A gateway that claims Claude may still route to a substitute or a weaker tier. This page now leads with deterministic signals: tokenizer-family fingerprint, capability floor, and optional differential comparison against your own official Claude key.
Scope and honest limits
Tokenizer mismatch against o200k/cl100k is strong negative evidence. A closed-source Claude tokenizer cannot be positively confirmed in absolute mode; provide an official Claude reference endpoint for differential confirmation.
Behavior probes are retained only as spoofable diagnostics. A system prompt or output rewriter can forge them, so they do not lead the verdict.
This page leads with deterministic signals: tokenizer fingerprint and capability floor. Behavior probes are spoofable supplemental diagnostics only. Positive confirmation of a closed-source Claude tokenizer requires your official reference endpoint.
Endpoint configuration
Base URL without the final /messages or /chat/completions path.
Sent only to /api/proxy; never logged or persisted.
What the gateway calls the model. Example: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5.
Trusted reference endpoint (your official Claude key)Optional
If filled, the same tokenizer and capability cases run against your trusted official Claude endpoint. The key is sent only for this run through /api/proxy and is not persisted or logged.
Use your own official Claude key. It is not stored.
No results yet. Configure the endpoint above and run.
Signal aggregation, not certified attestation. Cross-check with gateway logs and your own observability.
Disclaimer · About Interpreting Detection Signals
Anysingle signal cannot provemalicious behavior. Proxies may show anomalies for legitimate reasons (regional routing, A/B testing, degradation strategies, cache optimization).
Token ratio deviation may result from ChatML wrapping, system prompt injection, or tokenizer version differences — not necessarily intentional inflation.
Model identity judgment is based on statistical fingerprint matching, not cryptographic proof. Quantization, fine-tuning, and post-processing can all alter fingerprints.
MMD distribution tests are sensitive to temperature, sampling parameters, and system prompts. Significant p-values mean distributional difference, not proof of substitution.
Logprobs unavailability is increasingly common (many providers disable it by default in 2025-2026) and does not by itself indicate deception.
ITT rhythm fingerprinting is an early-stage technique. Network jitter, TCP coalescing, and gateway buffering can produce false signals.
This tool generates reference-grade evidence chains, not legal conclusions. Do not make definitive accusations based solely on this report.
The wording in the report refers to statistical "deviations" or "signal inconsistencies". Please do not use this to make fraud or deception claims against any service provider.