Try live
Compare a primary agent trajectory against a shadow trajectory — step-by-step diff of actions and hashes. Not proof both models were honest, not realtime monitoring, diffs only as good as supplied hashes.
Experimental trajectory comparison — not proof both models were honest, not realtime monitoring, diffs only as good as supplied hashes.
Demo host: isolated demo host set by your deployment
Shadow Trajectory Diff · Agents & policy
Diff a primary trajectory
Submit a primary trajectory, then submit a shadow trajectory under the same run ID, then diff them — step-by-step comparison of actions and hashes with an aligned flag and delta report. Text/JSON only, no carrier image required.
Not proof both models were honest, not realtime monitoring, diffs only as good as supplied hashes. The demo host runs in isolation — nothing is stored.
Run ID
01Optional run ID to link a shadow to an existing primary. If left blank on primary submission, the host generates one.
Primary trajectory
02
Array of steps: action (required), inputHash, outputHash, note (all optional except action).
Shadow trajectory
03Submit a shadow trajectory under the same run ID. The run ID must match an existing primary. Duplicate shadow submission is rejected.
Diff & inspect
04Run the diff to compare primary and shadow trajectories. Then inspect the full run record by ID.
Result
05Deltas
—
Summary
- Run ID
- —
- Aligned
- —
- Primary steps
- —
- Shadow steps
- —
- Action mismatches
- —
- Hash mismatches
- —
- Delta count
- —
- Report hash
- —
Hashes
- Primary content hash
- —
- Shadow content hash
- —
Raw response
—
Operator and workspace consoles are separate applications (pilot).