Unlike mock sandboxes that return canned JSON, this workbench executes the exact production TypeScript cores in-process against synthetic data: the 6-condition gate conjunction, the NCCI/MUE claim scrubber, the 835 ERA parser and conserving integer-cent waterfall, and the RFC 8785 hash-chain verifier.
POST /api/developers/sandbox/run and POST /api/mcpToggle any of the 6 canonical conditions below and run the live conjunction. Across all 2⁶ = 64 truth combinations, only all-6-true returns `PASS → GATED`. Failing only condition 5 (`reserve_capacity`) returns `DEFERRED_TO_DAY_N → DEFERRED`.
Encounter signed; addenda append-only; PUT returns 409.
Same-day 271 verified active commercial/Medicare coverage.
All CPT & ICD-10-CM pointers carry chart-grounded citations.
0 blocking NCCI PTP, modifier-25, laterality, or MUE errors.
Practice reserve sits above underwriting floor.
Rendering NPI 1999999907 active in NPPES, zero OIG exclusions.
How the shteg.ai sandbox guarantees that a test integration can exercise every deterministic validator while remaining mathematically incapable of touching live money or real patient records.
assertSandboxKey) refuses to initialize if a non-sandbox key is supplied — there is no boolean environment flag that can accidentally flip a sandbox tenant onto a live banking rail. TEST-SBX-0001.. MRNs, RFC 2606 @example.test emails, and Luhn-valid synthetic NPIs. Cross-clinic links are refused by assertSandboxClinic.evaluateGateDecision, scrubClaim, and computeWaterfall modules used in production. If an edit set is missing or a condition fails, the sandbox halts or marks not_verified rather than faking a pass.FM-015-001 through FM-015-026) test stacked CAS groups, CARC 97 bundling, PLB takebacks, COB secondary remits, and reversal claims.