Grouped by what each document is for, not when it was written.
Numbers are permanent — they are how these get linked — so a document never gets renamed or renumbered, and currency is shown as a status instead.
current current kept current revised in place still valid point-in-time, still true superseded by NNN do not quote figures
Start here
The two current documents. Everything else is context.
015The engine now checks its own work
Management view. Why the numbers can be trusted now and could not before — ten defects, none of which threw an error — plus the finding that reframes the roadmap: cognigy.com ranks worse than nice.com and converts 2.5× better, which rules out domain authority and makes this page-level work measurable in weeks. Ends with three decisions we need.
014Twelve queries where we already rank and earn almost nothing
The working document. 786 queries on page two with real volume and near-zero clicks, the 180 queries where our own pages compete with each other (
ai call center: six pages, 40,557 impressions, one click), the eight pages losing ground, and the queries you must open in a browser before treating them as opportunities at all. Every figure queried live on the day.Living reference — updated, not dated
These describe how things work now and are revised in place. If one contradicts a dated report below, this wins.
009How the system actually works
The whole machine, for the people who need to optimise it together: every workflow and the number it decides on, the scoring model exactly, both AI prompts verbatim, how a brief becomes an article and where the machine stops, what runs itself and the 19 steps that don't, and where every dial lives.
001Manager Portal
The briefing and the atlas in one: what the system is, where to look, how to interact with it, the filterable configuration table, and how to add a source. The best starting point for anyone new to this work.
002Manager Dashboard — Integration Spec
Hand-off document for the separate dashboard project: what it may read, the few things it may write, and why it owns no state of its own.
Open asks
Time-bound. This disappears when the asks are answered.
006Three asks, none of them engineering
What is still waiting on a person rather than on code, with the cost of each delay. Revised: the AI credit resolved itself, and the Teams alerting channel is now marked deferred-by-choice — with a four-day silent outage as the evidence for what that choice costs.
Findings and narrative — dated, still standing
Point-in-time, and nothing in them has since been contradicted.
007We are visible in the wrong places
What 500 days of first-party Search Console data actually says. 88% of our non-brand visibility sits below the top three, and careers pages earn 26× the click-through of glossary pages at identical positions — which rules out the domain, the brand and the templates as causes. Includes the method, the limits, and where to confirm each finding by hand.
008The same findings, drawn
Five pictures: the 10M→51k funnel, the 26× gap between page types at identical positions, the six-page pile-up on
ai call center, decay before/after, and the three competing explanations for our zero-click queries. For a one-pass read.013Your brief, and the code it became
The story from the original AI brief to the running system: which specs shipped near-verbatim (the data shape, the five trigger types, the thresholds, the 7am schedule), the one thing we changed on purpose and the number that forced it, the three things built that were never asked for, and why everything still unbuilt is waiting on a decision rather than engineering.
Superseded — kept because the links were shared
Do not quote figures from these. They are here because they were sent to people and those links still have to resolve; each says at the top what replaced it.
012What it watches, what it found, what it still can't see
Status and purpose briefing for SEO and Web managers, written the week Search Console access landed: three findings from the first week of first-party data, what was connected, and four decisions it needed. No workflow numbers.
003From data to action
Source coverage, what SEMrush unlocks, and how a number becomes a prediction, a piece of content, or a decision.
004Response to the brief
Point-by-point answer to the manager's AI brief: what already exists, what was genuinely new in it, and what was declined and why.
005Your diagram, lit up
The manager's own architecture diagram redrawn with live build status overlaid on each box.
011From market noise to a publish-ready brief
How a raw market signal becomes a brief a writer can act on. The clearest walkthrough of the content engine as it stood in early July.
010Remediation Plan: NiCE Signal Spine
The fix list from the July audit of the signal spine — what was wrong, what was changed, and why. Engineering record rather than a briefing.