fix(ms-ai-architect): 58-klassen er lukket — og den siste keep-linja var usann

De tre siste medlemmene (idx-26ab, idx-26ae, idx-26ah) baerte hver sitt flagg
som holdt dem utenfor klasse-entryen. Ingen av flaggene overlevde maalingen som
den saken de var bokfoert som.

MAALT, IKKE ARVET: 58 av 58 medlemmer i r11-footer-class-2026-08-11.json har
naa ingen ordrett ankerlinje igjen i korpus. Klassen er lukket.

idx-26ab (prose_repeat). Entryens egen note sa at andre lokator «is a rewrite,
not a line deletion». Det ble falsifisert ved aa skrive ut resultatstrengen:
sletting av de 16 sammenhengende tegnene «12 MCP-kall til » gir en grammatisk
norsk setning som fila selv baerer (8 nummererte kilder, 20 kodeblokker), og
forfatter null ord. Argumentet sa rewrite; strengen sa sletting.
=> RATIFISERT #25: en delete-klasse-telling gjentatt INNE i en linje (hodefelt
eller loepende prosa) lukkes ved aa slette det sammenhengende fragmentet.
Tre vilkaar som ALLE maales: resten er grammatisk, resten er grunnet i fila,
null ord forfattes. Holder ett av dem ikke, kommer saken tilbake som spoersmaal.
Widening av #24 fra den ene parentesen den ble skrevet for.

idx-26ae (label_value_mismatch). Spoersmaalet — lukkes en feilmerket linje av aa
slettes, naar det er etiketten og ikke tallet som villeder — trengte ingen ny
form. Linja har INGEN keep-verdi (3 docs_search + 2 docs_fetch, og dens eget
resultat sier «= 5 MCP-kall»), saa den er ikke en #23-blandet linje; slettingen
tar etiketten med seg; og den eneste alternative reparasjonen, aa doepe om
«MCP-kilder» til «MCP-kall», er noeyaktig det #22 forbyr.

idx-26ah (label_value_mismatch). Den smale editen var tilgjengelig og ble
forkastet med grunn: :653 er eneste medlem innenfor de 58, men sletting av den
alene ville latt :651 (naar genereringen kjoerte) og :652 (hvilket verktoey)
staa rett over editen med samme referent. #22s tidsstempel-klausul sier INSIDE
THE BLOCK, og her finnes ingen merkelapp — den rekker ikke, og kunne ikke
strekkes uten den stille utvidelsen #22 selv nektet.
=> RATIFISERT #26, SCOPET TIL ÉN FIL: en avsluttende umerket rekke der HVER
linje feiler referent-testen slettes hel. Den generelle klassen staar fortsatt
aapen paa idx-26ar.

NY DEFEKT FUNNET VED AA MAALE NABOEN FOER DEN BLE STOLT PAA (#21/#23-plikten):
rag-document-preprocessings :792 «8 Microsoft Learn-artikler + 4 GitHub-repos
= 12 kilder» er keep-klasse og to av tre deler er usanne. Maalt over to
populasjoner som er enige (Kilder-seksjonen og hele fila): 8 distinkte
learn.microsoft.com-artikler stemmer, men det er 2 distinkte github.com-URLer,
ikke 4, og dermed 10 navngitte kilde-URLer, ikke 12 — eller 13 om de tre
pris-URLene teller som kilder, en lesning linja ikke oppgir. Ingen lesning gir
12. Aritmetikken stemmer bare fordi GitHub-halvdelen er feil.
BOKFOERT SOM idx-26at, IKKE REPARERT: aa rette 4 til 2 forfatter tall inn i
korpus (idx-26v-faren), og ingen ratifisert form dekker en keep-telling som
bare er gal — #24 slettet en slik telling kun fordi BEGGE halvdeler feilet.
Beslutningen operatoeren skylder entryen er om dette programmet faar skrive et
tall det har maalt.

Koe: 54 entries (13 aapne, 41 resolved). Suite 1052/1052. Alle _meta-pekere
sjekket for haand mot noekkelsettet — ingenting validerer _meta.

De 5 utrackede maaledatafilene i scripts/kb-eval/data/ er tracket etter
operatoerbeslutning (spurt 2026-08-11, avgjort 2026-08-12), paa linje med
r11-footer-class-2026-08-11.json.

scripts/kb-eval/data/g7-review-queue.json

@@ -11,7 +11,9 @@
       "#21": "PROVENANCE QUALIFIER ON A KEEP-CLASS LINE. Ratified 2026-08-11.\nSCOPE FIRST, because the question was raised over four entries and only ONE survived measurement. idx-26y, idx-26af and idx-26ag do NOT fall under this form: in each, the anchored line is itself a call count or a call enumeration, which the idx-26j form already deletes, so the provenance overclaim dies WITH the deletion and needs no separate decision. Measured before the scope was narrowed rather than after: microsoft-graph-api-copilot-integration.md carries 13 distinct URLs and ALL 13 are on learn.microsoft.com, so its keep-class neighbour \"**Total kilder:** 11 MCP-verified URLs\" carries no overclaim at all; the keep-class neighbours in ai-act-annex-iii-checklist.md and ai-act-compliance-guide.md (\"**Total sources referenced:** 8\" and \"16\") carry no provenance qualifier in the first place.\nTHE FORM: where a KEEP-class line (a count of the artifact's own sources or URLs) carries a provenance qualifier that overclaims, DELETE THE QUALIFIER AND KEEP THE COUNT. Do NOT name the actual provenance mix.\nTHE GROUND is idx-26j's ground: naming the mix requires asserting which tool produced which URL, and no artifact in this repo records that. Rewriting \"fra MCP-research\" as \"5 fra MCP-research, 2 fra web\" would replace an unverifiable qualifier with a MORE specific unverifiable claim - the idx-26v hazard, where the repair authors the defect class it closes. Deleting the qualifier removes tokens and adds none, leaving only what the file itself can adjudicate.\nGOVERNS: idx-26w.",
       "#22": "WHAT \"DELETE THE LINE\" MEANS FOR BLOCK FORM. Ratified 2026-08-11. The widest form ratified since idx-26j; it governs 14 entries.\nTHE REFERENT TEST, applied line by line inside the block:\n - Referent is THE GENERATION (calls made, calls enumerated, tools used, when the document was generated) -> DELETE. Nothing can adjudicate it. Re-verified 2026-08-11 rather than inherited: the only files under scripts/, docs/, tests/ and hooks/ that mention MCP calls are this queue, r11-footer-class-2026-08-11.json, docs/r11-pilot-results.md and docs/ref-kb-direction-note-2026-06.md - all analysis, none a run record.\n - Referent is THE ARTIFACT (a count of sources, URLs or documents the file itself lists) -> KEEP. It is countable against the file.\nTHE LABEL RULE, which is what the question actually turned on. The label or heading above the block is ITSELF a member of the delete class when its own words name the generation - \"MCP Calls: 6\", \"Total MCP Calls: 4\", \"MCP-kall utfoert\", \"MCP-statistikk\", \"MCP-research metadata\", \"MCP research summary\", \"MCP-soek utfoert\", \"MCP Calls Summary\", \"Verifiseringsmetode\". Delete it together with the lines it labels, and PROMOTE any surviving keep-class line to the enclosing level. The label is KEPT UNCHANGED when its words name content or coverage rather than process - \"Research Coverage\" (idx-26am) is the only such label in the 14.\nNEVER RENAME A LABEL. Renaming authors a new claim (idx-26v). This is not a side condition, it is what makes the form safe - and it is why idx-26aq needs no decision about its parenthetical: \"### MCP-soek (7 unique sources)\" is FALSE read as sources (the block above enumerates 10) and true only read as calls, but the heading is DELETED rather than renamed, so the reading never has to be chosen.\nGENERATION TIMESTAMPS INSIDE THE BLOCK are delete class under the same test (\"**Dato for research:** 2026-06-19\", \"**Research timestamp:** April 2026\", \"**Dato generert:** 2026-02-04\"). One-step consequence of idx-26j's referent, not a new form - the reasoning idx-26u and idx-26ac each recorded and neither applied.\nEXPLICITLY NOT COVERED, and this is a scope decision rather than a judgement on merit: idx-26ar (zero-etl-fabric-patterns.md) carries the enumeration with NO count and sits outside the 58 this question was scoped over. It was put to the operator as requiring an explicit decision rather than silent inheritance, and is left open on that ground - the discipline idx-26j's scope addendum named after booking stopped at 7.\nGOVERNS: idx-26x, idx-26z, idx-26aa, idx-26ac, idx-26ad, idx-26af, idx-26ag, idx-26ai, idx-26aj, idx-26ak, idx-26al, idx-26am, idx-26ao, idx-26aq.",
       "#23": "A LINE THAT MIXES THE DELETE CLASS AND THE KEEP CLASS. Ratified 2026-08-11.\nTHE FORM: do NOT split the line. MEASURE the keep-class half against the file first, then DELETE THE WHOLE LINE.\nWHY THAT LOSES NOTHING - measured in both members rather than argued:\n - idx-26ao, genaiops-llm-specific-practices.md:383. \"18 kilder\" is TRUE (18 numbered sources) and restated TWICE more, at :6 in the header block and at :356 in the heading \"### Microsoft Learn-kilder (18 dokumenter)\". Deletion removes nothing the file does not still say.\n - idx-26z, chain-of-thought-prompting.md:500. \"3 unike Microsoft Learn-kilder\" is FALSE - the file's own \"### Microsoft Learn (Verified via MCP)\" table has 5 rows and 5 distinct Learn URLs. \"4 MCP-kall\" is also wrong against 3 enumerated. BOTH numbers in that line are wrong, so splitting it would have preserved a falsehood.\nTHE POINT: the hazard was that deletion takes a true claim with it. MEASUREMENT DISSOLVES THE HAZARD instead of trading it for authoring. Where the keep half is redundant, deletion is free; where it is false, deletion is a correction.\nNOT RATIFIED, BECAUSE NOT INSTANTIATED: the residual case where a keep half is BOTH true AND unique in its file. Neither member is that case. A future member that is must come back as a new question rather than inherit this form.\nGOVERNS: idx-26z, idx-26ao. idx-26ao ALSO falls under #22 - the heading and the three call bullets above its line go in the SAME edit, or the deletion produces the idx-26aq residue.",
-      "#24": "A DELETE-CLASS COUNT RESTATED IN THE DOCUMENT HEADER BLOCK. Ratified 2026-08-11.\nTHE FORM: delete the whole parenthetical \"(8 MCP-oppslag, 16 kilder)\" at mlops-security-access-control.md:5, leaving \"**Confidence:** HIGH - Basert paa offisiell Microsoft Learn dokumentasjon\".\nTHE MEASUREMENT THAT TURNED A RE-AUTHORING INTO A DELETION: the header was treated as mixed - one delete-class half, one keep-class half - which would have forced a rewrite inside the reference contract's structured header. It is NOT mixed. \"16 kilder\" is FALSE: the file enumerates 8 numbered sources and carries 10 distinct URLs, 9 of them on learn.microsoft.com. No reading yields 16. \"8 MCP-oppslag\" is unverifiable on the idx-26j ground. BOTH halves fail, so there is nothing to preserve and no replacement to write. What remains asserts no count and is grounded.\nZERO WORDS AUTHORED - the header keeps its field name and its claim; only the false and the unverifiable are removed.\nGOVERNS: idx-26ap, whose second locator (the footer \"**MCP Calls:** 8\" at :744) is deleted by the idx-26j form in the same edit."
+      "#24": "A DELETE-CLASS COUNT RESTATED IN THE DOCUMENT HEADER BLOCK. Ratified 2026-08-11.\nTHE FORM: delete the whole parenthetical \"(8 MCP-oppslag, 16 kilder)\" at mlops-security-access-control.md:5, leaving \"**Confidence:** HIGH - Basert paa offisiell Microsoft Learn dokumentasjon\".\nTHE MEASUREMENT THAT TURNED A RE-AUTHORING INTO A DELETION: the header was treated as mixed - one delete-class half, one keep-class half - which would have forced a rewrite inside the reference contract's structured header. It is NOT mixed. \"16 kilder\" is FALSE: the file enumerates 8 numbered sources and carries 10 distinct URLs, 9 of them on learn.microsoft.com. No reading yields 16. \"8 MCP-oppslag\" is unverifiable on the idx-26j ground. BOTH halves fail, so there is nothing to preserve and no replacement to write. What remains asserts no count and is grounded.\nZERO WORDS AUTHORED - the header keeps its field name and its claim; only the false and the unverifiable are removed.\nGOVERNS: idx-26ap, whose second locator (the footer \"**MCP Calls:** 8\" at :744) is deleted by the idx-26j form in the same edit.",
+      "#25": "A DELETE-CLASS COUNT RESTATED INSIDE A LINE. Ratified 2026-08-12. Widens #24 from the single header parenthetical it was written for to any in-line restatement - header field OR running prose.\nTHE FORM: delete the CONTIGUOUS delete-class fragment and leave the rest byte-identical. Three conditions, ALL MEASURED rather than assumed: (1) the remainder is grammatical, (2) the remainder is grounded in what the file itself lists, (3) zero words are authored.\nWHY IT IS NOT A REWRITE - which is what idx-26ab's own booking called it. At inferencing-optimization-caching.md:1029 the fragment is \"12 MCP-kall til \", 16 contiguous characters. Removing them leaves \"**Confidence niva: HIGH** - Denne referansen er basert pa offisiell Microsoft-dokumentasjon og kodeeksempler.\", a grammatical Norwegian sentence whose claim the file's own 8 numbered sources and 20 fenced code blocks carry. Nothing is composed. The booking's premise was falsified by writing the resulting string out and reading it, not by argument - the same move that turned #24 from a re-authoring into a deletion.\nTHE LIMIT, which is what keeps this from swallowing the corpus: where the three conditions do not ALL hold, this form does not apply and the case comes back as a new question. A remainder that needs one word added is authoring, and authoring is the idx-26v hazard.\nGOVERNS: idx-26ab.",
+      "#26": "AN UNLABELLED TRAILING FOOTER RUN. Ratified 2026-08-12, SCOPED TO ONE FILE.\nSCOPE FIRST, because the scope is the whole decision: this governs ai-impact-assessment-framework.md:651-653 and nothing else. #22's timestamp clause says INSIDE THE BLOCK, and its precedent (idx-26ac, applied 2026-08-12) deleted a timestamp that sat under the label \"**MCP-research metadata:**\". This run carries no heading and no label above it, so #22 does not reach it and could not be made to without the silent widening #22 itself refused.\nTHE FORM: where a file ends in a run of sibling \"**Label:** value\" lines with no heading and no label above them, and EVERY line in the run fails idx-26j's referent test, the run is deleted whole.\nMEASURED LINE BY LINE, not inherited: \":651 **Sist oppdatert via MCP:** 2026-02-04\" is when the generation ran; \":652 **MCP-kilder brukt:** microsoft-learn (docs.microsoft.com, microsoft.com/ai)\" is which tool was used and carries no number at all; \":653 **Antall dokumenter sokt:** 4 (search queries) + 2 (deep fetch)\" puts a document label over values that count operations - a search query is not a document. All three referents are the generation. Nothing countable against the file is in the run: the file's sources are the numbered lists above, untouched, as is the \"**Confidence markers:**\" block.\nWHY THE RUN AND NOT THE COUNTING LINE ALONE: :653 is the only member inside the 58 counting class, so the narrow edit was available. It would have left two lines with the same referent immediately above the edit - the residue class G7 exists for, caught at the edit site this time rather than by a later sweep.\nWHAT STAYS OPEN: the general class - an unlabelled run, and an enumeration with no count - remains open at idx-26ar. This form is an instance, not the class, and a second instance must come back as a question rather than inherit it.\nGOVERNS: idx-26ah."
     }
   },
   "entries": [
@@ -483,14 +485,15 @@
       "id": "idx-26ab",
       "file": "skills/ms-ai-engineering/references/mlops-genaiops/inferencing-optimization-caching.md",
       "class": "multi-locator",
-      "status": "open",
+      "status": "resolved",
       "raised": "2026-08-11",
       "summary": "Booked 2026-08-11 from the hand-classified footer class measured in docs sec. 9.16: 58 members in 58 files, dataset scripts/kb-eval/data/r11-footer-class-2026-08-11.json (tracked). Operator decision #17 that day split the class - the 42 line-form unflagged members take one class-wide record, and the 16 that carry a decision the ratified form does not settle get their own entry. This is one of those 16. Reason: flag prose_repeat. Bucket konsistent - MEASURED at :1005, \"**Total MCP-kall:** 7 (docs search) + 3 (docs fetch) + 2 (code samples) = **12**\", and 7+3+2 does equal 12. CONSISTENCY IS NOT A REASON TO KEEP IT: the ground for deletion is unverifiability, not arithmetic, so this line goes for the same reason the inconsistent ones do. WHY ITS OWN ENTRY: the same 12 is restated in PROSE elsewhere in the file, VERIFIED at :1029 - \"**Confidence nivå: HIGH** — Denne referansen er basert på 12 MCP-kall til offisiell Microsoft-dokumentasjon og kodeeksempler.\" Deleting the footer line alone leaves the number alive in the same file. That is the residue class G7 was built for, and it is the sec. 9.16 method finding \"a deletion can leave the number alive somewhere else in the same file\" with a measured instance. Both locators are anchored so neither can be edited around. THE RATIFIED FORM (idx-26j, 2026-08-09): current provenance, one referent for the whole footer block, and the unverifiable MCP-calls line is DELETED rather than corrected to the sum. The ground is UNVERIFIABILITY, not arithmetic - the consistency bucket governs BOOKKEEPING, never whether something is deleted. NOTE the prose locator is not a bare count - removing \"12 MCP-kall\" from :1029 leaves a confidence claim that needs a grammatical sentence, so this one is a rewrite at the second locator, not a line deletion. Neighbour :1007 \"**Kilder totalt:** 8 Microsoft Learn-artikler + 15+ kodeeksempler\" is KEEP class and is not part of this defect.",
       "evidence": "skills/ms-ai-engineering/references/mlops-genaiops/inferencing-optimization-caching.md:1005 (footer total) and :1029 (prose repeat of the same 12); dataset r11-footer-class-2026-08-11.json (form=line, b=konsistent, flags=[prose_repeat]); docs sec. 9.16 (rest-search-per-file requirement)",
       "anchors": [
         "**Total MCP-kall:** 7 (docs search) + 3 (docs fetch) + 2 (code samples) = **12**",
         "**Confidence nivå: HIGH** — Denne referansen er basert på 12 MCP-kall til offisiell Microsoft-dokumentasjon og kodeeksempler."
-      ]
+      ],
+      "resolution": "APPLIED 2026-08-12, one edit, two locators: :1005 under the idx-26j form and :1029 under #25.\n\"**Total MCP-kall:** 7 (docs search) + 3 (docs fetch) + 2 (code samples) = **12**\" is deleted with its trailing blank line. At :1029 the contiguous fragment \"12 MCP-kall til \" is deleted, leaving \"**Confidence niva: HIGH** - Denne referansen er basert pa offisiell Microsoft-dokumentasjon og kodeeksempler.\"\nTHIS BOOKING'S OWN NOTE WAS FALSIFIED. It said the second locator \"needs a grammatical sentence, so this one is a rewrite\". Written out, the edit is a 16-character contiguous substring deletion that authors zero words - which is what #25 was ratified on. The note was an argument; the string was a measurement.\nRESIDUE SWEPT WHOLE-FILE: the digit 12 with this referent occurred at exactly these two locators and now occurs at neither. Keep-class neighbour :1007 was measured before it was trusted - 8 numbered sources OK, \"15+ kodeeksempler\" against 20 fenced code blocks OK - and left untouched. NOT REPAIRED AND NOT BOOKED, named so it is not mistaken for unseen: the provenance qualifiers on the content headings at :997 \"### Code Samples (MCP microsoft-learn)\" and :1032 \"- Verified (MCP 2026-04)\" are #21-shaped but carry no count, so they fall outside both the 58 and #21's stated scope.\nVerified by whole-line comparison: both pre-edit lines occur 0x, the post-edit line exactly 1x."
     },
     {
       "id": "idx-26ac",
@@ -530,14 +533,15 @@
       "id": "idx-26ae",
       "file": "skills/ms-ai-engineering/references/rag-architecture/rag-document-preprocessing.md",
       "class": "replacement",
-      "status": "open",
+      "status": "resolved",
       "raised": "2026-08-11",
       "summary": "Booked 2026-08-11 from the hand-classified footer class measured in docs sec. 9.16: 58 members in 58 files, dataset scripts/kb-eval/data/r11-footer-class-2026-08-11.json (tracked). Operator decision #17 that day split the class - the 42 line-form unflagged members take one class-wide record, and the 16 that carry a decision the ratified form does not settle get their own entry. This is one of those 16. Reason: flag label_value_mismatch. Bucket konsistent. MEASURED at :791: \"**Totalt antall MCP-kilder:** 3 docs_search calls + 2 docs_fetch calls = **5 MCP-kall**\", and 3+2 does equal 5. THE DEFECT IS IN THE LABEL, NOT THE NUMBER: the label promises a count of SOURCES (\"MCP-kilder\") while every value in the line counts CALLS, and the line says so itself in its own result (\"= **5 MCP-kall**\"). It is the mirror image of idx-26w, where the number is right and the provenance qualifier overclaims. A reader who trusts the label reads this file as having 5 sources; the file's actual source count is the NEXT line. THE RATIFIED FORM (idx-26j, 2026-08-09): current provenance, one referent for the whole footer block, and the unverifiable MCP-calls line is DELETED rather than corrected to the sum. The ground is UNVERIFIABILITY, not arithmetic - the consistency bucket governs BOOKKEEPING, never whether something is deleted. The call count therefore goes under the ratified form. What is NOT settled is whether a mislabelled line is closed by deleting it, given that the label - not the count - is what misleads. Neighbour :792 \"**Totalt antall unike URLer:** 8 Microsoft Learn-artikler + 4 GitHub-repos = **12 kilder**\" is KEEP class and anchored so it cannot be taken along. SEPARATELY MEASURED AND DELIBERATELY NOT BOOKED HERE: :793 \"**Konfidensnivå totalt:** 95% Verified (fra MCP), 5% Baseline (norske forhold og priser)\" is the idx-26j locator-4 / idx-26v class - a percentage with no stated denominator - and it is NOT in the footer counting class, so booking it would widen decision #17 unasked. It is named here and raised to the operator rather than left to live only in prose.",
       "evidence": "skills/ms-ai-engineering/references/rag-architecture/rag-document-preprocessing.md:791-793 (file ends at :805); dataset r11-footer-class-2026-08-11.json (form=line, b=konsistent, flags=[label_value_mismatch]); idx-26w (mirror case); idx-26v (the percentage class named at :793)",
       "anchors": [
         "**Totalt antall MCP-kilder:** 3 docs_search calls + 2 docs_fetch calls = **5 MCP-kall**",
         "**Totalt antall unike URLer:** 8 Microsoft Learn-artikler + 4 GitHub-repos = **12 kilder**"
-      ]
+      ],
+      "resolution": "APPLIED 2026-08-12 under the idx-26j form, with #22's NEVER RENAME A LABEL settling the question this booking left open. :791 is deleted whole.\nTHE OPEN QUESTION WAS whether a mislabelled line is closed by deleting it, given that the label rather than the count is what misleads. It DISSOLVES IN MEASUREMENT instead of needing a new form, which is #23's move applied to a different shape: the line carries NO keep-class value at all (3 docs_search + 2 docs_fetch, and its own result says \"= **5 MCP-kall**\"), so it is not a #23 mixed line and there is nothing to preserve; the deletion takes the misleading label away with it; and the only alternative repair - renaming \"MCP-kilder\" to \"MCP-kall\" - is precisely what #22 forbids. What the label promised, a count of sources, the next line delivers.\nNEW DEFECT FOUND AT THAT NEXT LINE AND BOOKED RATHER THAN REPAIRED: measuring :792 before relying on it (the #21 and #23 discipline) showed two of its three parts false. Booked as idx-26at.\nVerified by whole-line comparison: the pre-edit line occurs 0x; :792 and :793 still occur 1x each."
     },
     {
       "id": "idx-26af",
@@ -577,13 +581,14 @@
       "id": "idx-26ah",
       "file": "skills/ms-ai-governance/references/responsible-ai/ai-impact-assessment-framework.md",
       "class": "replacement",
-      "status": "open",
+      "status": "resolved",
       "raised": "2026-08-11",
       "summary": "Booked 2026-08-11 from the hand-classified footer class measured in docs sec. 9.16: 58 members in 58 files, dataset scripts/kb-eval/data/r11-footer-class-2026-08-11.json (tracked). Operator decision #17 that day split the class - the 42 line-form unflagged members take one class-wide record, and the 16 that carry a decision the ratified form does not settle get their own entry. This is one of those 16. Reason: flag label_value_mismatch. Bucket ikke_sjekkbar - no total. MEASURED at :653: \"**Antall dokumenter søkt:** 4 (search queries) + 2 (deep fetch)\". THE LABEL COUNTS DOCUMENTS; THE VALUES COUNT OPERATIONS. A search query is not a document - one query can return ten documents or none - so the line's own parenthetical falsifies its own label. Same class as idx-26ae, on the opposite pairing: there a source label sat over call values, here a document label sits over query values. This file is also one of the fourth-escape-form instances in sec. 9.16 point 2: the count rides on the action word (\"dokumenter søkt\"), with no \"kall\" and no tool identifier beside a digit. THE RATIFIED FORM (idx-26j, 2026-08-09): current provenance, one referent for the whole footer block, and the unverifiable MCP-calls line is DELETED rather than corrected to the sum. The ground is UNVERIFIABILITY, not arithmetic - the consistency bucket governs BOOKKEEPING, never whether something is deleted. Neighbour :651 \"**Sist oppdatert via MCP:** 2026-02-04\" raises the generation-timestamp category question booked at idx-26ac and idx-26u, and :652 \"**MCP-kilder brukt:** microsoft-learn (docs.microsoft.com, microsoft.com/ai)\" carries no number and so falls outside the counting class. Neither is repaired here.",
       "evidence": "skills/ms-ai-governance/references/responsible-ai/ai-impact-assessment-framework.md:651-653 (file ends at :654); dataset r11-footer-class-2026-08-11.json (form=line, b=ikke_sjekkbar, flags=[label_value_mismatch]); docs sec. 9.16 point 2; idx-26ae (mirror pairing); idx-26ac, idx-26u (timestamp category)",
       "anchors": [
         "**Antall dokumenter søkt:** 4 (search queries) + 2 (deep fetch)"
-      ]
+      ],
+      "resolution": "APPLIED 2026-08-12 under #26. :651-653 are deleted whole together with the blank line above them; the file now ends at the \"**Confidence markers:**\" block, at 650 lines instead of 653.\nTHE NARROW EDIT WAS AVAILABLE AND WAS REJECTED WITH A REASON: :653 is the only member of the run inside the 58 counting class, so deleting it alone would have been the scope-faithful move - and would have left \"**Sist oppdatert via MCP:** 2026-02-04\" and \"**MCP-kilder brukt:** microsoft-learn (...)\" standing immediately above the edit, both with the same generation referent. The widening was put to the operator as an explicit ratification rather than inherited from #22, whose timestamp clause is scoped INSIDE THE BLOCK and does not reach an unlabelled run. Ratified file-scoped; the general class stays open at idx-26ar.\nVerified by whole-line comparison: all three pre-edit lines occur 0x and the preceding bullet occurs 1x."
     },
     {
       "id": "idx-26ai",
@@ -937,6 +942,18 @@
           "text": "- **Metode:** MCP microsoft-learn server"
         }
       ]
+    },
+    {
+      "id": "idx-26at",
+      "file": "skills/ms-ai-engineering/references/rag-architecture/rag-document-preprocessing.md",
+      "class": "replacement",
+      "status": "open",
+      "raised": "2026-08-12",
+      "summary": "Raised 2026-08-12 while applying idx-26ae, out of the duty to measure a keep-class neighbour BEFORE relying on it (the #21 and #23 discipline). :792 \"**Totalt antall unike URLer:** 8 Microsoft Learn-artikler + 4 GitHub-repos = **12 kilder**\" is keep class - every part of it is countable against the file - and TWO OF ITS THREE PARTS ARE FALSE.\nMEASURED over two populations that agree, the \"Kilder og verifisering\" section (:748-793) and the whole file: 8 distinct learn.microsoft.com articles, TRUE (the section's two tables carry 6 + 2; the body's one extra occurrence is the same search-indexer-overview article without /en-us/, so it is not a ninth). 2 distinct github.com URLs, NOT 4 - Azure/azure-sdk-for-python and microsoft/Form-Recognizer-Toolkit. Therefore 10 named source URLs, not 12; or 13 if the three azure.microsoft.com pricing URLs at :783-785 count as kilder, a reading the line does not state. NO READING YIELDS 12, and the arithmetic is internally consistent only because the GitHub half is wrong.\nWHY IT IS NOT REPAIRED HERE: correcting 4 to 2 and 12 to 10 authors numbers into the corpus - the idx-26v hazard - and no ratified form covers a keep-class count that is simply wrong. #24 deleted such a count only because BOTH halves failed and there was nothing left to preserve; here \"8 Microsoft Learn-artikler\" is true, so #24's ground does not transfer, and #23 does not reach it because this is not a mixed delete/keep line. The line survived the idx-26ae edit intact and anchored, exactly as that entry required.\nTHE DECISION OWED: delete the whole line, correct the two false parts, or delete only the false parts and keep the true count. Each is a different answer to whether this programme may write a number it has measured, which has not been settled and should not be settled silently by whichever session gets here first.\nSEPARATELY, still not booked and named so it is not mistaken for unseen: :793 \"**Konfidensniva totalt:** 95% Verified (fra MCP), 5% Baseline (norske forhold og priser)\" is the idx-26j locator-4 / idx-26v class - a percentage with no stated denominator - already named in idx-26ae and deliberately left where it was.",
+      "evidence": "skills/ms-ai-engineering/references/rag-architecture/rag-document-preprocessing.md:792 (kept when :791 was deleted under idx-26ae, 2026-08-12); URL census of the file: 8 distinct learn.microsoft.com, 2 distinct github.com, 3 azure.microsoft.com pricing URLs; source tables at :750-777; idx-26ae (the application that surfaced it); #21 and #23 (measure-the-keep-half discipline); #24 (why deletion does not transfer); idx-26v (the authoring hazard)",
+      "anchors": [
+        "**Totalt antall unike URLer:** 8 Microsoft Learn-artikler + 4 GitHub-repos = **12 kilder**"
+      ]
     }
   ]
 }

skills/ms-ai-engineering/references/mlops-genaiops/inferencing-optimization-caching.md

@@ -1002,8 +1002,6 @@ Diagnostikk:
 - Autoscaling configuration (YAML)
 - Databricks disk cache configuration
 
-**Total MCP-kall:** 7 (docs search) + 3 (docs fetch) + 2 (code samples) = **12**
-
 **Kilder totalt:** 8 Microsoft Learn-artikler + 15+ kodeeksempler
 
 ---
@@ -1026,7 +1024,7 @@ Diagnostikk:
 - Deploy **ONNX models til edge** (Azure SQL Edge, Windows ML) for low-latency/privacy-kritiske scenarios
 - Monitor **cache hit rate** og **autoscaling metrics** kontinuerlig
 
-**Confidence nivå: HIGH** — Denne referansen er basert på 12 MCP-kall til offisiell Microsoft-dokumentasjon og kodeeksempler.
+**Confidence nivå: HIGH** — Denne referansen er basert på offisiell Microsoft-dokumentasjon og kodeeksempler.
 
 
 ### ONNX Inferencing Optimization for Computer Vision (Azure ML AutoML 2026) — Verified (MCP 2026-04)

skills/ms-ai-engineering/references/rag-architecture/rag-document-preprocessing.md

@@ -788,7 +788,6 @@ Images (JPEG/PNG)?
 
 ---
 
-**Totalt antall MCP-kilder:** 3 docs_search calls + 2 docs_fetch calls = **5 MCP-kall**
 **Totalt antall unike URLer:** 8 Microsoft Learn-artikler + 4 GitHub-repos = **12 kilder**
 **Konfidensnivå totalt:** 95% Verified (fra MCP), 5% Baseline (norske forhold og priser)
 

skills/ms-ai-governance/references/responsible-ai/ai-impact-assessment-framework.md

@@ -647,7 +647,3 @@ Purview SDK-integrasjon gir:
 - **Verified:** Informasjon hentet direkte fra Microsoft MCP-kilder (microsoft-learn)
 - **Baseline:** Informasjon basert på modellens treningsdata (pre-Jan 2025), men validert mot kjente Microsoft-rammeverk
 - **Inferred:** Logiske utledninger basert på verified sources, markert eksplisitt der det brukes
-
-**Sist oppdatert via MCP:** 2026-02-04
-**MCP-kilder brukt:** microsoft-learn (docs.microsoft.com, microsoft.com/ai)
-**Antall dokumenter søkt:** 4 (search queries) + 2 (deep fetch)