Skip to content
Lorca Labs
Live demo
Agents & policy

Shadow Trajectory Diff

Not whether the models agree — but whether they agree on what they did.

Try live Shadow Trajectory Diff Contact

The problem

When a primary agent and a shadow agent should produce the same trajectory, you need to surface exactly where they diverged — which action differed, which hash changed, and whether the overall run is aligned. Token-by-token diffs hide operational divergence in action ordering and output provenance.

What it does

01

Submits a primary trajectory as a series of steps (action, inputHash, outputHash, note) with optional agentId and runId.

02

Submits a shadow trajectory under the same runId for comparison; duplicate role submission is rejected.

03

Diffs the two trajectories, producing an aligned flag, delta list (mismatched actions and hashes at each step index), and a report hash.

04

Lists runs (metadata only) and fetches a full run record with both trajectories and their content hashes.

Who it’s for

  • Agent safety teams auditing primary vs shadow deployments
  • SREs verifying shadow-model parity
  • Compliance reviewers comparing execution traces

In context

Complements Sovereignty (agent governance) and the Stego Audit Carrier (carrier-bound audit).

Try it live in the studio

A no-signup, in-browser experiment against an isolated demo host. Load sample data, run the flow, and see the result — nothing is stored.

Try live Contact