schematron 0.3.2

# SVRL — Schematron Validation Report Language

SVRL is the standard XML vocabulary for a Schematron validation report, in
namespace:

```
http://purl.oclc.org/dsdl/svrl
```

conventionally bound to `svrl`. Emitting it is what makes this crate's output
consumable by existing Schematron tooling.

## Document shape

```xml
<svrl:schematron-output
    xmlns:svrl="http://purl.oclc.org/dsdl/svrl"
    title="..." phase="..." schemaVersion="...">

  <svrl:ns-prefix-in-attribute-values prefix="..." uri="..."/>*

  <svrl:active-pattern id="..." name="..." documents="..."/>
  <svrl:fired-rule id="..." context="..." role="..." flag="..."/>
  <svrl:failed-assert test="..." location="..." id="..." role="..." flag="...">
    <svrl:text>...</svrl:text>
    <svrl:diagnostic-reference diagnostic="..."><svrl:text>...</svrl:text></svrl:diagnostic-reference>*
    <svrl:property-reference property="..." role="..." scheme="..."><svrl:text>...</svrl:text></svrl:property-reference>*
  </svrl:failed-assert>
  <svrl:successful-report .../>

</svrl:schematron-output>
```

## Flat, not nested

`active-pattern`, `fired-rule`, `failed-assert`, and `successful-report` are
all siblings. The structure is implied by order, not by nesting: every
`fired-rule` belongs to the most recent `active-pattern`, and every
`failed-assert` and `successful-report` belongs to the most recent
`fired-rule`. This is what the reference implementation emits, because it is
what falls out of an XSLT streaming transform, and consumers depend on it.

The crate's internal report is a tree — `Report { patterns: [ { rules: [ { assertions } ] } ] }` —
and is flattened on the way out. That way the JSON output can keep the
structure while the SVRL output stays wire-compatible.

## Elements

| Element | Emitted when |
|---|---|
| `svrl:active-pattern` | A pattern begins running against a document. `@documents` is present only for `@documents` patterns. |
| `svrl:fired-rule` | A rule matches a node. Emitted once per matching node. |
| `svrl:failed-assert` | An `assert` whose test evaluated false. |
| `svrl:successful-report` | A `report` whose test evaluated true. |
| `svrl:text` | The instantiated human-readable message. |
| `svrl:diagnostic-reference` | Per `@diagnostics` reference on the assertion. |
| `svrl:property-reference` | Per `@properties` reference on the assertion. |
| `svrl:ns-prefix-in-attribute-values` | Once per schema `ns`, so a consumer can interpret `@location` and `@test`. |

## Attributes on failed-assert / successful-report

| Attribute | Value |
|---|---|
| `test` | The assertion's XPath source text, verbatim |
| `location` | Absolute XPath to the subject node, see [validation.md](validation.md) |
| `id` | The assertion's `@id`, if any |
| `role` | Resolved role: assertion's, else rule's |
| `flag` | Resolved flag: assertion's, else rule's |
| `see`, `icon`, `fpi` | Passed through when present |

## Verbosity

Emitting a `fired-rule` for every matching node is correct but verbose — a
large document can produce far more `fired-rule` elements than findings. The
writer therefore has two modes:

- **Full** (default, standard-conformant): every event.
- **Findings only**: `failed-assert` and `successful-report` only, plus the
  `active-pattern` elements that contain them.

`--svrl-findings-only` on the CLI selects the second.