Measurement — is the doctrine-declared-without-mechanism family shrinking in share?

This is a dated record, measured 2026-09-20. It states a measurement and its method; it amends no plan and ratifies no sequencing. The figures below are the record of one pass, not a live value: re-run the evidence scripts rather than transcribing a number from this file. Evidence and re-run instructions: f1-family-share-2026-09-20-evidence/.

Why this was taken

docs/rnd/ghi-landscape-reorganization.md closed on 2026-09-20 naming exactly one measurement it did not take, and the session handoff carried it forward as the single highest-value open loop. The R&D record's own words:

The campaign records 19 of 32 (59%) on 2026-09-02; this pass counts 14 of 57 (~25%). Different readers, different dates, no shared method.

Two judgments by different readers are not a measurement, and the active campaign's sequencing — Movement C's Close the doctrine-declared-without-mechanism family box, pulled to NEXT-IN-PRIORITY on 2026-09-02 — rests on the older figure. Taken on the operator's work selection, 2026-09-20.

Method

The defect in the prior figures is method, not arithmetic. This pass fixes the three variables that were free: one reader, one criterion, one day, applied to both cohorts.

The criterion is the family's own, from GHI #537 as the campaign box quotes it:

Layer X declares a discipline that Layer X does not mechanically enforce.

Applied operationally. An issue is a member when its body establishes both:

• (a) a declared discipline — a binding claim, contract, mandate or doctrine stated in some layer (rule text, docstring, skill mandate, doctrine document, CLI contract, REQ);
• (b) a witness gap — nothing observes that claim, or the witness's subject is narrower than the claim, or the mechanism exists and is never invoked.

Excluded, with the reason each exclusion is principled rather than taste:

Excluded shape	Why
missing capability, nothing previously declared	no (a) — nothing was promised
two witnesses disagreeing	the campaign's own #877 exclusion precedent: a coherence defect, not an unwitnessed discipline
plain code defect (wrong branch, import cycle, hardcoded literal)	no (a)
forward edge with no governed reversal	no declared discipline goes unenforced
a mandate already demoted to advisory in its own text	the family's completion criterion names this as discharged — "no third state"

The cohorts are reconstructed from GitHub rather than assumed. The 2026-09-02 cohort is every issue open at that pass's own measurement instant, pinned to 2026-09-02T05:00:45Z — the creation time of #933, the youngest issue that pass named, so the instant is a lower bound the campaign's own evidence fixes. Today's cohort is the 56 issues open on 2026-09-20.

Self-declaration is not the criterion. 56 of 56 open bodies carry a Class of failure section, extracted mechanically as an input. It is evidence a reader weighs, not a verdict: #998 self-describes as "the inverse of the doctrine-declared-without-mechanism family" and is a member under the criterion, because a witness that gates volume while coverage dies is precisely a witness whose subject is narrower than its claim.

The campaign's figure: numerator sound, denominator wrong

The 2026-09-02 pass is reproducible, because it stated its criterion and listed its 19 members. Both halves were checked.

• Numerator — verified. All 19 named issues were open at that instant.
• Denominator — wrong by 5. The open queue held 37 issues at that instant, not 32. The queue never held 32 on that date; it passed through 32 around 2026-08-25/26, a week earlier. The claim as written is 19 of 37 = 51.4%, not 59.4%.

The pass's own section heading — "Half the open GHI queue is this family" — survives the correction. The stated percentage does not.

Five surfaces carry the uncorrected figure. Two are immutable historical records and are correct as records (docs/reports/big-picture/2026-09-19.md, the 2026-09-20 handoff). Three are live: build-to-1.0-campaign-2026-08-16.md (three places), capability-control-review-2026-09-12.md:410, and docs/rnd/ghi-landscape-reorganization.md (twice, quoting it).

Result

One reader, one criterion, both cohorts, 2026-09-20:

	2026-09-02	2026-09-20	change
open queue	37	56	+19
family members	28	38	+10 (+36%)
share	75.7%	67.9%	−7.8 points

Sensitivity. Eleven calls were borderline. Flipping every one of them, in either direction, does not change the finding:

borderline treatment	2026-09-02	2026-09-20	change
as judged	75.7%	67.9%	−7.8 pts
all lean-includes excluded	67.6%	57.1%	−10.4 pts
all lean-excludes included	81.1%	76.8%	−4.3 pts

Under every treatment the family is between 57% and 81% of the open queue at both dates, the share declines by 4 to 10 points, and the absolute count rises.

Flow over the 18 days. 10 members closed; 20 new members were filed. The class discharges instances and produces them at twice the rate.

What this answers

The hypothesis is not supported. The R&D record's suggested direction — the family "shrinking in share while the queue grows", 59% → ~25% — does not survive a shared method. The apparent collapse was reader variance, not a trend: a 34-point drop measured across two readers becomes a 7.8-point drift when one reader applies one criterion to both dates.

The campaign's sequencing premise holds, and holds more strongly than the campaign claimed it. The box was pulled to NEXT-IN-PRIORITY on the strength of "19 of 32" and "instances close same-session, the class keeps producing." The first figure was overstated as a share and understated as a fact: the family is roughly two-thirds to three-quarters of the open queue, at both dates. The second is confirmed and quantified — 10 closed against 20 produced.

The share decline is real but small, and it is not discharge. Share fell 7.8 points while the stock grew 36%. The family is not shrinking; the queue is growing slightly faster than the family is, and the growth arriving from elsewhere is what moves the ratio.

Disposition (operator ruling, 2026-09-20)

This record was taken as a measurement and amended nothing on its own. The operator then ruled on its two routing questions, and the rulings are recorded here as provenance:

• The campaign's live sequencing text becomes a pointer, not a corrected number. Ruling, verbatim: "Pointer, not value", selected over correcting 32 to 37 in place. The two live sites in build-to-1.0-campaign-2026-08-16.md now cite this record instead of carrying a transcribed share, and a dated § Amendments 2026-09-20 entry records the re-measure. Correcting the integer would re-create the same defect one value later, with nothing observing it.
• The box does not move. The 7.8-point drift is not discharge, and the measurement strengthens the premise that placed the box at NEXT-IN-PRIORITY rather than revising it.

The archives are deliberately untouched. The 2026-09-02 amendment block, capability-control-review-2026-09-12.md:410 and both citations in docs/rnd/ghi-landscape-reorganization.md each accurately quote what their author found on their date. Editing evidence inside a record falsifies it; this record is their linked follow-up, which is the same route a published report's correction takes.

One thing this record does not settle. A transcribed measurement, binding on what sessions draw, with no witness over whether it is still true, is an instance of the family measured here. Whether that warrants its own work order — and how it sits against the queue-staleness chore the R&D run funded, and against GHI #1009 — is unrouted. ghi-author Step 0 would settle the prior art before anything is filed.