Tell people which controls spend money before they press them

[ThinkingSpade]

This app is careful about not spending without a click — useMeteredQuery
disables mount/focus/reconnect refetching so a restored page can't fire a paid
request. This closes the other half: six controls that do spend and didn't
say so.

Look up · up to 1 paid request
Run · 4 paid requests
Build brief · up to 4 paid requests
Fetch reviews · 1 paid request
Update keyword stats · uses credits
Analyze <domain> · may use credits      + "If not cached: est. $0.85."

One helper, not six suffixes

The wording encodes two facts that are easy to get wrong independently.

How much. A count of upstream requests where that's knowable, a measured
dollar figure where one exists, otherwise just "uses credits". Never an invented
number — shared/analysis-costs.ts already says why: showing a guessed price
immediately before spending someone's money is worse than showing none. The
brand estimate reuses that file's constants and its markup rule rather than
hardcoding a figure.

Whether it might cost nothing. Most of these read a server cache first, so a
second press inside the window spends zero. Saying "1 paid request" there would
be a lie in the expensive-sounding direction — which teaches people to distrust
the labels that are accurate. Hence "up to" / "may", and hence the brand price
sitting beside the button as a conditional rather than on it as a promise.

Counts were verified against the actual fan-out, not assumed: Prompt Explorer
dedupes models before spending (new Set(...)), Local SEO caches a successful
profile for 12h, the brief caches 7 days over a 6h SERP read.

Two things deliberately unchanged

• Small rank checks stay as they are. The header already shows a per-check
  estimate, and the <50-keyword threshold deliberately avoids modal fatigue.
• No new blocking confirmations. The app reserves those for 50+ keyword
  checks and batch analysis. These six are deliberate submits, so an inline label
  is proportionate — a modal in front of a cheap, cached click trains people to
  dismiss modals.

Both came from the Codex audit, which correctly refuted one of its own
earlier items once it read RankTrackingDetailHeader.

A caught mistake worth recording

knip passed while meteredEstimateNote was exported, tested, and rendered
nowhere — the test's import satisfied the unused-export check. A green gate
was hiding the fact that the price never reached a user. It's rendered now, and
browser-verified showing If not cached: est. $0.85.

Verification

Browser-verified: four of the six labels render correctly (Local SEO, Prompt
Explorer, Content brief, Brand analyze). Fetch reviews and Update keyword
stats need a business profile and rank data respectively.

7 new tests, 2,129 passing, pnpm ci:check clean.

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	flyrocketseo	3576e7e	Jul 31 2026, 06:45 AM