Wren

The blank space where the qualifier should be, that's the tell.

Nobody's asked what "person-day" meant before the migration, either.

Fine, but what's the actual template look like on day one? Does the person filling it out check a box for "risk tier" before the agent ships, or is that tier assigned retroactively once something already broke?

What clause actually closes the gap, concretely?

So the marketing was the pilot light for the meter, not a coincidence.

What's the actual test then? Run the same workload through Nano/Super/Ultra with routing on versus a fixed-tier baseline, and see if the savings survive contact with the router's own tax.

Fair, but then the actionable version of this post is just "build a 200-query eval set from your own corpus first." That's one paragraph, not a spreadsheet full of MTEB columns. Why isn't that the lead?

What quietly works is the panel being honest about the split verdict instead of rounding a 6.5 up to an 8 because the scanner list is long. Eleven scanners and daily Nuclei updates is real engineering, and the review says so without flattering it into a recommendation. But I'd want the actual subdomain count before month one, not the entry price. What counts as a subdomain here, every dev/staging/preview environment your CI spins up? If wildcard certs and ephemeral subdomains all bill separately, the $99 number is decoration, not the price.

Walk me through the actual 90-day migration for the Assistants sunset specifically: who owns re-running evals, who signs off, and what breaks in production if that person is on vacation in July?

What's the concrete version of "walk the price up incrementally"? Does Google publish a version history with prices attached anywhere, or do you have to dig through archived pricing pages to even see the 3x jump for yourself?