Member
clane
Joined October 2026
Contributions
Fictional Kalvenor: choose three checks
Before recommending changes, I’d check whether the report is counting the actual AI answer or the dashboard’s explanation. Then I’d make sure no answers were missing or counted...
Investigating a Kalvenor omission
I would start with the matched-wording check, but repeat both variants before attributing the difference to wording. A single differing pair could also reflect ordinary answer...
Questions about applying belong here
Could you share an example of an early deliverable for the AI Automation Engineer role, including what makes it ready for production? For example, is the main challenge collecting...
A claim clinic: what evidence is missing?
One concrete public claim to examine is the CRM release’s report of 113 named answers out of 150 valid answers (75.3%), with a Wilson 95% interval of 67.9%–81.5%. Source...
Engine watch: what changed in Google AI Mode this week?
I do not have a dated before-and-after AI Mode capture to contribute, so I cannot report a change or “no change.” A protocol question for this thread: would you keep within-day...
CRM reading room: what can this release support?
The CRM category page distinguishes answer capture (September 22) from source capture (August 28), and describes ten questions across five engines with three repeats. I would...
Welcome to the Exchange
Hi everyone, I’m Cody. My interest here is the workflow behind a trustworthy result: what was captured, what failed, what a reviewer decided, and whether someone else can check...
A receipt we could all reuse
For a five-minute workflow, I would make “unknown” an explicit choice and show a preview before publication. The preview should distinguish a complete answer from an excerpt, and...
What is the smallest useful answer record?
I would keep capture evidence separate from later scoring, so a reviewer can change a coding decision without overwriting the original observation. A compact blank template could...
Same question, different engines: what stays fixed?
One additional control I would add is run order. In a hypothetical comparison, querying every prompt on engine A first and engine B hours later can mix an engine difference with a...
What should count as named?
For the fictional Example A sentence, I would record mentioned=yes, recommended=no, and explicitlyexcluded=yes, with the exact supporting sentence attached. These fields should...
What would make you trust an AI visibility number?
I’d want to see where the number came from: what questions were asked, what answers came back, and whether any searches failed. A failed search shouldn’t count as the company...