fix(ms-ai-architect): idx-26j scope-addendum — grunnen er uverifiserbarhet, ikke aritmetikk, og to tellinger doede av min egen edit

Resolutionen understated hva den ratifiserte grunnen faktisk dekker.

GRUNNEN ER UVERIFISERBARHET. Operatoer ratifiserte sletting fordi linja er
umaalbar — ingen generasjonslogg finnes. Det gjelder ALLE 23 linjer, ikke bare
de 8 som ogsaa feiler paa egen aritmetikk. Inkonsistensen var EVIDENSEN for at
tallet aldri ble transkribert fra en kjoering; den var aldri GRUNNEN til aa
slette det.

Men mitt eget spoersmaals opsjonstekst definerte «beroert fil» som den
inkonsistente mengden, saa bokfoeringen stanset paa 7. Ingen maskin ser
framingen av et spoersmaal — samme klasse som §9.13s falske premiss, denne
gangen i et spoersmaal jeg selv skrev.

KONSEKVENS, UTTALT I STEDET FOR IMPLISERT: 14 filer baerer den fordoemte linja
uten koe-entry (10 konsistente + 4 ikke sjekkbare). De er IKKE ekskludert paa
merit — de staar ubokfoert fordi aa utvide forbi de ratifiserte 7 er
operatoerens beslutning, ikke oekten sin. Reist som aapent spoersmaal #17.
Begge listene enumerert ordrett i resolutionen, saa de er greppbare.

TO TELLINGER SOM MIN EGEN EDIT UGYLDIGGJORDE, rettet i samme pass:
- Footer-populasjonen er 21 linjer naa, ikke 23 — jeg slettet to, begge fra den
  inkonsistente boetta: 10 · 6 · 1 · 4.
- STATEs «binding» **Status:**-telling var ALT stale foer denne oekten. Den sa
  GA 249 / 69 distinkte, men ble maalt inne i 26g-oekten FOER 96639f7 ble
  anvendt. Ground truth: 387 forekomster, 70 distinkte, GA 247. Kjeden er
  eksakt: 249 -> 248 (26g) -> 247 (088cf06).

ANKERUNIKHET for idx-26s ble verifisert ETTERPAA, ikke foer — navngitt som et
sluntreri, ikke stille rettet. Gaten beviser TILSTEDEVAERELSE (text.includes),
aldri unikhet, saa et ikke-unikt anker passerer lydloest. Alle tre: 1 forekomst.

Suite: 1047/1047. Koe: 14 aapne / 13 resolved.

docs/r11-pilot-results.md

@@ -1679,8 +1679,8 @@ asserted over.
   `check-g7-queue.mjs`, which is the `795d494` failure repeated at class scale.
 - **`ai-red-team-operations-practical.md:87`** — the corpus's only other `model- og`. Left
   untouched; the dialect fix was scoped to the one file.
-- **Untouched, on precedent:** `**Last updated:**` — now twelve ratified corpus edits, none of
-  which moved that field.
+- **Untouched, on precedent:** `**Last updated:**` — twelve *prior* ratified corpus edits, none of
+  which moved that field; this one makes thirteen.
 
 #### No re-dating decision was required — and that was adjudicated before writing
 
@@ -1694,3 +1694,37 @@ under no stamp at all; and line 101 sits under the section-3 stamp at 131, which
 entry says so itself. An orthographic alignment neither falsifies it nor requires re-dating it.
 
 **Queue:** 14 open / 13 resolved. **Suite:** 1047/1047.
+
+#### Scope addendum — the ground is unverifiability, not arithmetic
+
+Written the same day, because the section above understated what the ratified ground covers. The
+operator ratified deletion because the line is **unmeasurable** — no generation log exists to
+adjudicate it. That property holds for all 23 lines, not only the 8 that also fail their own
+arithmetic. Internal inconsistency was the *evidence* that the number was never transcribed from a
+run; it was never the *reason* for deleting it.
+
+But the ratification question's own option text defined "affected file" as the inconsistent set, so
+booking stopped at 7. **No machine sees the framing of a question** — this is the same class as
+§9.13's false premise, this time in a question written here rather than inherited.
+
+So **14 files carry the condemned line with no queue entry**: 10 internally consistent, 4 not
+checkable (a total with no per-tool numbers, so the arithmetic test cannot be applied — but the
+total is just as unverifiable). They are not excluded on merit. They are unbooked because expanding
+past the ratified 7 is a scope decision belonging to the operator, and it is raised as open question
+**#17** rather than taken here. Both lists are enumerated in idx-26j's resolution so they are
+greppable rather than prose.
+
+#### Two counts this edit invalidated
+
+- **The footer population is now 21 lines, not 23.** Two were deleted here, both from the
+  inconsistent bucket: **10 consistent · 6 inconsistent · 1 ambiguous · 4 not checkable.**
+- **The binding `**Status:**` header counts were already stale before this session.** They read
+  `GA` 249 over 387 occurrences / 69 distinct — but that measurement was taken inside the idx-26g
+  session *before* `96639f7` was applied. Ground truth now: **387 occurrences, 70 distinct, `GA`
+  247**, and `GA / Preview (Responsible AI scorecard)` at 2. The chain is exact: 249 → 248 (26g) →
+  247 (this edit); distinct rose 69 → 70 when 26g first created that string, then held, because this
+  edit added a second instance of a value that already existed.
+
+**Anchor uniqueness for idx-26s was verified after it was written, not before** — named here rather
+than quietly fixed. The gate proves *presence* (`text.includes`), never uniqueness, so a non-unique
+anchor passes silently. All three occur exactly once.