API Magnifier does not treat the claimed model name as a verdict. The engine collects signals layer by layer, then aggregates an honesty score and finding list. Detection depth comes in onlytwo modes(selectable on the check page) to balance runtime and evidence completeness.
Network, protocol, metadata, tokenizer fingerprint, capability probes, and performance baseline — enough for everyday authenticity checks.
Runs every available layer on top of the standard check (including pool and consistency signals). Longer runtime, fuller evidence.
Conceptual descriptions below; probes and thresholds evolve by version — rely on findings in your report.
Enter your relay URL and key. Keys for one-off checks exist only in memory for that run.