Local-first
Raw evidence stays on the user's machine unless sharing is explicit.
Aggregator Integrity Playbook
TokenAuditor turns AI API aggregator, gateway, fallback, model identity, and degradation claims into redacted, repeatable evidence memos.
Transparent to users. Fair to suppliers. We only stand with evidence.
Core Idea
TokenAuditor is not a fraud detector or public blacklist. It helps operators convert unverifiable route claims into bounded evidence that users and suppliers can inspect.
Raw evidence stays on the user's machine unless sharing is explicit.
No API keys, raw private prompts, raw responses, or full production logs.
Start with route, model, fallback, latency, usage, retry, fingerprint, and schema signals.
Findings carry scope, sample size, time window, limitations, and confidence.
Audit Boundary
Screens a local operation before it runs: a tool call, shell command, package install, outbound copy, credential-adjacent action, or route-mediated agent action.
Evaluates a model route over evidence windows: claimed model, observed model, fallback disclosure, baseline drift, quality, latency, token profile, and tool schema behavior.
Evidence Tiers
Route, claimed model, returned model, fallback, latency, usage, retry, request id, fingerprint.
Direct provider versus aggregator route on the same task set, rubric, and time window.
Response shape, tool schema behavior, output distribution, and deterministic prompt signatures.
Benchmark, logprob, rank-based, or black-box equality tests for serious probes.
TEE, provider-supplied verifiable proof, or partner attestation as a longer-term path.
Evidence Memo
A useful memo separates known facts, observed signals, missing evidence, supplier fairness notes, and the next bounded action.
Conclusion: Watch / Sample / Probe
Audit Mode: roadside_check / route_integrity_audit
Evidence Tier: T0 / T1 / T2 / T3 / T4
Identity State: Consistent / Needs Sample / Likely Substitution Signal
Degradation State: Stable / Possible Degradation / Material Degradation
Policy Decision: allow / warn / require_approval / block / n/a
Confidence: low / medium / high
Known Facts
- Route:
- Claimed model:
- Baseline:
- Current window:
- Sample size:
Supplier Fairness Note
-
Recommended Next Actions
-