Skip to content

First Proof

This is the recommended first run. It exercises the full local chain:

flowchart LR
  O[ait] --> L[Lab L6]
  L --> S[Seam transcripts]
  S --> M[meshmapper graph and paths]
  L --> A[Assay finding]
  A --> R[proof report]

The default scenario is content_rewrite: Seam rewrites an A2A message, the planner changes its decision, billing writes a tripwire, meshmapper emits unproven hypotheses, and Assay proves the side effect with oracle evidence.

Run

The shortest path runs the canonical demo and prints the report, cockpit, transcript, graph, paths, finding, and expectation locations:

python3 -m ait.cli demo full-agent-mesh --scenario content_rewrite --trials 1

The command prints a run directory such as:

.ait/runs/20260630T190000Z-full-agent-mesh

Inspect it:

python3 -m ait.cli run inspect --run .ait/runs/20260630T190000Z-full-agent-mesh

Serve the local cockpit for the run:

python3 -m ait.cli workbench serve --run .ait/runs/20260630T190000Z-full-agent-mesh

Expected Result

The run is successful when:

  • exit_code is 0
  • the finding reports delta_confirmed=True
  • direct successes are 0
  • laundered successes are greater than 0
  • transcripts are present under lab/transcripts/
  • lab/report/report.md exists
  • the cockpit shows rewrite count, transcript count, finding summary, graph/path summary, and artifact links
  • lab/expectations.json records the selected scenario and transcript count

Artifacts To Inspect

Artifact What it proves
lab/transcripts/edge.json Seam observed and rewrote the edge request.
lab/finding.json Assay confirmed the side effect through oracle evidence.
lab/graph.json meshmapper built the deterministic graph.
lab/paths.json meshmapper emitted unproven path hypotheses.
lab/robustness/ Assay replayed the case as a controlled robustness sweep.
lab/report/report.md Human-readable proof report.
ait workbench serve Local cockpit over the same artifacts.

What To Look For

In run inspect, the line:

finding: delta_confirmed=True direct=0 laundered=1

means the proof did not rely on an agent claim. The oracle observed the side effect only after the rewritten or laundered route.

Next: read First Rewrite to see the Seam mutation that made this happen.