# v0.5.0 Release Notes

## Version

`v0.5.0`

## Dist Tag

`latest`

This is the stable release record for the gate-first `0.5.0` line.

Post-publish registry state:

- `persona-harness@latest`: `0.5.0`;
- `persona-harness@next`: `0.5.0-rc.2`;
- `persona-harness@alpha`: `0.3.9-alpha.8`;
- `persona-harness@0.5.0` gitHead:
  `c0f1085a5182cdd17411bd043173aabc9a76b30e`;
- dist.shasum: `3a7c43e4807e7cc8bd1b6c697746d6334ee56b09`;
- local/remote `v0.5.0` tag:
  `c0f1085a5182cdd17411bd043173aabc9a76b30e`;
- Trusted Publisher run `28611027369`: SUCCESS;
- Release workflow run `28611144533`: SUCCESS;
- GitHub release:
  `https://github.com/jyt6640/persona-harness/releases/tag/v0.5.0`.

## Summary

`0.5.0` makes Persona Harness gate-first by default.

The product center is the local CLI completion gate:

- workflow/check/finish/archive gates;
- PH-generated evidence and explicit verification records;
- deterministic conformance blockers;
- TDD evidence rail and read-only TDD status helper;
- read-only evidence reports, including A/B and P-minus decision support.

OpenCode runtime guidance remains available as explicit opt-in preview behavior,
not the default product path.

## Default Runtime Behavior

Default backend bootstrap is gate-first:

```sh
npx ph bootstrap backend
```

The default setup keeps model-facing runtime guidance off:

- `features.runtimeInjection=false`;
- `enforce.systemConstitution=false`;
- `enforce.executeVerification=false`.

To opt into model-facing runtime guidance:

```sh
npx ph bootstrap backend --runtime-injection-preview
```

Strict mode remains an intentional heavier opt-in:

```sh
npx ph bootstrap backend --strict
```

Strict mode opts into runtime guidance, system constitution, and direct
verification. It is still not generated-app product-quality certification or a
closure guarantee.

## Evidence And P-minus Interpretation

`0.5.0` includes the accepted P-minus aggregate-vs-paired interpretation
cleanup.

`ph evidence pminus-report --json` reports paired consistency metadata for key
metrics, including candidate-lower/candidate-higher/tied counts and sign-test
p-values. Aggregate-lower provider-token evidence with weak paired consistency
is reported as descriptive local evidence only and lowers the decision hint to
`keep-gathering`.

For the accepted clean-exit calibrated fixture, aggregate provider-token mean
was lower for PH ON, but paired evidence was inconsistent: ON was lower in
3/10 pairs with sign-test `p=0.34375`. That is not token-saving or
product-efficacy evidence.

## Optional Developer Integrations

Optional integrations remain explicit:

- `ph bootstrap backend --codegraph-preview` registers the external CodeGraph
  wrapper integration when requested. It is not PH-owned CodeGraph, not a
  codegraph replacement, and not default behavior.
- `ph bootstrap backend --lsp-preview` registers the LSP MCP wrapper preview.
  Missing dependencies keep a protocol-alive unavailable facade. No real Java
  LSP effectiveness or default LSP claim is made.
- `ph bootstrap backend --runtime-injection-preview` enables optional
  model-facing PH guidance.

## Accepted Evidence Summary

The stable direction is based on accepted local and registry evidence from the
`0.5.0-rc.2` cycle:

- registry `0.5.0-rc.2` package/runtime smoke verified gate-first default
  bootstrap and explicit runtime-injection preview behavior;
- timeout-bounded 10-pair PH OFF/ON A/B produced negative/partial scoped
  evidence;
- clean-exit calibrated 10-pair fixture passed evidence integrity, but paired
  statistics were inconsistent and did not support token-saving or product
  efficacy claims.

These results justify the default-off runtime guidance policy and the
gate-first release message. They do not prove a universal effect.

## External Stable Registry Smoke

QA accepted the official `0.5.0` External registry smoke narrowly as registry
package-runtime evidence.

- source: registry `persona-harness@latest` only;
- archive:
  `/Users/yongtae/Desktop/persona-harness-artifacts/archive/2026-06-24-desktop-persona-runs/v050-stable-registry-smoke-20260702T180404Z`;
- `RESULT.md`: PASS;
- `summary-final.json`: parsed and confirmed command exits/surfaces.

Accepted package-runtime observations:

- package entries included localized READMEs, the logo,
  `docs/current/release/v0.5.0-release-notes.md`, P-minus paired/status dist
  files, instruction infer/adopt/check dist files, and CodeGraph/LSP wrappers;
- basic CLI/help/version/init/doctor/bootstrap surfaces passed;
- default bootstrap was gate-first/default-off:
  `features.runtimeInjection=false`, `enforce.systemConstitution=false`,
  `enforce.executeVerification=false`;
- default developer MCPs were only `grep_app`/`context7`; CodeGraph and LSP
  were not default; no `.codegraph` auto-init occurred;
- `--runtime-injection-preview` enabled runtime injection only;
- `--strict` intentionally enabled runtime injection, system constitution, and
  execute verification;
- CodeGraph/LSP opt-in facades passed with no token-savings, default, or
  effectiveness claim;
- P-minus paired consistency passed:
  aggregate-lower/paired-inconsistent evidence lowered to `keep-gathering` and
  `aggregate-lower-but-paired-inconsistent`;
- report/status/metrics surfaces stayed read-only;
- instruction preview passed, including Controller→Repository drift detection
  after adopted rule, with no harness config or closure mutation;
- workflow spot checks passed: `workflow tdd` exited 0 and incomplete strict
  `workflow finish implement` exited 1.

This is registry package-runtime smoke only. It is not product-efficacy,
app-quality, token-saving, full-TDD, CodeGraph/LSP effectiveness, broad
reliability, closure guarantee, Codex/code-nav replacement, generated-app
certification, or automatic downgrade/removal evidence.

## Claim Audit

Do not claim:

- token saving or provider-token saving;
- product efficacy or navigation benefit;
- app/product quality guarantee;
- full TDD framework, scaffolding, coverage, mutation testing, or test
  sufficiency;
- CodeGraph or LSP default behavior or effectiveness;
- broad AST/linter product or general architecture correctness;
- broad reliability, closure guarantee, or generated-app certification;
- Codex support or code-nav replacement;
- automatic downgrade/removal beyond read-only P-minus decision support.

Allowed wording:

- gate-first default;
- runtime guidance default-off and opt-in preview;
- P-minus reports aggregate-vs-paired consistency and lowers weak aggregate-only
  evidence to `keep-gathering`;
- finish/evidence gates block explicitly defined unverified completion paths.