Summary
RankQuantFastscan is #[doc(hidden)] but publicly reachable as ordvec::RankQuantFastscan. It's an optional b=2 FastScan kernel (block-32 PQ-LUT) offering a real scan-latency win at ~2× the RankQuant b=2 footprint (dim/2 bytes/doc) — compelling for latency-critical callers — but the "hidden + not yet stable" status makes it un-adoptable in production code.
Ask
Give callers a way to opt in knowingly. Preferred path (pre-1.0): gate it behind --features fastscan with a documented stability caveat, rather than stabilising it onto the default surface. Keep it off the default docs build.
Open question for the maintainer: feature-gate vs. promote to a documented ordvec::fastscan::RankQuantFastscan with an explicit "experimental API, may change before 1.0" banner.
Notes
Filed from an external technical review (May 2026); verified against v0.2.0.
Summary
RankQuantFastscanis#[doc(hidden)]but publicly reachable asordvec::RankQuantFastscan. It's an optional b=2 FastScan kernel (block-32 PQ-LUT) offering a real scan-latency win at ~2× the RankQuant b=2 footprint (dim/2bytes/doc) — compelling for latency-critical callers — but the "hidden + not yet stable" status makes it un-adoptable in production code.Ask
Give callers a way to opt in knowingly. Preferred path (pre-1.0): gate it behind
--features fastscanwith a documented stability caveat, rather than stabilising it onto the default surface. Keep it off the default docs build.Open question for the maintainer: feature-gate vs. promote to a documented
ordvec::fastscan::RankQuantFastscanwith an explicit "experimental API, may change before 1.0" banner.Notes
#[doc(hidden)]path.RankQuant::new(dim, 2)(PR fix: pre-public public-API contract hardening (release-readiness triage) #26), so the surface is small.Filed from an external technical review (May 2026); verified against v0.2.0.