[ffroliva]

Summary

Two $0 probes and a corrected comment. No behavioural change — deliberately.

#692's reporter gets RecaptchaError + exit 1 where they expect FlowHostMigratedError + exit 36. #694 hardened that path, but the failure still cannot be reproduced here, so this PR does not stack a speculative fix on top of it. Per the Iron Law, a fix for a failure never observed is a guess with formatting.

What it does add is evidence that changes what we can tell the reporter — and corrects a comment that would misdirect whoever ports gflow image next.

flow.google.com CAN mint

Measured on ci-probe, on the same page pool the client uses:

route	recaptcha scripts	site key	grecaptcha.enterprise	mint
flow.google.com/ (root grid)	0	—	false	❌ RecaptchaError
flow.google.com/project/<id>	2	6LdsFiUsAAAA…	true	✅ OK, 2404-char token

The mint runs on the pooled bootstrap page, which the client-side handoff leaves on the root grid — and only the root grid carries no script. That is also why the reporter's bundle shows route: "/": diagnostics.py maps both hosts to host_category: flow_app, so the route is the only tell.

So the guard's stated reason — "the migrated host has no recaptcha/enterprise.js" — is false about the host and true only about one route on it.

The outcome is still right, for a different reason

gflow image is UI-driven (mint → drive the composer), and the migrated project composer has no image-generation mode. Enumerated by opening the closed surfaces rather than inferring from the default view:

• the add menu is a media library: movie Scenes · image Images · videocam Videos · upload Upload media
• the settings overlay's radios are dashboard Grid / campaign_all Batch and S/M/L — layout, not mode
• [role=tab] = 0, and no image-generation ligature anywhere

Image generation does exist on that host — but only inside the character editor (see #703).

So exit 36 is correct today. When the composer gains an image mode, the fix is to route the mint to a project page, not to keep refusing the origin. That sentence is now in the code, where the wrong one used to be.

Test plan

• scripts/dev/spike_migrated_recaptcha_mint.py — A/B of both routes, re-runnable, $0
• scripts/dev/spike_migrated_image_capability.py — opens every closed surface (add menus, settings overlay) rather than reading the default view, $0
• ruff, ruff format --check, pyright 0 errors, repo hygiene over 1012 files
• tests/api/test_client_migrated_mint.py — 10 passed (unchanged behaviour, as intended)
• Comment-and-scripts only; no Flow surface behaviour changed, so no e2e applies

Refs #692 — stays open. The reporter's failure is still unreproduced, and this changes nothing about whether #694 fixes it.