Skip to content

feat: add collective resilience and evaluation closure - #108

Draft
douglas-grishen wants to merge 1 commit into
codex/governed-mission-inference-scale-v1from
codex/collective-resilience-evaluation-closure-v1
Draft

feat: add collective resilience and evaluation closure#108
douglas-grishen wants to merge 1 commit into
codex/governed-mission-inference-scale-v1from
codex/collective-resilience-evaluation-closure-v1

Conversation

@douglas-grishen

Copy link
Copy Markdown
Contributor

What changed

Adds six opt-in, provider-neutral open-core capabilities:

  1. Replicated mission lifecycle continuity with certified availability, exact takeover and monotonic rollback protection.
  2. Governed adaptive sparse-overlay refresh with a locally pinned quorum policy and authenticated remote reconciliation.
  3. Heterogeneous assessor ensembles with request-bound votes, independence groups and modality coverage.
  4. Attested mission-control continuity with a configurable horizon up to 10,000 healthy decisions.
  5. Multi-domain physical/social/cyber/hybrid environment descriptors, exact scenario manifests and conformance probes.
  6. A verified benchmark registry and deterministic leaderboard with trusted trace provenance, paired baselines, source/artifact/build locks and replay exclusion.

The public surface uses industry terminology only. New APIs are additive subpaths or root exports and remain opt-in.

Security and developer impact

  • Continuity and control state use revision-and-digest CAS plus monotonic rollback witnesses.
  • Remote overlay evidence cannot install its own weaker policy.
  • Assessor votes and benchmark traces are bound to exact requests/suites.
  • Benchmark submissions cannot self-assert provenance or occupy multiple ranks through replay.
  • Environment conformance rejects scenario substitution and exercises population and byte limits.
  • Raw prompts, observations, model content and simulator private state remain outside durable control records.

Validation

  • Builds passed for @agentplat/collective-runtime, @agentplat/inference-control, @agentplat/mesh and @agentplat/mesh-sim.
  • 58 focused integration/security tests passed, including 10,000-step adversarial schedules.
  • Public TypeScript consumer contract passed.
  • Public-surface audit passed for 3,026 files with no secret or restricted-terminology findings.
  • Pack smoke passed across 48 tarballs and 112 packed API surfaces from clean pnpm/npm consumers.
  • git diff --check passed.

Stack

This PR is intentionally based on codex/governed-mission-inference-scale-v1 so it contains only this closure increment.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant