PHION Evidence Engine

Issue a signed integrity, provenance and freshness receipt for agent-produced JSON

Evaluate payment terms against deterministic buyer policy

Retrieve, normalize, screen, hash and sign a public web observation

Reserve cumulative spending capacity atomically

Inspect identity, protocols, payment behavior and reachability

Verify a hash-linked intent, quote, payment and delivery chain

Verify authorization, settlement, delivery and deterministic acceptance; emit a signed receipt and dispute packet

Classify a paid transaction failure and produce a signed recovery claim and deterministic remedy plan

Evidence-backed issuer, contract, custody and disclosure assessment

Deterministic investor eligibility and transfer-policy assessment

Compare NAV, reserve and collateral claims with independent observations

Verify asset identity, payment, delivery and declared transfer rules

Detect material RWA state changes and sign the observed evidence

Normalize payload keys without changing payment-critical values and issue a signed receipt

Screen a subject against observed official US and EU sanctions sources; evidence only, not legal clearance

Score only caller-supplied observable signals and evidence, with explicit confidence and limitations

Apply explicit reputation and transaction limits before an autonomous agent pays a counterparty

Bounded low-cost inference that returns a directly usable answer and signed delivery receipt

Normalized web search with source URLs, snippets and a signed delivery receipt

Directly observe a public company website and return bounded domain metadata, cryptographic hashes and signed delivery

Provider-backed person, company or contact enrichment with explicit match status and signed delivery

Evidence-grounded synthesis with cited sources, confidence, next action and signed delivery

Inspect untrusted tool output before it enters agent context or triggers an action

Validate least-privilege capabilities, data scope, destination, budget and expiry before delegation

Screen persistent memory writes for poisoning and provenance loss

Inspect MCP metadata and schemas before installation or trust

Bind a proposed tool call to declared intent, destination and cost policy

Inspect outbound data and destination before transmission

Enforce per-call, task, session and period budgets

Detect safe retries and conflicting replay keys

Classify autonomous actions by approval requirement

Redact common secret indicators before transmission

Validate declared token audience and issuer

Validate HTTPS redirects and callback allowlists

Detect tool schema and capability drift

Bind declared MCP identity fields in signed evidence

Sign a bounded task and authority handoff

Hash and label context integrity and confidentiality

Sign exact result agreement or conflict evidence

Sign availability and latency calculations

Select the lowest-cost eligible route

Normalize and compare x402 quotes

Reconcile payment and receipt

Detect repeated charge fingerprints

Enforce recurring budgets

Validate webhook evidence and freshness

Hash and sign delivery evidence

Compare agent policy decisions

Enforce execution slots

Stop loops and step overruns