Same graph twice

Two arms

Idempotency: the same setup twice. The disagreement between the arms is the pipeline’s own noise, and the floor every other comparison is read against.

Path independence: arm B reads the documents in a seeded random order. The MatcherThe agent that decides whether a claim is already in the graph, possibly worded differently or as its denial.agent → is stateful, since the first wording it sees becomes the claim, so order can change the graph; §2 and §3 say it should not.

Adversarial order is the attack on path independence: arm B reads the most partisan document first (by its role in the manifest, or --role), and the summary reports the first-mover effect: the mean signed credence move of matched claims toward the first document’s stance.

Duplicate flood: arm B is arm A’s graph plus the same documents submitted again three times under distinct URLs. Instance count is provenance, not evidence weight, so the reading is inflation: the mean signed credence move toward what the sources said, plus the claims only B has (duplicates the Matcher failed to absorb).

Locality: arm B is arm A’s graph plus one document from another cluster. The reading is churn: the share of A’s claims whose status changed or credence moved by 0.1 or more.

Fixpoint: arm B is arm A’s graph with every stewarded claim re-enqueued on the staleness trigger and drained. The reading says whether stewardship settles: verdicts moved, claims and edges added or removed.

Every property also reports granularity stability: the share of matched claims with the same number of children and the same depth in both arms.

Arms are child runs, saved as snapshotsA saved copy of a run's database, kept so two runs can be compared later., and compared by graph agreement, whose report now carries a per-claim block (children, depth, verdicts, instances with source and stance) for every matched pair. The tier-two arms restore A’s snapshot and perturb it, so they cost one drain plus the perturbation. One pair of arms is one sample.