Rendered from docs/obligations/0168-a-saturated-pattern-set-and-a-clean-corpus-are-the-same-zero-and-no-report-separates-them.md in the Headwater corpus. Every document on this half of the site is typed by the taxonomy the descriptor names: corpus.json.

A saturated pattern set and a clean corpus are the same zero, and no report separates them

Context

voice.forbidden_construction reads a curated pattern set for each category that a voice regime forbids. Edition one of that set holds forty patterns across three categories. Over this corpus it generates 258 instances and reports nothing.

A zero from a lexical rule carries two readings, and the report states neither. The first reading is a corpus that holds none of the fault. The second is a pattern set whose entries are absent from the language the corpus writes. A direct count separates them here: not one of the forty patterns of edition one occurs in a declarative document of this corpus.

HW-OBL-0002 took the first reading. It asks for an adjudicated sample of at least 50 findings per category. On the day this record was written, that record read every category as an order of magnitude short of the bar. It now carries the reading of 2026-09-11 instead: 36 findings for one category, and none for the other two. It concludes that no larger corpus under the declarative regime exists to run against, so the claim waits on an adopter.

Edition two of the change_narration set reported 60 findings over the same documents on the day it landed. The reading on commit 58f46a8d, made on 2026-09-11, is 36. Two merged rewrites of the corpus removed the sentences that the set was curated against. So the widened set carried the bar of HW-OBL-0002 from out of reach to within reach. The corpus then carried it back out, at 36 findings against a required 50. The bar answers to the pattern set and to the corpus together, and a reading of either one alone dates within days.

The same reading shows that edition two leaves the other two categories saturated. On that commit future_intent matches on 0 of its 14 patterns over the 291 documents that bind the declarative regime. phased_rollout matches on 0 of its 13 over the 293 that bind the declarative or the prospective regime. The three corpus occurrences of either set sit under docs/reviews/ and docs/evaluations/, which bind narrative and forbid nothing. So the shape this record describes is live for two categories of the three. A hand count is still the only instrument that separates the two readings.

The reading of 2026-09-27, on commit 0ee88a2d, replaces the figures of 2026-09-11 above. HW-OBL-0002 holds it, and this record does not restate it. change_narration reports 22 findings, not 36. future_intent reports one finding, and that finding is false. So a zero became a one, and the one does not show that the set detects the fault. A count of findings alone cannot show that difference, which is the gap that this record names.

Curation against a corpus rejects candidates that intuition admits. Three entries measured badly enough to stay out, and each failed for a reason a reader of the phrase alone would miss. the retired matches four sentences of this corpus and none is a narration, because "retired term" is the vocabulary of a rule here. at one point matches the positional sense in "worth making at one point in a text". was replaced matches three sentences that state a measurement procedure rather than a change to the system.

Obligation

The corpus owes a reading, per lexical rule, of how many of the patterns of that rule match at least one sentence. A rule whose whole set matches nothing is reported as such on the run that reports its zero.

This reaches past voice. language.retired_term.used reads a list from the taxonomy and carries the same shape. A term retired long ago, and edited out of every document, reports zero on every run after that. The report says only that the rule passed.

Discharge

A coverage line that names, for each rule with a pattern set, the count of patterns that matched and the count declared. Nothing here asks for a verdict on that ratio. A set with dead entries is legitimate, and a set with none alive is the finding.