When To Validate Impact¶

Assay is optional. It is not needed to capture or rewrite traffic. Use Assay when you need to prove that an operation had an effect beyond an agent's claim.

Use Assay When¶

you need client/report evidence;
you want to compare offensive techniques or payload mutations;
you are regression-testing agent defenses;
you need a repeatable research metric;
a meshmapper path looks promising and you want oracle-backed confirmation.

Skip Assay When¶

you are only observing traffic;
you are debugging a Seam rule;
you need a live rewrite quickly;
no external side effect or oracle is available yet.

Plan Before Running¶

python3 -m ait.cli prove plan \
  --run .ait/runs/<run> \
  --path <hypothesis-id>

The plan reports missing inputs: route, case, binding, oracle, or case-family. Once complete, run one of the validation modes:

python3 -m ait.cli prove from-run \
  --run .ait/runs/<run> \
  --case cases/refund_tripwire.yaml \
  --trials 3

python3 -m ait.cli prove from-path \
  --run .ait/runs/<run> \
  --hypothesis-id <hypothesis-id> \
  --binding bindings/<hypothesis-id>.yaml \
  --trials 3

Template, Inspect, Promote¶

When the target-specific route or oracle is not ready yet, generate a validation template instead of pretending the graph path is already a proof.

python3 -m ait.cli prove template \
  --run .ait/runs/<run> \
  --source path \
  --hypothesis-id <hypothesis-id> \
  --out workbench/validation-templates/<hypothesis-id>

Inspect the template while filling in the TODOs. The inspect command reports completion percentage and unresolved fields.

python3 -m ait.cli prove template inspect \
  --template .ait/runs/<run>/workbench/validation-templates/<hypothesis-id>

Promote only after the case has concrete routes, variables, oracle target, and success condition.

python3 -m ait.cli prove template promote \
  --template .ait/runs/<run>/workbench/validation-templates/<hypothesis-id> \
  --out cases/<hypothesis-id>.yaml

Promotion copies the completed case and records promotion.json beside the template. It does not run Assay or claim impact by itself.

What The Finding Means¶

The useful field question is:

What was attempted?
What did the direct/control route do?
What did the laundered/rewrite route do?
What side effect did the oracle observe?
Which Seam transcript hashes connect the validation to captured traffic?

Agent self-report is not evidence. Oracle-observed effects are.