# v0.3.9-alpha.0 Release Notes

## Version

`v0.3.9-alpha.0`

## Dist Tag

`alpha`

During the alpha pilot, `latest` may be synchronized to the current alpha build to avoid stale default installs. This is not a stable support claim.

## Summary

This prerelease tightens the pre-eval workflow rail and backend-shape surfaces before the project moves toward `0.4` eval work.

Persona Harness remains an AI coding workflow rail + evidence + continuation harness. It is not a Java Clean Code quality guarantee, generated app product-quality certification, AST/linter, or enforcement engine.

## What Changed

### Backend-shape Java field parser hardening

- Reduced naive `private` line text scan pollution in backend-shape checks.
- Reduced false WARNs caused by comments, text blocks, strings, lambdas, and nested generic mentions.
- Preserved WARNs for real multiline Service fields such as `Map`, `AtomicLong`, and `nextId`.
- Added adversarial fixture coverage so these parser boundaries stay fixed.

### Spring Boot Gradle dependency-management guidance

- Strengthened guidance for Boot plugin-managed dependencies and `io.spring.dependency-management`.
- Clarified that `spring-boot-starter-*`, JDBC/JPA/validation starters, and Flyway dependencies should not use blank versions or dot-only versions.
- Added recovery guidance: when dependency resolution fails, fix `build.gradle`, then rerun wrapper-backed `test` and `build`.

### Report coverage continuation hardening

- Added report coverage WARN behavior when a report says `Status: filled` but coverage/checklist body content is blank or template-like.
- Surfaced report coverage missing state and next actions through `check`, `continue`, and `finish`.
- This is a workflow completion guard, not generated app quality certification.

### v0.4 eval preparation docs

- Added or refined v0.4 eval preparation records for:
  - three README fixture candidates;
  - plain prompt / AGENTS / CLAUDE / cursorrules / PH baseline conditions;
  - baseline kill-gate;
  - preregistered thresholds;
  - primary and secondary metrics;
  - blind / second reviewer rubric.
- Added and restored the v0.4 evaluation runbook, evaluation methodology, fixture files, and evaluation plan in the release target HEAD.
- The runbook records fixture order, baseline setup, metadata capture, artifact archive naming, metrics, blind package preparation, second reviewer handoff, and kill-gate calculation timing.
- Release references to the runbook are guarded: release prep may mention the runbook only when it exists in the target HEAD being prepared.

### AST verified report schema research

- Refined the verified report schema research direction with candidate fields:
  - `ruleId`;
  - `result`;
  - `targetFile`;
  - `evidence`;
  - `limitations`;
  - `confidence`;
  - `source`.
- Recorded stable rule ID candidates.
- Added the verified report manual backfill plan:
  - artifact selection;
  - rule ID evidence source mapping;
  - PASS / WARN / FAIL / UNKNOWN judgment criteria;
  - confidence and source handling;
  - false-positive review;
  - v0.5 parser spike decision criteria.
- AST/linter/enforcement implementation remains a non-goal for this release.

### 0.4 eval stop-gate

- Documented that HQ must stop before 0.4 eval and ask the user for a `0.3.9-alpha` publish or release decision.
- Completing pre-eval cleanup, QA eval fixture/kill-gate docs, and verified report schema docs does not authorize publish, push, tag, or direct 0.4 eval start.

### Docs stale guard

- Added release checklist guardrails for package version, lockfile, CHANGELOG, release note, and develop README/current-status alignment.
- Added release-prep checks for surface-verified smoke versus full generated-app behavior verification.
- Added release-prep checks for generated app product quality certification wording.
- Added registry `gitHead` / current `HEAD` mismatch recording and smoke install-source recording.
- Added develop-doc commit/no-commit reporting guidance, including the case where the develop directory is not a git worktree.

### Skills guidance brevity audit

- Tightened Gradle guidance wording while preserving semantics:
  - fake Gradle shim remains forbidden;
  - wrapper-first verification remains preferred;
  - Spring Boot dependency-management guidance remains in force;
  - dependency-resolution failures still require fixing `build.gradle` and rerunning wrapper-backed verification.

### Runtime structured warning output

- Runtime warnings now use the structured shape `[Persona Harness Runtime Warning] kind=... scope=...`.
- Host hook throws remain guarded so warning output does not reject the host hook.
- Evidence writes remain best-effort; this is host stability work, not generated app quality certification.

### CLI report coverage helper refactor

- Split workflow report coverage finding logic into `workflow-report-coverage.ts`.
- Workflow semantics are unchanged.
- Added structured summary assertions for report coverage findings.

### External develop artifacts

- Lane retrospectives and templates were prepared outside this repository under the develop operating context.
- They are not committed here and are not package contents.

## External Smoke

Pre-alpha9 5-run smoke result: partially successful.

Observed:

- Parser hardening local/Windows adversarial fixture succeeded.
- Clean tarball install, `init`, `bootstrap`, and `doctor` succeeded.
- PH ON bounded OpenCode run confirmed workflow rail/profile/evidence/pending block behavior.
- PH ON stopped before implementation files were generated.
- PH OFF baseline used the same README but drifted to CommonJS/in-memory Node HTTP server.

Focused re-smoke result: successful for targeted surfaces.

Observed:

- Current local tarball surfaced report coverage guidance through `check`, `continue`, and `finish`.
- Spring Boot dependency-management guidance surfaced through README/build.gradle injection.
- `init`, `bootstrap`, and `doctor` passed.

Important limits:

- Focused smoke is surface verified.
- This smoke did not directly verify a full OpenCode continuation applying dependency guidance, filling reports by itself, and reaching `finish` PASS.
- This release does not certify generated app product quality.
- AST/linter/enforcement is still absent.
- Registry `alpha`/`latest` before this release prep pointed to `0.3.8-alpha.5` with gitHead `d99df54`, so current commits were verified with local tarballs.

## Verification

Repository verification from the release prep worktree:

```bash
npm test
npm run typecheck
npm run build
npm run check:docs
npm pack --dry-run
```

Observed result on 2026-06-25:

- `npm test`: passed. Scope diagnostics and docs taxonomy diagnostics passed; 57 files / 361 tests passed.
- `npm run typecheck`: passed.
- `npm run build`: passed.
- `npm run check:docs`: passed.
- `npm pack --dry-run`: passed for `persona-harness@0.3.9-alpha.0`.
  - Filename: `persona-harness-0.3.9-alpha.0.tgz`
  - Package contents: 340 files
  - Shasum: `7f7328886b8ab6fbb89bea907a08fcc2d1232440`
- `git diff --check`: passed.

## Known Gaps

- Full generated-app/OpenCode finish behavior is not certified by this release.
- Focused smoke verified guidance and command surfaces, not complete generated application behavior.
- Generated app product quality is not certified.
- Evidence means read/injection/workflow traces, not a quality score.
- AST/linter/build-failure enforcement remains a non-goal.

## Recommended Before Publish

- Publish only as a prerelease alpha after explicit user/HQ release decision.
- Keep messaging focused on workflow rail, parser hardening, report coverage, dependency guidance, and pre-eval preparation.
- Do not describe focused smoke as full OpenCode implementation success or generated app product-quality certification.