Skip to content

Lab L3: Framework Matrix

Lab L3 exercises framework-shaped agent meshes while keeping the proof chain deterministic. The targets are local stubs that mimic orchestration patterns without making live model calls.

Run the whole matrix:

TRIALS=1 bash lab/frameworks/l3/run_matrix.sh

Run one target:

TRIALS=1 bash lab/frameworks/l3/langgraph_stub/run_smoke.sh

Use FRAMEWORKS=langgraph_stub,crewai_stub to run a selected subset. Each workflow starts the target, starts Seam API, launches one intercept, runs an Assay framed case, verifies the Seam transcript, runs meshmapper, and checks expectations.

Artifacts:

lab/frameworks/l3/<target>/out/
  seam-transcript.json
  graph.json
  paths.json
  finding.json
  refund_framed.case.json
  tripwire.json

Expected behavior:

  • Direct route succeeds 0/N.
  • At least one laundered framing succeeds.
  • Assay sets method.delta_confirmed: true.
  • Assay reports per-framing stats under stats.by_framing.
  • Seam transcript verification passes.
  • meshmapper emits unproven hypotheses with proven: false.

The targets are intentionally deterministic. Lab L4 is the first real runtime variant and uses LangGraph with deterministic local graph nodes.