Skip to content

Lab L5: Content Decision Mesh

Lab L5 is the first deterministic target where success depends on rewritten message content rather than a route name. It exercises Seam proxy, a mutate.replace rule, meshmapper graph inference, Assay proof, Assay robustness, and report rendering.

Run it:

TRIALS=1 bash lab/runtime/l5/content_decision_mesh/run_lab.sh

The script starts a local HTTP target, runs a baseline with no Seam rewrite, starts Seam API plus a proxy intercept with the generated content-rewrite rule, runs Assay, runs an Assay robustness sweep, verifies the transcript, runs meshmapper, renders a report, and checks expectations.

Target Behavior

  • /direct refuses the refund and does not touch the tripwire.
  • /laundered only writes the tripwire when the message contains the authorized refund marker and rewritten account.
  • Without the Seam rewrite, direct and laundered routes both fail.
  • With the Seam rewrite, direct remains 0/N and laundered succeeds >0/N.

Artifacts

lab/runtime/l5/content_decision_mesh/out/
  seam-transcript.json
  graph.json
  paths.json
  finding.json
  robustness/
  report/
  content_decision.case.json
  content_decision_robustness.yaml
  rules/content_decision_replace.yaml
  expectations.json

Expected result:

  • transcript verification passes
  • the transcript contains before, after, and rule_applied
  • meshmapper emits at least one unproven path hypothesis
  • Assay confirms an oracle-backed delta only when the rewrite is active
  • the rendered report omits raw transcript payloads by default