Roadmap

go-ruby-abbrev/abbrev is grown test-first, each capability differential-tested against MRI rather than built in isolation. Ruby's abbrev — the deterministic, interpreter-independent prefix-abbreviation algorithm — is complete.

Stage	What	Status
Unambiguous abbreviations	Every prefix that identifies exactly one word plus each full word; a prefix shared by two or more words never appears.	Done
Optional prefix filter	Abbrev(words, prefix) keeps only words starting with prefix, treated as a literal string (MRI anchors it as /\A<quoted>/), not a pattern.	Done
Both Ruby entry points	The Go shape of Abbrev.abbrev(words[, prefix]) and the Array#abbrev core extension, collapsed into one variadic Abbrev.	Done
MRI edge cases	Empty words map to themselves with no prefixes; duplicate words collapse; multibyte words split on characters, not bytes, matching String#[].	Done
Differential oracle & coverage	Runtime-free cases drive coverage to 100%; a differential oracle runs the same inputs through ruby (Abbrev.abbrev) and asserts byte-identical maps. gofmt + go vet clean, green across six arches and three OSes.	Done

Documented out-of-scope boundaries

These are deliberate, recorded so the module's surface is unambiguous:

• No Regexp prefix. MRI's optional pattern may be a Regexp; this library models the common literal String-prefix case (what rbgo binds). A host that needs the regexp form filters the word list before calling Abbrev.
• No interpreter. The library implements the deterministic algorithm; it never runs arbitrary Ruby. Wiring it into a live object model is the consumer's job — that is why rbgo binds this module rather than the reverse.
• Reference is reference Ruby (MRI). Byte-for-byte conformance targets MRI's Abbrev.abbrev, pinned by the differential oracle.
• Standalone & reusable. The module has no dependency on the Ruby runtime; the dependency runs the other way.

See Usage & API for the surface and Why pure Go for the deterministic/interpreter split.