Paraphrase-Difference Test¶

Fidelity review — instantiates Textual Close-Reading Mode

Holds a finished summary against the evidence map and flags what the compression dropped — modality, exception, speaker, ambiguity, sequence, or relation — before the summary is allowed to stand in for the source.

Close reading has to end in a paraphrase, but the paraphrase must earn its place. The Paraphrase-Difference Test is the terminal gate that sets the proposed summary beside the inspected marks and relations and asks what the compression lost. Its defining stance is that a summary is provisional until proven faithful: it may not silently become the answer. Where the analysis has surfaced modality, exceptions, speaker, preserved ambiguity, sequence, and relation, the test checks each of those against the summary; any material loss forces a revised paraphrase or an explicit compression note. It produces no new interpretation — it is the fidelity check that stands between the evidence and the gloss that will travel in its place.

Example¶

A product team publishes a plain-language summary of its privacy policy. The difference test lines the summary against the source, clause by clause. It catches a load-bearing drift: the source says the company "may share data with partners," while the summary says it "shares data with partners only when required." Two losses, both material — the modality has flipped from a permission to a described practice, and a limiting exception ("only when required") has been invented that the source does not state. The test's verdict is not "close enough": the summary is either rewritten to preserve the may and drop the phantom exception, or it carries an explicit note that it simplifies a broader permission. Either way, a convenient gloss is stopped from quietly narrowing what the policy actually reserves the right to do. Modality (may / shall / must) is a real linguistic feature; the company and clause are illustrative.

How it works¶

What distinguishes the test is that it runs last and compares — it does not draft the summary and does not gather evidence. It takes the finished paraphrase and walks the evidence map feature by feature, checking that each load-bearing property survived: modality, qualification, exception, speaker and attribution, any ambiguity that ought to be preserved, sequence, and formal relation. Each material loss is dispositioned — repaired in a new paraphrase, or recorded as an explicit compression note — and genuine residual ambiguity is logged rather than smoothed into false clarity. A summary passes only when its differences from the evidence are either eliminated or openly declared.

Tuning parameters¶

Loss-materiality threshold — how large a compression loss must be before it forces a rewrite rather than a footnote. Tight thresholds preserve fidelity but resist any summary at all.
Feature-checklist scope — which features are load-bearing for this question (modality for obligations, speaker for testimony, sequence for narrative). Checking everything is thorough but slow.
Repair versus annotate — whether a detected loss triggers a rewritten paraphrase or a retained compression note. Annotation is cheaper but leaves the imperfect gloss in view.
Ambiguity-preservation stance — how hard to keep genuine ambiguity intact versus resolving it for readability.
Round-trip depth — a single pass versus iterating until the paraphrase survives. More passes converge on fidelity but can over-hedge the summary into unreadability.

When it helps, and when it misleads¶

Its strength is catching the specific, high-consequence losses — a dropped may, a vanished exception, a flattened speaker — that make a paraphrase quietly false while it still reads faithful.^[1] It is the archetype's guarantee that synthesis stays accountable to grain.

Its failure modes are pedantry and evasion. Pushed too hard it blocks every usable summary over trivial losses; pushed too little it is simply skipped or run backwards — the summary is written first and the "source reading" is tuned to match it, the bad paraphrase that "becomes the answer first and recruits quotations afterward." The discipline is to run the test against an evidence map produced before the summary, and to treat a failed test as a reason to change the summary, never the evidence.

How it implements the components¶

paraphrase_deferral_gate — it is the gate in operation: the summary is kept provisional and admitted only once it has demonstrated fidelity to the inspected marks and relations.
ambiguity_and_uncertainty_ledger — material losses and genuine residual ambiguity that a summary cannot preserve are recorded here as compression notes instead of being smoothed away.

It checks a summary against the evidence but does not build that evidence: warranted claims come from Micro-to-Macro Claim Matrix, recurrence and placement from Motif Concordance and Recurrence Map and Sequence and Placement Map. It gates the paraphrase, not the whole analysis — the stop rule for the reading as a whole is Interpretive Saturation Log.

Instantiates: Textual Close-Reading Mode — the test enforces the archetype's demand that a summary retain the grain it compresses.
Consumes: Micro-to-Macro Claim Matrix and the evidence maps — the grain-level record the summary is tested against.
Sibling mechanisms: Micro-to-Macro Claim Matrix · Interpretive Saturation Log · Counterreading Stress Review · Motif Concordance and Recurrence Map · Sequence and Placement Map · Parallel-Passage Comparison · Omission and Negative-Evidence Audit · Diction, Syntax, and Punctuation Trace · Line-by-Line Annotation Protocol · Close Reading Table · Contextual Reentry Memo · Edition or Variant Collation Table

Notes¶

The test is only meaningful if a real grain pass produced the evidence map first; run without one, it becomes a rubber stamp on whatever summary it is handed. It is deliberately the last gate before a paraphrase is trusted to travel on its own.

References¶

[1] Epistemic modality — the grammatical marking of how certain, permitted, or obligatory a proposition is (may / might / must / shall). It is the feature a hurried paraphrase most often drops or shifts, and because a single modal can turn a permission into a requirement, it is the first thing this test checks. ↩