Release AKTA v0.6.0#3
Merged
Merged
Conversation
Pin AKTA at v0.6.0 and declare security optional deps so downstream installs match the release train.
Centralize review packet assembly so evaluate and adapters share one grant-aware shape for v0.6 review loops.
Grant-scoped evaluation must reuse the same admissibility surface as live SCOPE decisions for reproducible v0.6 benchmarks.
Gate checks now align with iterative review so blocked transitions fail before expensive SCOPE round-trips.
Contract additions keep Python and adapter boundaries aligned with SCOPE v0.6 grant shapes.
Cryptographic policy integrity closes the gap between hashed bundles and operator-supplied trust anchors in v0.6.
Regeneration must emit the same signed manifest shape CI and operators verify at install time.
New registry entries document adapters shipped in v0.6 so PCS hashes and admissibility stay in sync.
Manifest pins must match the expanded registry before release artifacts are cut.
Stricter classification guards reduce prompt-injection and schema-evasion paths in production adapters.
Overlay loader changes enable domain overlays without weakening default admissibility checks.
Protocol types mirror SCOPE v0.6 so client and tests share one import surface.
The SCOPE client must orchestrate review packets and narrow grants used by v0.6 demos and PCS export.
Schema-backed VSA imports keep rich reports valid before they enter admissibility evaluation.
VSA adapter upgrades let PCS bundles include structured evidence without silent field loss.
Memory entries become first-class evidence inputs for v0.6 reconstructable experiment demos.
Trace import connects PF artifacts to AKTA records for reconstructable experiment workflows.
Gym scenarios feed adversarial and holdout evals without hand-maintained JSON fixtures.
Export must emit the full v0.6 evidence chain so external verifiers reproduce bundle semantics.
Runner automates PCS-Bench against out-of-tree checkouts for cross-repo CI gates.
REST surface gains validation and error semantics expected by external orchestrators.
MCP clients need parity with CLI evaluate paths for agent integrations.
Middleware enforces admissibility between graph nodes so agents cannot skip review gates.
Guardrails adapter applies AKTA decisions before model outputs reach downstream tools.
Inter-rater metrics quantify classifier stability before holdout and adversarial gates.
Adversarial transitions stress review-loop edges that linear scenario suites miss.
Holdout runner keeps private labels out of training fixtures while reporting parity metrics.
Oracle-independent evals must cover expanded scenario corpora shipped in this release.
Central runner dispatches holdout and adversarial corpora alongside public scenarios.
Domain overlay encodes materials science admissibility without forking core policy.
Tests lock grant-aware review behavior before adapter and PCS layers diverge.
Signature tests prevent shipping manifests that fail operator trust verification.
Security tests assert rejected inputs cannot bypass admissibility classification.
Adapter tests exercise scientific_memory, labtrust, and PF import paths together.
Eval tests ensure harness outputs remain stable as scenario corpora grow.
Shared helpers keep live SCOPE contract tests reproducible across checkouts.
Existing adapter tests align with hardened REST and MCP behavior shipped in v0.6.
Contract test ensures bench runner works against real suite layouts.
Shape tests catch breaking SCOPE exports before they reach integrated demos.
Demo contract test validates end-to-end drift narrative against live SCOPE stubs.
Updated contracts document grant-aware import expectations for operators.
Top-level drift examples reflect v0.6 review-loop fields for documentation walkthroughs.
Sidecars stay aligned with grant-aware review packets used in the drift narrative.
Manifest pins must match v0.6 policy hashes bundled with the drift demo.
PCS bundle decisions include review metadata required for v0.6 verification.
PF obligation and review trigger artifacts complete the drift PCS chain.
SCOPE grant, decision, and review packet entries match live import expectations.
End-to-end artifacts demonstrate memory, VSA, PF, and PCS-Bench linkage for v0.6.
PCS bundle provides hashed manifest members for external reconstructability checks.
Script ties adapters and PCS export into a reproducible v0.6 integration story.
CI runs Ed25519, holdout, and external bench steps required before v0.6 release.
Drift demo script follows grant-aware SCOPE client behavior introduced in v0.6.
Operators need sibling-repo paths and secrets guidance for green cross-repo jobs.
Make targets mirror GitHub Actions so local verification matches CI.
Integrity doc explains how to rotate keys and verify manifests in v0.6 deployments.
Bridge documentation reflects SCOPE client changes integrators rely on in v0.6.
Limitations set expectations for holdout, adversarial, and cross-repo coverage gaps.
Changelog captures review loops, policy signing, adapters, and eval expansions.
Top-level docs advertise v0.6 integrations and the hardened trust model.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
evaluate_with_grant(), fullreview_decision.py, human review packet export/import, and Ed25519 policy signing with HMAC deployment attestationscripts/demo_reconstructable_experiment.pyend-to-end demomaterials_expert_v0overlay.github/CROSS_REPO_CI.mdfor cross-repo checkout expectationsTest plan
release/v0.6(unit, contract, and v0.6 security/cross-repo gates where configured)makev0.6 CI/eval targets pass locally or in CIscripts/demo_reconstructable_experiment.pyand verify PCS bundle verification0.6.0/akta-policy-manifest-v0.6