Find your next thread

Evidence & studies

Inspect the result.
Know its limits.

Start with the release references, then explore earlier studies. Each result belongs to its recorded version, scope and method; historical measurements are not current-release performance guarantees.

A curious cat following a bright blue thread of yarn
Fix release · v0.11.1

Faster saved reads, visible phases.

Snapshot reads share object-store admission while preserving payload integrity checks. Documentation commands report phases and heartbeats on stderr. Proposal publication preserves short sequences instead of inventing disconnected overview nodes. Full CI and all three platform builds passed; an isolated macOS arm64 installation reports READY. The reported workplace stall still requires phase-level diagnosis; this release does not establish complete service documentation.

Downloads and limitations ↗ · Full CI ↗ · Release checks ↗

Release reference · v0.11.0

Internal flows in the generated documentation.

The release adds native source-linked flow diagrams, dependency maps and linked decision tables, recovery from saved service captures, shared publication influence and a fix for false conflicts from generated sibling placeholders.

Release qualification passed. The release commit passed full CI; packaged builds passed for macOS arm64/x86_64 and Linux x86_64. An isolated macOS arm64 installation passed runtime verification. Earlier study results below remain historical. This scope does not establish exhaustive business-flow coverage, independently verified meaning, deployed behavior or 40-service capacity.

v0.11.0 downloads and limitations ↗ · Release checks ↗ · Full CI ↗ · Native visual publication tests ↗ · Visual authoring workflow →

Historical release reference · v0.8.1

Local documentation you can inspect.

The release covers source documentation without an application build, local unassessed publication, service sections, processes, entity dataflows, portable evidence and publication history. Release qualification passed for macOS arm64/x86_64 and Linux x86_64. Three local Kotlin service reviews exercised the documentation workflow. Version 0.8.1 adds bundled offline help and runbooks, shared reader navigation and visible compiler-provider evidence.

These checks do not establish exhaustive callable coverage, independently reviewed business meaning, deployed-system behavior, GitLab readiness or 40-service capacity.

Release scope and downloads ↗ · Release checks ↗ · Executable fixtures ↗

Historical saved-edit qualification

Retained reports and measurements.

Earlier source-build runs checked Kotlin consequences, Rust syntax, repeat rendering and freshness after saved edits. Their timings include development conditions and should not be used to predict release latency.

Fixture measurements → · Repository measurements → · Interactive report →

Historical v0.3.1 service pilot

31 Spring entry points across two services.

Independent inventories matched the selected annotation-declared roots, including one scheduled job missing from existing documentation. The inputs are private; published aggregates do not form a reproducible public dataset.

Method, result and limits →
Historical Kotlin study · revision 6281138

16/16 focused checks passed

Compatibility policy, Unicode coordinate normalization, Maven reactor authority, bounded local CFGs and semantic-input hashing were each traced to exact source. All sixteen selected checks passed with zero failures and zero skips.

Inspect source fragments and digests →
Historical navigation receipt

3 of 9 candidates, 115 exact source lines

A bounded Kotlin navigation request returned three candidates, an exact source window, K2 FIR declaration identity, a source digest and an evidence digest. The receipt explicitly remained PARTIAL/UNSURE because one requested term was unmatched—useful evidence without silent overclaim.

Explore evidence-bound documentation →
Historical prepared warm Q1 · tokens

3.795× fewer effective tokens

In one prepared warm Q1-SINGLE-ANALYSIS C→D run, both arms were source-correct. Codeclew consumed 10,560 effective tokens; the counterbalanced default arm consumed 40,075.

Machine-readable evidence →
Historical prepared warm Q1 · tool output

40.28× less tool output

Within the same run, Codeclew returned 15,307 bytes of tool output and the default arm returned 616,573 bytes. This measures the evidence envelope, not universal task performance.

Machine-readable evidence →
Technical paper

Results on the site. Method and limitations in the PDF.

The paper records the five-slice selection, exact test gate, deterministic text-search baseline, preserved receipt authority, resource-limit exclusion, claim boundary and the smallest useful next comparison.

Download the Kotlin evidence study →