v0.3.3 — the eval-driven resolver

Resolver: nine eval-driven rounds, four languages, five SCIP-verified corpora

Corpus	Language	Gated precision	Gated recall
gin v1.10.0	Go	99.3%	82.6%
hono v4.6.14	TypeScript	98.8%	63.8%
click 8.1.8	Python	98.7%	94.6%
ripgrep 14.1.1	Rust	98.5%	75.3%
abyss (dogfood)	Rust	100.0%	90.9%

Gated = --min-confidence 0.7 (default). Full method, per-tier tables, and the round-by-round chronicle: eval/RESULTS.md. Reproduce: eval/run.sh.

Highlights since v0.3.0

• The 1.0 tier earns its name — full confidence only for call shapes measured ≥98% correct; qualified calls with unknown receivers never pose as facts (hono's 1.0 tier was 81.9% precision; now 97.9–100% everywhere).
• Named-import binding tiers — import { x } from './m' (TS, barrel chains chased), from m import x (Python), import com.f.X (Java), use crate::m::x incl. pub use re-exports (Rust).
• Receiver-type lite inference for Rust — typed params, let x = T::new(), self → impl type, associated functions (T::new()), workspace crate roots via Cargo.toml discovery.
• Type-grade evidence beyond exact scope — the receiver type's import binding or unique defining file resolves aliases, split impls, and trait-scoped methods.
• Chunker fixes that benefited every language — oversized functions keep their symbols; non-exported const f = () => are symbols; impl Trait for Type scopes methods to the type.

Install

curl -fsSL https://raw.githubusercontent.com/telagod/abyss/main/install.sh | bash
npm install -g @code-abyss/cli   # 0.3.3
cargo binstall code-abyss        # or: cargo install code-abyss

74 tests. APIs and index format may still change before 1.0.

v0.3.0

release: v0.3.0

- version 0.3.0 (drop -dev)
- add Apache-2.0 LICENSE file
- cargo publish --dry-run verified