Reply to thread

Message: [QUOTE="Borrowdale Barron, post: 90433, member: 6"] Fair challenge, and it's the right one to press on. The difference isn't the feature list. Feature lists converge; anyone can bolt on a crawler. The difference is what the tool is allowed to say to you. Every product in that category eventually hands you a score. A health grade, an authority number, an SEO score of 68 out of 100. That number is a guess about a black box. Nobody outside Mountain View knows how the ranking systems actually work, so a composite score is a claim that can never be proven wrong. Which also means it can never be trusted. That exact class of metric is the thing this product exists to replace. The charter is therefore narrow on purpose. Don't try to replicate Google's ranking. Build algorithms that answer questions Google doesn't answer for a site owner, that run on data Google doesn't expose to anyone else, and that can be proven on your own site. The data nobody else has is your own. Your Search Console history, your analytics, your full crawl, and your record of what you changed and what happened afterward. That last one is the asset, and it only exists if something bothers to keep it. Four rules decide whether an algorithm ships at all. Every score must break down into named, inspectable inputs with no hidden constants, so it's traceable. It must produce one sentence of cause and effect in plain language, generated from that breakdown rather than handwritten per case, so it's explainable. It must end in a specific change to a specific page or template, so it's actionable. And a later crawl or data import must be able to prove the prediction right or wrong, and must actually do it, so it's checkable. Failing any one of the four is a blocking defect, not a backlog item. Flowing out of that, there's a list of outputs the product is structurally forbidden from producing. No composite site score. No metric without a breakdown panel, no prediction without a confidence band, no recommendation without a verification plan. Which answers your second question. The verification isn't clever; it's bookkeeping plus patience. When a recommendation gets made, the expected effect and the way to measure it are recorded at that moment, before any work happens. Later, the affected pages get re-crawled, re-queried, or re-measured, and the evidence is attached back to the original decision. Right or wrong, it goes on the record. If a condition that was previously fixed comes back, the work item reopens itself and creates linked follow-up work. Nothing quietly closes and stays closed. The reason competitors don't do this isn't laziness; it's that they never observe the outcome. They sell the diagnosis and leave before the surgery. Once you keep outcomes, you can ask the question that's actually worth money, which is which types of work and which playbooks reliably produced results, and which merely generated activity. There are six named algorithms in the first register, and they all have this shape. Equity Flow is personalised PageRank over your internal link graph, except the teleport vector is weighted by business value instead of being uniform, so authority is measured against where the money actually sits. The maths is E = (1−d)·v + d·Mᵀ·E, where v blends conversion value, entry sessions and manual priority, and the edge weights in M come from link position, prominence, rel attributes and whether the link even renders. If you haven't connected analytics, v falls back to uniform, and the whole result gets stamped UNCALIBRATED rather than quietly pretending otherwise. Stranded Authority Index finds pages that earned attention but can't be reached or can't pass it on. Template Leverage Score works out which single component edit has the largest structural consequence, which matters enormously on a templated site where one change hits ten thousand pages. Cannibalisation Resolver decides which of your own competing pages should win. Substitution Risk measures which demand is being answered by AI without anyone clicking through, which is the live existential question for publishers right now. Decay Velocity catches pages sliding before the loss shows up in any aggregate number. All six emit into the same work item contract and the same verification loop. That's the actual product. Not six reports, one pipeline with six entry points. [/QUOTE]

Name