DaoyuanLi2816 · DaoyuanLi2816 · Aug 12, 2026 · Aug 12, 2026 · Aug 12, 2026 · Aug 12, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -6,6 +6,27 @@ All notable changes to miniVERL are recorded here. The format follows
 
 ## [Unreleased]
 
+## [0.7.1] - 2026-08-11
+
+### Product correction
+
+- Reordered the English, Chinese, PyPI and documentation landing pages around
+  the installable single-GPU runtime, its hardware boundary, the pinned verl
+  artifact bridge and measured systems evidence. Research studies and their
+  negative results remain intact under Research Notes rather than preceding
+  the quickstart.
+- Updated package metadata, CLI help and diagnostics to describe the current
+  single-GPU alignment and distillation runtime without claiming the planned
+  verl-style OPD execution layer already exists.
+- Added `miniverl evidence show/validate alignment-external-v1` and
+  `miniverl pilot --builtin-study alignment-external-v1`. The wheel now carries
+  the typed result, schema, preregistration and all 512 task-evidence rows with
+  byte-bound validation, so the primary pip journey no longer depends on a Git
+  checkout.
+- Carried the corrected post-v0.7.0 evidence digest labels into a new immutable
+  stable release without changing the v0.7.0 tag or any frozen scientific
+  result.
+
 ## [0.7.0] - 2026-08-10
 
 ### External Alignment Gate result
@@ -802,7 +823,7 @@ Same-tokenizer only; one trajectory per forward pass; `swap` unavailable for
 quantized models; only Qwen3 and Qwen2 architectures tested; single-seed GPU
 results. The full list is in `docs/limitations.md`.
 
-[Unreleased]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.7.0...HEAD
+[Unreleased]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.7.1...HEAD
 [0.7.0]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.6.3...v0.7.0
 [0.6.3]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.6.2...v0.6.3
 [0.6.2]: https://github.com/DaoyuanLi2816/mini-verl/compare/v0.6.1...v0.6.2

diff --git a/CITATION.cff b/CITATION.cff
@@ -2,15 +2,15 @@ cff-version: 1.2.0
 title: "miniVERL: Auditable single-GPU alignment and distillation runtime"
 message: "If you use miniVERL in your work, please cite it as below."
 type: software
-version: 0.7.0
+version: 0.7.1
 date-released: 2026-08-11
 license: Apache-2.0
 repository-code: "https://github.com/DaoyuanLi2816/mini-verl"
 url: "https://github.com/DaoyuanLi2816/mini-verl"
 abstract: >-
-  miniVERL is a single-GPU runtime for comparing alignment and distillation
-  methods under conditions strict enough that the comparison means something.
-  Its core is multi-turn, tool-aware on-policy distillation: a student language
+  miniVERL is an auditable single-GPU alignment and distillation runtime with a
+  bounded artifact bridge to one pinned verl profile. Its native core is
+  multi-turn, tool-aware on-policy distillation: a student language
   model samples its own tool-using trajectories against deterministic local
   environments, a teacher scores exactly the states the student visited, and
   the student is updated with token-level distributional supervision on its own
@@ -20,8 +20,8 @@ abstract: >-
   never become a training label, and stores teacher targets in a versioned,
   checksummed, pickle-free cache with policy-version enforcement.
   Around that core it provides supervised fine-tuning, recorded-provenance
-  preference optimization and knowledge distillation as directly comparable
-  arms; shared-backbone execution that switches student, teacher and reference
+  preference optimization and knowledge distillation; shared-backbone
+  execution that switches student, teacher and reference
   roles across adapters on one set of base weights to fit low-memory hardware;
   padded update batching; deterministic replay; transactional checkpoints and
   cross-process run locks; and mechanism studies such as RecoveryBench that are
@@ -32,7 +32,7 @@ abstract: >-
   algorithmic parity with PPO. miniVERL is designed for one personal CUDA GPU,
   automatically selects bf16 or fp16, and requires neither Ray nor a cluster.
   Published performance is measured on one RTX 4080; other GPU models use the
-  same code path but remain unmeasured. The v0.7.0 external-alignment study
+  same code path but remain unmeasured. The v0.7 external-alignment study
   terminated at its preregistered checkpoint-selection gate: two declared
   lineages and eight candidates all scored 0/64 retained JSONNav utility, so
   no teacher qualification, continuation method comparison or reserved final

diff --git a/PROJECT_STATE.md b/PROJECT_STATE.md
@@ -6,11 +6,35 @@ and what it printed.
 
 Last updated: 2026-08-11.
 
-Canonical release state: stable `v0.7.0` (`148822964dbb73e97ce06ef740f907364166a724`), development `0.7.1.dev0`.
+Canonical release state: releasing `v0.7.1`.
 Every public version claim is generated from `release-state.yaml` and gated by
 `python scripts/release_state.py --check`.
 
-## v0.7.0 External Alignment Gate — EVIDENCE RELEASE IN PROGRESS
+## v0.7.1 Product correction — RELEASE CANDIDATE
+
+Branch `v0.7.1-product-correction` starts from synchronized main
+`3ec042e829b4193aab363ab97e3786a1cd2621c0`. It changes product hierarchy,
+installed evidence UX and metadata only; no scientific experiment, frozen
+result, model revision or existing tag changes.
+
+The wheel carries the typed v0.7 result, schema, preregistration and 512
+privacy-safe JSONNav rows. Each file is checked against its recorded SHA-256,
+the result is parsed through the impossible-state validator, task rows are
+structurally checked, and `pilot --builtin-study` exposes the early-stop result
+without a checkout. The planned verl-shaped executable OPD profile remains a
+v0.8 development objective and is not claimed by this release.
+
+Local release-candidate validation at implementation commit `5142ed6` passes
+2110 non-GPU/non-network tests with 86.19% branch coverage, 8 RTX 4080 GPU
+tests, 15 network tests, Ruff over 332 files, mypy over 147 source files,
+actionlint 1.7.12, strict MkDocs and 36 rendered SVG instances at four
+viewports. Twenty-four screenshots are under the ignored
+`artifacts/docs-visual-v0.7.1/` directory and the 1440, 820 and 390 px home
+views were manually inspected. Clean core and `[train]` wheel installs pass;
+core keeps torch absent and validates the built-in evidence, while `[train]`
+completes the real fast demo under latest resolved dependencies.
+
+## v0.7.0 External Alignment Gate — RELEASED
 
 Branch `v0.7.0-evidence-release` starts from the exact post-PR-#52 main commit
 `a8272e2b5674e12107461a81d285f1a3d56588a5`. PR #51 merged the public