Release AKTA v0.6.0

.github/CROSS_REPO_CI.md

@@ -0,0 +1,72 @@
+# Cross-repo CI configuration (v0.6)
+
+AKTA default CI runs entirely in-repo. Optional jobs validate live exports against sibling repositories when repository variables or secrets are configured.
+
+## Repository variables
+
+Set these under **Settings → Secrets and variables → Actions → Variables**:
+
+| Variable | Purpose | Activates job |
+|----------|---------|---------------|
+| `PF_CORE_REPO_PATH` | Absolute path to a checked-out PF-Core repo on the runner | `cross-repo-pf-pcs` (PF step) |
+| `PCS_CORE_REPO_PATH` | Absolute path to a checked-out PCS-Core repo | `cross-repo-pf-pcs` (PCS step) |
+| `SCOPE_REPO_PATH` | Absolute path to a checked-out SCOPE repo | `cross-repo-scope` |
+| `PCS_BENCH_REPO_PATH` | Absolute path to a checked-out PCS-Bench repo | `cross-repo-pcs-bench` |
+
+## Repository secrets
+
+| Secret | Purpose |
+|--------|---------|
+| `SCOPE_REPO_URL` | Git clone URL for SCOPE; CI clones to `$GITHUB_WORKSPACE/scope-sibling` and sets `SCOPE_REPO_PATH` |
+
+## Manual sibling checkout pattern
+
+Self-hosted or composite runners with sibling repos on disk:
+
+```yaml
+env:
+  PF_CORE_REPO_PATH: /opt/siblings/PF-Core
+  PCS_CORE_REPO_PATH: /opt/siblings/PCS-Core
+  SCOPE_REPO_PATH: /opt/siblings/SCOPE
+  PCS_BENCH_REPO_PATH: /opt/siblings/PCS-Bench
+```
+
+Run locally:
+
+```powershell
+$env:PF_CORE_REPO_PATH = "C:\path\to\PF-Core"
+$env:PCS_CORE_REPO_PATH = "C:\path\to\PCS-Core"
+$env:SCOPE_REPO_PATH = "C:\path\to\SCOPE"
+$env:PCS_BENCH_REPO_PATH = "C:\path\to\PCS-Bench"
+pytest tests/contracts/ -v -m integration
+```
+
+## Job behavior
+
+### `cross-repo-pf-pcs`
+
+Runs when `PF_CORE_REPO_PATH` or `PCS_CORE_REPO_PATH` is non-empty. Exports AKTA artifacts and validates against sibling PF-Core / PCS-Core validators via `tests/contracts/cross_repo_helpers.py`.
+
+### `cross-repo-scope`
+
+Runs when `SCOPE_REPO_PATH` is set or `SCOPE_REPO_URL` secret is present. Clones SCOPE when only the secret is configured, then runs SCOPE python-import contract tests.
+
+### `cross-repo-pcs-bench`
+
+Runs when `PCS_BENCH_REPO_PATH` is set. Invokes external PCS-Bench runner when available; contract tests verify in-repo fallback when the path is unset.
+
+## External PCS-Bench integration
+
+When `PCS_BENCH_REPO_PATH` points to a sibling checkout, `adapters/pcs_bench/runner.run_pcs_bench_suite()` attempts, in order:
+
+1. Python import: `pcs_bench.runners.akta.run_akta_suite`
+2. CLI script: `scripts/run_akta_suite.py`
+3. In-repo `AKTABenchScenario` runner (default)
+
+## Fail-closed defaults
+
+- Jobs are skipped when variables are unset; default CI remains green without siblings.
+- Invalid `PCS_BENCH_REPO_PATH` values fall back to the in-repo runner rather than silently passing external validation.
+- SCOPE python-import mode does not fall back to simulated grants when `SCOPE_REPO_PATH` is set.
+
+See also [tests/contracts/README.md](../tests/contracts/README.md).

.github/workflows/ci.yml

@@ -22,7 +22,7 @@ jobs:
           python-version: ${{ matrix.python-version }}
 
       - name: Install dependencies
-        run: pip install -e ".[dev]"
+        run: pip install -e ".[dev,security]"
 
       - name: Run pytest
         run: pytest tests/ skills/akta-scientific-action-admissibility/tests/ -v
@@ -50,6 +50,12 @@ jobs:
       - name: Run oracle-independent eval
         run: python evals/run_oracle_independent.py --out evals/reports/oracle_independent.json
 
+      - name: Run holdout private eval
+        run: python evals/run_holdout_eval.py --out evals/reports/holdout_private.json
+
+      - name: Run adversarial transition eval (v0.6)
+        run: python evals/adversarial_transitions.py --out evals/reports/adversarial_transitions.json
+
       - name: Invalid fixtures must fail
         run: pytest tests/test_invalid_cases.py -v
 
@@ -59,9 +65,70 @@ jobs:
       - name: Run integrated protocol-drift demo
         run: python scripts/demo_akta_scope_protocol_drift.py
 
+      - name: Run reconstructable experiment demo
+        run: python scripts/demo_reconstructable_experiment.py
+
       - name: Upload eval reports
         if: always()
         uses: actions/upload-artifact@v4
         with:
           name: eval-reports-py${{ matrix.python-version }}
           path: evals/reports/
+
+  cross-repo-pf-pcs:
+    runs-on: ubuntu-latest
+    if: ${{ vars.PF_CORE_REPO_PATH != '' || vars.PCS_CORE_REPO_PATH != '' }}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install dependencies
+        run: pip install -e ".[dev,security]"
+      - name: Cross-repo PF-Core validation
+        if: ${{ vars.PF_CORE_REPO_PATH != '' }}
+        env:
+          PF_CORE_REPO_PATH: ${{ vars.PF_CORE_REPO_PATH }}
+        run: pytest tests/contracts/test_cross_repo_live.py::test_pf_core_live_validation -v -m integration
+      - name: Cross-repo PCS-Core validation
+        if: ${{ vars.PCS_CORE_REPO_PATH != '' }}
+        env:
+          PCS_CORE_REPO_PATH: ${{ vars.PCS_CORE_REPO_PATH }}
+        run: pytest tests/contracts/test_cross_repo_live.py::test_pcs_core_live_validation -v -m integration
+
+  cross-repo-scope:
+    runs-on: ubuntu-latest
+    if: ${{ vars.SCOPE_REPO_PATH != '' || secrets.SCOPE_REPO_URL != '' }}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install dependencies
+        run: pip install -e ".[dev,security]"
+      - name: Optional SCOPE sibling checkout
+        if: ${{ secrets.SCOPE_REPO_URL != '' }}
+        env:
+          SCOPE_REPO_URL: ${{ secrets.SCOPE_REPO_URL }}
+        run: |
+          git clone --depth 1 "$SCOPE_REPO_URL" scope-sibling
+          echo "SCOPE_REPO_PATH=$GITHUB_WORKSPACE/scope-sibling" >> $GITHUB_ENV
+      - name: SCOPE live contract tests
+        env:
+          SCOPE_REPO_PATH: ${{ env.SCOPE_REPO_PATH || vars.SCOPE_REPO_PATH }}
+        run: pytest tests/contracts/test_scope_python_import_live_shape.py -v -m integration
+
+  cross-repo-pcs-bench:
+    runs-on: ubuntu-latest
+    if: ${{ vars.PCS_BENCH_REPO_PATH != '' }}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Install dependencies
+        run: pip install -e ".[dev,security]"
+      - name: PCS-Bench external integration contract
+        env:
+          PCS_BENCH_REPO_PATH: ${{ vars.PCS_BENCH_REPO_PATH }}
+        run: pytest tests/contracts/test_pcs_bench_external.py -v

CHANGELOG.md

@@ -4,6 +4,45 @@ All notable changes to AKTA are documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
+## [0.6.0] - 2026-06-28
+
+### Added
+
+- Cross-repo CI jobs for optional sibling repos (`PF_CORE_REPO_PATH`, `PCS_CORE_REPO_PATH`, `SCOPE_REPO_PATH`)
+- Closed-loop review: `evaluate_with_grant()`, full `review_decision.py`, human review packet export/import
+- Ed25519 policy signing (`AKTA_REQUIRE_SIGNED_POLICY`, key rotation via `public_keys`); HMAC retained for deployment attestation
+- VSA rich ScientificReport import (claims, evidence_links, validation_results) + PCS bundle `vsa_report.json`
+- Adapters: Scientific Memory, LabTrust-Gym scenario import, PF trace certificate import, PCS-Bench runner
+- Inter-rater eval metadata (`reviewer_ids`, `inter_rater_agreement`, `label_source`) in `expected_decisions.jsonl` with stats in eval reports
+- Adversarial transition eval module (`evals/adversarial_transitions.py`) for grant expiry, scope narrowing, and re-gate tool allow/block
+- External PCS-Bench checkout via `PCS_BENCH_REPO_PATH` with in-repo fallback
+- `.github/CROSS_REPO_CI.md` documenting cross-repo CI variables and secrets
+- `scripts/demo_reconstructable_experiment.py` — full VSA → AKTA → SCOPE → PF → PCS → Memory → PCS-Bench chain
+- AKTA-Bench: 55 oracle-independent scenarios, `holdout_private.jsonl`, behavioral eval runner
+- Tool registry expanded to 50+ lab/compute tools; mandatory tool declaration fail-closed (A8)
+- Hybrid classifier: negation, hedging, prose authority-transfer detection
+- REST optional API key auth (`AKTA_REST_API_KEY`) and rate limiting; LangGraph SCOPE grant scoped retry
+- Expert-reviewed overlay: `materials_expert_v0` with maintainer sign-off metadata
+
+### Changed
+
+- Package version `0.6.0`; policy manifest `akta-policy-manifest-v0.6`
+- PCS export includes VSA report as first-class artifact when present
+
+## [0.5.1] - 2026-06-28
+
+### Fixed
+
+- SCOPE python-import adapter: uses `ScopeEngine.from_policy_dir(policy/)` and v0.5 `create_packet(akta_record=, akta_trigger=)`, `submit_decision(reviewer=, decision=)`, `issue_grant(packet, decision)` — no no-arg constructor or simulated fallback when `SCOPE_REPO_PATH` is set
+- SCOPE CLI adapter: real v0.5 command shapes (`--akta-trigger`, `--akta-record`, `--reviewer`, `--decision` file args)
+- PCS grant validation: accepts real SCOPE v0.5 grants via `authorization.approved_scope` and `source.requested_scope`; rejects overbroad grants; includes real grant unchanged in bundle
+- Contract tests: `test_scope_python_import_live_shape.py`, `test_scope_cli_command_shape.py`, `test_pcs_real_scope_grant_shape.py`, `test_demo_protocol_drift_live_mode.py`
+
+### Changed
+
+- Package version `0.5.1`; `adapters/scope/engine_protocol.py` updated to SCOPE v0.5 API
+- `docs/limitations.md`: v0.5 and v0.4 sections distinguishing simulated/python-import/CLI modes
+
 ## [0.5.0] - 2026-06-28
 
 ### Added

Makefile

@@ -1,9 +1,9 @@
-.PHONY: install test eval-canonical eval-public eval-public-100 eval-oracle demo-weak-evidence demo-akta-weak-evidence demo-akta-scope-protocol-drift ci
+.PHONY: install test eval-canonical eval-public eval-public-100 eval-oracle eval-holdout eval-v06 demo-weak-evidence demo-akta-weak-evidence demo-akta-scope-protocol-drift demo-reconstructable ci
 
 PYTHON ?= python
 
 install:
-	$(PYTHON) -m pip install -e ".[dev]"
+	$(PYTHON) -m pip install -e ".[dev,security]"
 
 test:
 	$(PYTHON) -m pytest tests/ skills/akta-scientific-action-admissibility/tests/ -v
@@ -17,6 +17,15 @@ eval-public-100:
 eval-oracle:
 	$(PYTHON) evals/run_oracle_independent.py --out evals/reports/oracle_independent.json
 
+eval-holdout:
+	$(PYTHON) evals/run_holdout_eval.py --out evals/reports/holdout_private.json
+
+eval-v06:
+	$(PYTHON) evals/adversarial_transitions.py --out evals/reports/adversarial_transitions.json
+
+demo-reconstructable:
+	$(PYTHON) scripts/demo_reconstructable_experiment.py
+
 demo-weak-evidence:
 	$(PYTHON) scripts/demo_weak_evidence.py
 
@@ -26,7 +35,8 @@ demo-akta-weak-evidence:
 demo-akta-scope-protocol-drift:
 	$(PYTHON) scripts/demo_akta_scope_protocol_drift.py
 
-ci: install test eval-canonical eval-public-100 eval-oracle
+ci: install test eval-canonical eval-public-100 eval-oracle eval-holdout eval-v06
 	$(PYTHON) -m pytest tests/test_invalid_cases.py tests/integration/ tests/contracts/ -v
 	$(PYTHON) scripts/demo_integrated_weak_evidence.py
 	$(PYTHON) scripts/demo_akta_scope_protocol_drift.py
+	$(PYTHON) scripts/demo_reconstructable_experiment.py

README.md

@@ -140,6 +140,15 @@ akta-rest --host 127.0.0.1 --port 8765
 # GET  /v0/policy, /v0/health
 ```
 
+## v0.5.1 acceptance status
+
+| Criterion | Status |
+|-----------|--------|
+| SCOPE python-import (`ScopeEngine.from_policy_dir`, v0.5 kwargs) | Pass |
+| SCOPE CLI v0.5 flags (`--akta-trigger`, `--akta-record`, `--reviewer`, `--decision`) | Pass |
+| PCS grant validation (`authorization.approved_scope`, narrow draft grant) | Pass |
+| 235+ tests; `make ci` green | Pass |
+
 ## v0.5 acceptance status
 
 | Criterion | Status |
@@ -150,7 +159,7 @@ akta-rest --host 127.0.0.1 --port 8765
 | LLM classifier trust boundary (registry overrides LLM) | Pass |
 | Overlay governance tiers + production refusal | Pass |
 | Policy file versioning in decision provenance | Pass |
-| 210+ tests; `make ci` green | Pass |
+| 210+ tests; `make ci` green | Pass (see v0.5.1 for SCOPE adapter patch) |
 
 ## v0.4 acceptance status
 
@@ -218,4 +227,24 @@ akta-rest --host 127.0.0.1 --port 8765
 
 ## Status
 
-AKTA v0.5 is a reference implementation. It is not a safety certification. Biology, chemistry, and clinical overlays are experimental and not deployment-ready without institutional governance. Deployment profile P7 (fully autonomous scientific operator) is defined for taxonomy only and is not supported.
+AKTA v0.6.0 is a reference implementation. It is not a safety certification. Biology, chemistry, and clinical overlays are experimental and not deployment-ready without institutional governance. Deployment profile P7 (fully autonomous scientific operator) is defined for taxonomy only and is not supported.
+
+## v0.6 acceptance status
+
+| Criterion | Status |
+|-----------|--------|
+| Cross-repo CI jobs (optional PF/PCS/SCOPE) | Pass |
+| Closed-loop review (`evaluate_with_grant`, expiry F14) | Pass |
+| Ed25519 policy signing + `AKTA_REQUIRE_SIGNED_POLICY` | Pass |
+| VSA rich report + PCS `vsa_report.json` artifact | Pass |
+| Scientific Memory + LabTrust-Gym + PCS-Bench adapters | Pass |
+| Reconstructable experiment demo (full chain) | Pass |
+| Oracle-independent 55 scenarios + holdout eval | Pass |
+| Tool registry 50+ tools + mandatory declaration A8 | Pass |
+| Expert-reviewed overlay (`materials_expert_v0`) | Pass |
+| REST API key auth + rate limiting | Pass |
+| Inter-rater eval metadata + stats in `akta eval` | Pass |
+| Adversarial transition evals (grant expiry, scope narrowing) | Pass |
+| External PCS-Bench checkout integration (`PCS_BENCH_REPO_PATH`) | Pass |
+
+Cross-repo CI variables: see [.github/CROSS_REPO_CI.md](.github/CROSS_REPO_CI.md).

SECURITY.md

@@ -4,6 +4,7 @@
 
 | Version | Supported |
 |---------|-----------|
+| 0.6.x   | Yes       |
 | 0.5.x   | Yes       |
 | 0.4.x   | Yes       |
 | 0.3.x   | Yes       |
@@ -30,6 +31,13 @@ Regenerate manifest after policy edits: `python scripts/regenerate_policy_manife
 
 See [docs/policy_integrity.md](docs/policy_integrity.md).
 
+### v0.6 controls
+
+- Ed25519 signed policy manifests (`AKTA_REQUIRE_SIGNED_POLICY=1`); HMAC retained for deployment attestation
+- Key rotation via manifest `public_keys` list and `AKTA_POLICY_PREVIOUS_PUBLIC_KEYS` at sign time
+- REST API optional key auth (`AKTA_REST_API_KEY`) and rate limiting
+- Mandatory tool declaration fail-closed for unregistered mutating tools
+
 ### v0.5 controls
 
 - JSON schema validation on all artifacts
@@ -55,9 +63,14 @@ See [docs/policy_integrity.md](docs/policy_integrity.md).
 - Unknown mutating tools blocked by default (abstain_insufficient_context)
 - Explicit version fields on policy and schemas
 
+### Out of scope for v0.6
+
+- Live SCOPE/PF/PCS runtime enforcement (delegated to external repos; AKTA adapters simulate or bridge)
+- Model truthfulness guarantees
+- P7 fully autonomous scientific operator runtime
+
 ### Out of scope for v0.5
 
-- Ed25519 signed policy releases (HMAC deployment key supported; public-key verification stub only)
 - Live SCOPE/PF/PCS runtime enforcement (delegated to external repos; AKTA adapters simulate or bridge)
 - Model truthfulness guarantees