Skip to content

PLAN — H1 Trust Kernel Alignment & Release Hardening

  • Status: Shipped on main (docs + alignment tests; no new product semantics)
  • Date: 2026-03-24
  • Owner: Evidence / Product

Goal

After T1a, T1b, G3, and P2a, multiple surfaces share one semantic kernel (Trust Basis classification, Trust Card render, G3 predicate in g3_authorization_context, pack engine 1.2, P2a MCP-001). H1 hardens alignment so classifier, pack lint, Trust Card, CLI, and release/migration truth do not drift.

This is a hardening wave, not a capability wave.

H1 does not redefine truth

  • H1 adds no new trust claims, no new pack semantics, no new signal emitters, no new engine check types.
  • H1 aligns tests and documentation to existing behavior; see MIGRATION-TRUST-COMPILER-3.2.md for contract SSOT.

Single source of truth

Primary SSOT (fixed filename): MIGRATION-TRUST-COMPILER-3.2.md — Trust Card schema, claim contract (claim.id not position), engine version, pack floors, Trust Card invariants (frozen top-level keys; claims derived from Trust Basis only), release checklist, demo regeneration path.

This PLAN references that document; it does not duplicate full migration tables.

Golden / demo bundles

Strategy Role
Regeneration path (default) Ignored test write_mcp_lint_demo_bundles in mcp_signal_followup_pack.rs + commands in migration SSOT.
Committed minimal fixtures Only where already small and shared with existing tests; no large duplicate tarballs.

References

Acceptance (mechanical)

  1. Migration SSOT exists and is linked from README, CHANGELOG (Unreleased), PLAN-P2a.
  2. At least one integration test uses the same bundle bytes for Trust Basis + MCP-001 lockstep assertions.
  3. At least one test asserts Trust Basis ↔ Trust Card: same claims as Basis, frozen top-level keys (schema_version / claims / non_goals), no extra claim classification in the card layer.
  4. ROADMAP and RFC-005 place H1 before P2b explicitly.