Documentation

Structured summary of a threat-model test case.

Which list of a suite's ThreatModelsFor instance a threat model was run from. It decides how the counts in a ThreatModelSummary are to be read: the same tmsFailed (the attack's mutated transaction still validated) is a vulnerability for a Claimed model, the required outcome for an Expected one, and a known, tolerated artifact for an Accepted one. A consumer that alerts on failed > 0 must therefore filter on the category first.

Whose fault a failing threat-model test case is.

Only one cell of the slot-by-outcome matrix is the contract's fault; most red is a stale declaration. Saying which lets a reader — or a dashboard — tell "the contract regressed" from "somebody needs to update the instance", and stops a resolved vulnerability from reading as a broken contract.

The prefix a failure message leads with, so the fault is greppable.

The Tasty group that a category's per-model test cases live under. The single source of both the group the test tree is built with and the names the streaming reporter matches on, so that a category cannot end up in a group the reporter does not recognise (--test-id resolves a per-model test's Positive tests prerequisite from these names).

Mutable storage for threat-model summaries, owned by the reporter.

A recorder closure passed to test bodies via Tasty's option system. The default no-op makes summaries silently dropped when the streaming reporter is not active.

Internal option carrying the live store. Set by defaultMainStreaming alongside the recorder so the reporter can read summaries back out.

Callback for recording iteration traces as pre-serialized JSON. Arguments: group name, category ("positive"/"negative"), pre-serialized trace JSON. Default is a no-op (zero overhead when streaming is not active).

When trEnabled returns True, test bodies use the expensive traced code path (building IterationTrace values with UTxO snapshots, transaction summaries, and JSON serialisation). When it returns False (the IsOption default), the cheap runActions path is used instead, avoiding all that work.

trEnabled is an IO action so that the decision can be deferred until the streaming reporter has parsed --no-trace and written the shared IORef.

Internal option carrying the coverage index, i.e. all the possible code range that can be reported as covered.

Allocate fresh storage. Call once per reporter run.

Build a recorder that writes into the given store.

Look up a summary by key (does not delete).