/citation-integrity/verbatim-fragment-verification
Verbatim Fragment Verification
Verbatim fragment verification is the Citation Integrity check that asks whether a governed fragment actually appears in the extracted text of the source it cites. In the GARP rule vocabulary, this is the SR-07 boundary: a mechanical text-presence check, not a truth check.
The rule exists because a citation span can look plausible while still being miscopied, mapped to the wrong source, altered by extraction drift, or detached from its declared source. SR-07 makes that failure visible before promotion or export relies on the fragment.
What SR-07 checks
| Object | Role |
|---|---|
| fragment text | The exact span the governed record proposes to rely on. |
| source identifier | The registered source that the fragment claims to come from. |
| locator | The page, text span, section, timestamp, cell range, or profile-defined position used to make the citation inspectable. |
| extracted source text | The text representation of the registered source against which the fragment is checked. |
A fragment should point back to a registered source and a locator. The locator is not what makes the text true; it makes the claimed support inspectable by a reviewer.
Matching boundaries
The preferred pass condition is exact text presence. Because extracted text can carry line wrapping, spacing artifacts, or line-break hyphenation, SR-07 may also use documented normalization. A normalized pass should still say what kind of normalization was used.
| Mode | Meaning |
|---|---|
| exact text | The fragment appears in the extracted source text with the same characters and ordering. |
| whitespace-normalized text | Runs of whitespace are normalized so line wrapping and spacing artifacts do not cause a false failure. |
| dehyphenated text | Line-break hyphenation from extraction can be normalized when the rule documents that behavior. |
| documented profile normalization | A profile may add narrow normalization rules, but those rules must be named rather than hidden inside a pass result. |
Result states
| State | Meaning |
|---|---|
| pass | The fragment text is present in the extracted text under the documented matching rule. |
| warning | The fragment can be checked, but context such as locator precision, extraction confidence, or review state needs attention. |
| fail | The fragment does not match the declared source text, or the required source text is unavailable. |
Promotion block behavior
In workflows that require Citation Integrity, a failing SR-07 result blocks promotion of the affected fragment or record path. The block is a governance event: it preserves the fact that promotion was attempted and explains why the object did not cross the admission boundary.
A block does not erase the draft, the source, or the existing audit trail. It keeps unsupported material from becoming durable governed content until the source mapping, extracted text, locator, or fragment text is corrected and rechecked.
What SR-07 does not prove
- The source is authentic, admissible, complete, current, or legally significant.
- The quoted span is the best or only span that could support the proposition.
- The proposition attached to the fragment is true.
- The citation is well formed under a court, journal, editorial, or institutional standard.
- The extraction captured every visible mark in the original document.
- The mapping from citation label to source file is substantively correct.
Public fixture boundary
The public demonstration surface uses synthetic source material to show one passing fragment and one intentional drift case. That keeps the method inspectable without publishing private matter data or implying that private corpus results are part of this public page.
The important public claim is narrower: when a governed workflow says a fragment passed SR-07, it means the fragment text matched its declared extracted source text under the documented rule.