[Flotapponnier]

The flagship privacy board's spec, with the size selector you asked for: three rungs at 0.001, 0.01 and 0.1 BTC.

That selector is not decoration. Measured by hand on 2026-10-25, ZEC through Maya came back at 77.5 bps on the small rung, 40.7 on the middle one and 325 on the large one. The fixed withdrawal fee owns the small end and pool depth owns the large end, so the cheapest size is in the middle. A board showing one notional would have reported the shape of its own choice.

Held out of production, on purpose

In DEV_ONLY_BENCH_SLUGS. The harness is merged (#3295) but not deployed, so Prometheus holds no privacy_round_trip_bps series. A metric_panel with no value deletes its row rather than marking it, so shipping this today would publish an empty table under a confident title.

Staging still renders held-back benches — that is the point of the set. The copy, the JSON-LD and the selector can be reviewed there before any reader sees a board. The un-gate condition is written beside the entry.

Filed under Bridges, not a new Privacy category

These rails are cross-chain swap rails, so the category is accurate, and it avoids opening a tenth category while two of the existing nine already show nothing (Explorers has one bench that 410s, On-ramps one that is held back). A hub collects benches by slug list, not by category, so a future privacy hub needs nothing here.

Four things the validator caught, each a real constraint I had guessed

• source is a single repo link, not a list of named endpoints.
• A dimension value cannot contain a dot, so the rungs are sat100k, sat1m, sat10m. The size is kept in the value so retuning the ladder cannot silently keep an old series under a name like "small" — and the harness labels were renamed to match, because a spec filtering on a value nothing publishes renders an empty board.
• An aggregation needs an empty selector, avg_over_time(metric{}[24h]), or the loader appends the provider filter after the closing paren and Prometheus refuses to parse it.
• A new slug must be in bench-published.json or every datePublished, the citation date and the Dataset temporalCoverage silently fall back to 2025-01-01.

And one the SEO rules caught

seo_title was 72 characters against a 35 to 65 window. Now 62. The rest pass: no em dashes, no marketing adjectives, no hand-typed numbers, no pre-written winner, and the leader comes from {{best_name}} so the copy cannot contradict the table.

Cache keys

The provider set grew by two: bench-unfiltered v102, all-benchmarks v95, bench-filters v43 (the three rungs are filtered variants), provider-set hash 1838 → 1840.

Three more specs follow the same shape once this is reviewed: shielded payout coverage, exit cost, and the Monero maker book.

Gates: pnpm validate 254 specs, tsc --noEmit clean, bun test src/ 731 pass, and the harness's gofmt/build/vet/test all clean.