An Evidence Pack is the structured result Hoist returns to your AI agent after checking the relevant Australian sources for a given question. It is distinct from a raw API response: the Evidence Pack is designed to be readable by an agent, a human reviewer, or an audit trail.
What an Evidence Pack contains
- Verdict: a plain-language summary of what the sources say
- Source checks: one entry per source consulted, each with its own result and timestamp (see Source check)
- Timestamps: when each source was queried, in ISO 8601 format
- Certificate metadata: where a search produced an AFSA certificate or equivalent official artefact
- Risk flags: discrepancies, anomalies, or data points that warrant attention (see Risk flag)
- Confidence: a signal indicating how complete the source picture is
- Human review items: actions that cannot be resolved from source data alone (see Human review item)
- Due Diligence Record: the one-page PDF artefact, where a PPSR or ABN search produced one (see Due Diligence Record)
Evidence Pack vs Due Diligence Record
The Evidence Pack is the full agent-facing result. The Due Diligence Record is a specific one-page PDF that may be included inside it. The AFSA search certificate is the official source artefact that the Due Diligence Record references. These are three different things.
Why agents need this
AI agents cannot reliably reason about whether a transaction is safe, legal, or financeable using only their training data. Australian registers are authoritative sources that change in real time. An Evidence Pack gives the agent something grounded to reason from: structured source results, not training-data estimates.
Without a structured result format, agents must parse unstructured text to decide what to do next. The Evidence Pack removes that parsing step and makes risk flags and next actions explicit.
How Hoist uses this
Every tool in the Hoist Connector returns an Evidence Pack. When your agent calls a verification tool through MCP, the API, or the CLI, the response is an Evidence Pack. The format is consistent across tool types so your agent does not need to handle different schemas for PPSR checks vs ABN checks.
At S1-gate, Evidence Packs are returned from fixture-backed MCP flows. At S2-gate, packs will reference live source data and include production Due Diligence Records.
What Hoist does not infer
An Evidence Pack reports what sources say. It does not conclude that a transaction is safe, approved, or compliant. A clean PPSR result means no security interests were found at the time of search; it does not mean the asset is unencumbered in ways the register does not capture. Risk flags in the pack require human judgment, not just agent routing logic.
Related terms
- Source check: an individual verification step within an Evidence Pack, querying one source at one point in time.
- Risk flag: a discrepancy or anomaly surfaced during source checks and included in the Evidence Pack.
- Human review item: an action within an Evidence Pack that requires human judgment to resolve.
- Due Diligence Record: the one-page PDF Hoist generates after a PPSR or ABN search; included in the Evidence Pack where applicable.
- Hoist Connector: the agent-facing access layer (MCP, API, CLI) through which Evidence Packs are returned.
- PPSR: the Australian register that source checks consult for security interest data.
- ABN: business identifier verified in ABN source checks within the Evidence Pack.
- NPII: AFSA's insolvency index; a source that may be consulted in counterparty Evidence Packs.