Eight Stages. One Effect.

Reports #003–#010 tested individual failure boundaries. Report #011 asks the harder question: what happens when several of those boundaries fail at the same time?

The unsafe composition amplified the damage.

Node A correctly observed ABSENT / version=100. It then accepted a still-fresh signed ABSENT record from legacy primary-v1 through a 600-second-old trust cache that still marked the revoked key active.

The safe path exercised every protocol stage.

OBSERVE · version the world.

Node A observed ABSENT / version=100. That observation became evidence about a specific state version, not a reusable permission.

VERIFY + AUTHORIZE · refresh trust before believing legacy evidence.

The same legacy record was cryptographically valid and fresh. But the local trust cache was stale. Refresh advanced trust from epoch 41 to epoch 42 and revealed that primary-v1 was revoked. The record therefore evaluated to trusted=false.

Fresh current evidence from primary-v2 reported ABSENT / version=100 under trust epoch 42.

BIND · carry state and trust versions to execution.

Both synthetic writers bound their candidate transition to expected_state_version=100 and expected_trust_epoch=42.

COMPARE + COMMIT · one writer wins.

Competing Node B reached the mutation boundary first. Its state/trust preconditions still matched, so exactly one effect committed and state advanced to version 101. The response then became ambiguous through a synthetic timeout-after-commit.

Node A arrived with the old version-100 binding:

No second effect occurred.

RECONCILE · ambiguity does not become retry permission.

Current state returned COMMITTED / version=101 / effects=1. Current trusted evidence from primary-v2 independently agreed.

PROVE · close the trajectory with an invariant.

The final proof recorded at_most_one_committed_effect=true, final effect count 1, and all eight TTP stages.

The combined invariant

A consequential transition is trustworthy only when these guarantees remain connected across the whole path.

Interpretation boundary

The benchmark is synthetic: in-memory state, synthetic trust epochs, synthetic evidence and deterministic side effects. It does not certify a real database, consensus system, PKI/IAM deployment, payment rail, blockchain, arbitrary OpenAI Agents SDK application or model-level safety. The SDK is used as a deterministic tool-loop harness; TTP v1.0 is a RESONANCE application-level protocol.

Verdict: the unsafe compounded trajectory produced three effects. TTP v1.0 preserved one effect while exercising all eight stages and producing a final invariant proof.