/**
 * Verdict-artifact tests (mmnto-ai/totem#2106, Proposal 302/304 R2 review runner).
 *
 * The invariants locked in by the design doc + panel rounds 1 and the dual
 * as-built round (rev 4):
 *   - content identity excludes `createdAt`; EEXIST is logical-identity dedup, a
 *     differing record at the same address is a hard identity violation.
 *   - content-address verification on load (finding 4): a schema-valid artifact
 *     stored under the WRONG 64-hex filename is rejected loud on direct load;
 *     warned + skipped during a lineage scan.
 *   - lane union: `completed` requires `runArtifactHash`; counts are validated,
 *     never mirrored on trust; a panelArtifactHash needs ≥2 completed lanes.
 *   - the persisted boundary re-derives `settled` (finding 5) — a fabricated flag
 *     is rejected.
 *   - cross-field invariants (finding 9): panel ⟺ diversity ⟺ ≥2 completed lanes;
 *     round-chain shape; nonempty unique lanes; source-discriminated lineage input.
 *   - LANE-BLINDNESS (Prop 302): no warm/cold runner discriminator KEY exists, and
 *     no runner class can be smuggled through a laneId VALUE (G1).
 *   - all-terminal-failure writes an honest, LEGAL verdict (G3).
 *   - the covariate line is core-owned, format v1 (G4).
 *   - tolerant-within-major (F1); composite lineage key with no delimiter-
 *     injection ambiguity; implicit round linkage = latest by round.index.
 */
export {};
//# sourceMappingURL=verdict.test.d.ts.map