0.6.0 Capture Release Gates¶
This page records the final source-tree release gates for the 0.6.0 Capture
Intelligence milestone. It is evidence for release preparation, not the public
release record.
Publication and post-publication smokes happened after version preparation and npm publishing. See Release Checklist and Post-Release Validation.
Scope¶
The milestone proves that capture can classify a proposed memory before any write and then apply an explicit user decision through the confirmed capture boundary.
The release candidate must prove:
memory.suggest_captureis read-only;- bounded relationship evidence is scope-safe and output-bounded;
- exact duplicates, blocked candidates, rejected drafts, and unresolved ambiguous candidates remain write-free;
- confirmed create, update, keep-separate, clarify, and reject decisions go
through
memory.confirm_capture; - confirmed updates pass the displayed memory ID and revision;
- stale revisions return
MEMORY_REVISION_CONFLICTwithout silent retry; - Codex and Claude Code generated artifacts resolve the version-pinned Nuzo runtime and keep host wrappers thin.
Source-Tree Gate Evidence¶
These commands were run before closing the milestone implementation gate:
npm run check
npm test
npm run release:check
npm run build
npm run benchmark:recall
npm run benchmark:capture
npm run benchmark:capture -- --expect bounded
npm run validate:npm
npm run smoke:cli
npm run smoke:host-hooks
npm run smoke:claude-code-plugin
npm run smoke:codex-plugin
npm run package:plugins
.venv-docs/bin/mkdocs build --strict --site-dir /tmp/nuzo-release-gates-site
git diff --check
The build-mutating gates should be run sequentially because dist/ and
build/ are recreated by several commands.
Benchmark Evidence¶
Recall Benchmark¶
npm run benchmark:recall passed with:
top1=100.0%;expected_recall=100.0%;noise=0.0%;- English group
top1=100.0%andexpected_recall=100.0%.
Capture Baseline Profile¶
npm run benchmark:capture passed as the reproducible 0.5.0 baseline profile:
profile_accuracy=100.0%;contract_coverage=0.0%;exact_duplicate_recall=100.0%;policy_accuracy=100.0%;safety=100.0%;zero_writes=100.0%;memory_writes=0;audit_writes=0;scope_violations=0;archived_violations=0;bound_violations=0.
This baseline intentionally does not claim bounded relationship evidence.
Capture Bounded Profile¶
npm run benchmark:capture -- --expect bounded passed as the 0.6.0 release
quality profile:
target_relationship_accuracy=100.0%;profile_accuracy=100.0%;contract_coverage=100.0%;exact_duplicate_recall=100.0%;primary_accuracy=100.0%;policy_accuracy=100.0%;safety=100.0%;zero_writes=100.0%;evidence_reasons=100.0%;memory_writes=0;audit_writes=0;scope_violations=0;archived_violations=0;bound_violations=0.
English quality is gated independently and passed:
- English
target=100.0%; - English
profile=100.0%; - English
contract=100.0%; - English
safety=100.0%.
Staged Package And Host Evidence¶
npm run validate:npm passed against staged npm tarballs. It validates:
- installed CLI session continuity;
- installed MCP session continuity;
- read-only capture suggestions;
- explicit
memory.confirm_capturecreate, update, reject, duplicate-skip, and stale revision conflict behavior; - exact public MCP tool set;
memory.doctorwithout exposing stored memory content.
npm run smoke:host-hooks passed the lifecycle-hook matrix with 75 memories and
53 scenarios. It proves bounded read-only recall and no hook writes.
npm run smoke:claude-code-plugin and npm run smoke:codex-plugin passed
against generated host artifacts and staged npm tarballs. They prove the same
separate-session MCP continuity flow through the release-layout host packages.
Post-Publication Gates¶
After version preparation and npm publication, these commands passed:
npm run smoke:published:cli
npm run smoke:published:mcp
NUZO_PLUGIN_SMOKE_PUBLISHED=1 npm run smoke:claude-code-plugin
NUZO_PLUGIN_SMOKE_PUBLISHED=1 npm run smoke:codex-plugin
Those commands resolve public npm packages and should be treated as post-publication evidence, not as source-tree milestone closure.