, content collections, search, RSS, comments, and Cloudflare deployment.
An Astro-based bilingual personal blog with Markdown content, static search, RSS, SEO, and Cloudflare deployment.