open/ms-ai-architect
1
0
Fork 0
Code Issues Releases 1 Activity

docs(ms-ai-architect): G7 — 16 individuelle footer-entries bokfoert; klasse-entryen for de 42 er IKKE uttrykkbar i skjemaet

Browse source 
Beslutning #17 sa 1 klasse-entry (42 medlemmer) + 16 individuelle = 17.
16 av de 17 er skrevet. Den 17. kan ikke skrives, og grunnen er maalt.

BLOKKEREN: validateQueue leser ÉN entry.file og sjekker ALLE entry.anchors mot
den ene filas tekst (lib/g7-queue.mjs:83-89). Kjoert empirisk mot den ekte
validatoren med en hypotetisk klasse-entry over de 42: ok=false, ett
anchor_drift-funn med 41 av 42 ankere. Ikke utledet fra kodelesning — maalt.

DETTE ER GJENOPPDAGELSEN AV GRUNNEN TIL BESLUTNING #14. idx-26q-s egen summary
sier det ordrett: skjemaet baerer én file per entry, og en klassevid entry
ville gjort filene usynlige for check-g7-queue.mjs — 795d494-feilen, «en
lokator som bare lever i prosa mens begge gatene gaar groenne». #17 overstyrte
#14 uten aa baere den grunnen med seg. Begge grunnene er ekte og maalte; de
kolliderer.

DE 16: idx-26x, idx-26y, idx-26z, idx-26aa..idx-26am. Familien fortsetter
26-serien fordi dette ER samme defektfamilie som 26j/26k/26q; 26a har aldri
eksistert, saa 26aa kolliderer ikke.
Klasse: 11 multi-locator, 5 replacement (de fem der fiksen er en om-merking:
provenance-mix 26y/26af/26ag og label/value 26ae/26ah).

HVER SUMMARY ER MAALT MOT FAKTISK TEKST, ikke mot datasettets bokser. Funn som
kom ut av den lesningen:
- 26aj/26ak baerer totalen i en «###»-OVERSKRIFT, ikke i en broedtekstlinje.
  «Slett linja» betyr da aa slette en seksjonsoverskrift, og nabooverskriften
  («Unique Sources: 8») er behold-klassen paa samme nivaa.
- 26ac/26ad/26ah har en GENERERINGS-TIDSSTEMPEL-bulletpunkt i samme blokk —
  samme kategori som idx-26u-s filstoerrelse under den ratifiserte referenten.
  Navngitt, ikke anvendt.
- 26x/26am har en etikett som spenner over tre-fire ULIKE utsagnstyper, saa
  ingen scope fjerner klassen uten enten aa endre etikettens referent stille
  eller ta med ikke-klasse-innhold.
- 26ai-s behold-bullet teller ARTIKLER under en URL-etikett — samme form som
  26ae/26ah, men paa behold-siden. Notert, bevisst ikke bokfoert.

ANKERE: 63 stk, alle verifisert ordrett OG unike med split-telling foer
skriving (gaten beviser tilstedevaerelse, aldri unikhet). For blokkform er
HELE kall-blokken ankret, inkludert behold-klasse-naboene, slik at en
sletting ikke kan ta dem stille — 795d494-laerdommen anvendt.

Diff er ren tilfoeyelse (236 +, 0 −); round-trip-sjekk foer skriving sikret at
JSON-formatet ikke ble reskrevet. Koen: 31 -> 47 entries (27 open, 20 resolved).
Ingen korpusfil roert. Suite 1047/1047.
This commit is contained in:
Kjell Tore Guttormsen 2026-08-11 22:14:56 +02:00
commit 1c07afd056
1 changed files with 236 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

236
scripts/kb-eval/data/g7-review-queue.json
View file

@ -409,6 +409,242 @@
"anchors": [
"**Total sources cited**: 7 unique URLs fra MCP-research"
]
},
{
"id": "idx-26x",
"file": "skills/ms-ai-advisor/references/copilot-extensibility/enterprise-governance-copilot-deployment.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket ikke_sjekkbar - the block states per-tool counts with NO total, so no internal cross-check exists. MEASURED at :919-:924: a \"**MCP-statistikk:**\" label followed by five bullets - 3 microsoft_docs_search calls, 2 microsoft_docs_fetch calls, 1 microsoft_code_sample_search call, then \"25+ unike Microsoft Learn URLs referert\" and \"15+ PowerShell code samples inkludert\". 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. WHAT IS UNDECIDED: the last two bullets are NOT call counts. The URL bullet is the KEEP class (a source count). The code-samples bullet is neither a call count nor a source-URL count - it is a content inventory, a third category the ratified form has never addressed. Deleting only the three call bullets leaves the \"**MCP-statistikk:**\" label standing over two bullets that are not call statistics, so the label's referent silently changes under the edit - the same referent-drift the ratified form exists to stop. Also worth recording for whoever adjudicates: both survivors are approximations (\"25+\", \"15+\") with no exact referent, so they are not checkable either, but they are outside this class. Candidate scopes, NEITHER ratified: (a) delete the three call bullets and re-label the heading, (b) delete the heading and the three call bullets and re-attach the two survivors. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-advisor/references/copilot-extensibility/enterprise-governance-copilot-deployment.md:919-924 (the MCP-statistikk block, file ends at :925); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[]); idx-26j resolution (ratified form); docs sec. 9.16",
"anchors": [
"**MCP-statistikk:**",
"- 3 microsoft_docs_search calls",
"- 2 microsoft_docs_fetch calls",
"- 1 microsoft_code_sample_search call",
"- 25+ unike Microsoft Learn URLs referert",
"- 15+ PowerShell code samples inkludert"
]
},
{
"id": "idx-26y",
"file": "skills/ms-ai-advisor/references/copilot-extensibility/microsoft-graph-api-copilot-integration.md",
"class": "replacement",
"status": "open",
"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 provenance_mix, and the ONLY member of all 58 in bucket \"tvetydig\". MEASURED at :544: \"**MCP calls:** 7 (3x docs_search, 2x docs_fetch, 1x code_sample_search, 1x ToolSearch)\". The verdict FLIPS between two defensible readings. Read as \"tool calls of any kind\", 3+2+1+1 = 7 and the line is consistent. Read as \"MCP calls\" - which is what the label says - the microsoft-learn components sum to 6, ToolSearch is not a microsoft-learn MCP tool, and the stated 7 is ONE TOO HIGH. THAT DIRECTION MATTERS BEYOND THIS FILE: sec. 9.16 measured 10 of 11 inconsistent members as stated < sum, and this file under the MCP-only reading is one of only two places in the whole class where the deviation runs the other way. It is part of the evidence that killed \"all deviations run the same way\" as an ABSOLUTE claim - the 91% tendency survives, the absolute does not. 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. OPEN QUESTION #21 APPLIES: this is the idx-26w provenance class - the defect is not the number but what the number is asserted to be a count OF. STATE records that all four members of this class (idx-26w plus the three booked here) are to be decided TOGETHER, not one at a time. Neighbour :543 \"**Total kilder:** 11 MCP-verified URLs, 4+ code samples\" is the KEEP class (a source count) and is anchored here so a deletion cannot take it silently.",
"evidence": "skills/ms-ai-advisor/references/copilot-extensibility/microsoft-graph-api-copilot-integration.md:543-544 (file ends at :545); dataset r11-footer-class-2026-08-11.json (form=line, b=tvetydig, flags=[provenance_mix]); idx-26w (same provenance class, open); docs sec. 9.16 point 5 (the direction finding)",
"anchors": [
"**MCP calls:** 7 (3x docs_search, 2x docs_fetch, 1x code_sample_search, 1x ToolSearch)",
"**Total kilder:** 11 MCP-verified URLs, 4+ code samples"
]
},
{
"id": "idx-26z",
"file": "skills/ms-ai-advisor/references/prompt-engineering/chain-of-thought-prompting.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM plus flag direction_stated_gt_enumerated. Bucket inkonsistent. MEASURED at :494-:500: a \"### MCP-kall utført\" heading, three numbered items (microsoft_docs_search, microsoft_code_sample_search, microsoft_docs_fetch), then \"**Totalt:** 4 MCP-kall, 3 unike Microsoft Learn-kilder.\" THIS IS THE SINGLE COUNTEREXAMPLE IN THE CLASS: stated 4 EXCEEDS the enumerated 3, while all ten other inconsistent members state a total BELOW their sum. sec. 9.16 records it as the finding that falsified \"all deviations run the same way\" as an absolute; the tendency (91%) survives, the absolute does not, and the earlier \"zero the other way\" was an artefact of measuring half the population. 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. WHAT IS UNDECIDED: the total line is MIXED. One sentence carries a call count (4 MCP-kall) AND a source count (3 unike Microsoft Learn-kilder). Deleting the line as written would delete a KEEP-class claim along with the call count. Splitting it means AUTHORING a replacement sentence - which is precisely the hazard named in idx-26v, \"the replacement authors the defect class it closes\". Note also that the surviving source count was NOT verified here; this entry books the counting defect, it does not certify that 3 unique Learn sources is true. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-advisor/references/prompt-engineering/chain-of-thought-prompting.md:494-500 (file ends at :505); dataset r11-footer-class-2026-08-11.json (form=block, b=inkonsistent, flags=[direction_stated_gt_enumerated]); docs sec. 9.16 point 5; idx-26v (replacement-authors-the-defect hazard)",
"anchors": [
"### MCP-kall utført",
"**Totalt:** 4 MCP-kall, 3 unike Microsoft Learn-kilder."
]
},
{
"id": "idx-26aa",
"file": "skills/ms-ai-advisor/references/prompt-engineering/domain-specific-prompt-optimization.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket ikke_sjekkbar - no total, so nothing to cross-check against. MEASURED at :587-:591: a \"### Verifiseringsmetode\" heading and three bullets - \"**MCP-søk** — 3 søk mot microsoft-learn (2026-02-04)\", \"**Fetch** — 2 fullstendige dokumenter hentet via microsoft_docs_fetch\", and \"**Code samples** — Søk mot microsoft_code_sample_search (ingen direkte treff ...)\". THIS FILE IS PART OF THE EVIDENCE FOR THE FOURTH ESCAPE FORM (sec. 9.16 point 2): the count rides on the ACTION WORD (\"MCP-søk ... 3 søk\"), not on the word \"kall\" or a tool identifier, so sweep A was blind to it and :589 was rescued only by its neighbouring 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. WHAT IS UNDECIDED: the third bullet records a call WITH NO NUMBER AT ALL. \"Delete the line\" is undefined for a call record that carries no count - it is neither a false number nor a checkable one, and the ratified form was written for a stated count. Candidate scopes, none ratified: delete all three bullets and the heading; delete the two numbered bullets only; or treat the numberless bullet as outside the class. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-advisor/references/prompt-engineering/domain-specific-prompt-optimization.md:587-591 (file ends at :603); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[]); docs sec. 9.16 point 2 (fourth escape form, this file named)",
"anchors": [
"### Verifiseringsmetode",
"- **MCP-søk** — 3 søk mot microsoft-learn (2026-02-04)",
"- **Fetch** — 2 fullstendige dokumenter hentet via microsoft_docs_fetch"
]
},
{
"id": "idx-26ab",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/inferencing-optimization-caching.md",
"class": "multi-locator",
"status": "open",
"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."
]
},
{
"id": "idx-26ac",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/infrastructure-as-code-mlops.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket ikke_sjekkbar - per-tool counts with no call total. MEASURED at :932-:937: a \"**MCP-research metadata:**\" label and five bullets - search 4, fetch 3, code_sample 1, then \"**Total sources:** 10\" and \"**Dato for research:** 2026-06-19\". 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. WHAT IS UNDECIDED: the block mixes three categories under one label. The three call bullets are the class. \"**Total sources:** 10\" is the KEEP class. \"**Dato for research:** 2026-06-19\" is a GENERATION TIMESTAMP - a fourth thing. Under the referent ratified on idx-26j (current provenance, one referent for the whole block) a generation-time date falls in the same category as a generation-time call count, which would make its deletion a ONE-STEP CONSEQUENCE OF AN EXISTING RATIFICATION rather than a new form - exactly the reasoning recorded for the file-size line in idx-26u. That consequence is NAMED HERE, NOT APPLIED: extending the form unasked is what idx-26w refused to do, and the same restraint applies. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-engineering/references/mlops-genaiops/infrastructure-as-code-mlops.md:932-937 (file ends at :949); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[]); idx-26u (generation-time metadata, same category argument); idx-26j resolution",
"anchors": [
"**MCP-research metadata:**",
"- **microsoft_docs_search calls:** 4",
"- **microsoft_docs_fetch calls:** 3",
"- **microsoft_code_sample_search calls:** 1",
"- **Total sources:** 10"
]
},
{
"id": "idx-26ad",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-versioning-registry-management.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket ikke_sjekkbar - no call total. MEASURED at :579-:584: a \"### MCP research summary\" heading and five bullets - \"**Total searches:** 3\", \"**Document fetches:** 2\", \"**Code samples:** 1\", \"**Unique sources:** 7 Microsoft Learn articles\", \"**Research timestamp:** April 2026\". THIS FILE IS THE STRONGEST EVIDENCE FOR THE FOURTH ESCAPE FORM: sec. 9.16 point 2 records that it was invisible IN ITS ENTIRETY to sweeps A and B, because the counting rides on \"Total searches\" - no \"kall\", no tool identifier next to a digit - and it entered the class ONLY through sweep D. A single sweep cannot prove its own completeness; this file is what that costs when it is assumed. 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. WHAT IS UNDECIDED, AND THIS ONE IS SHARP: the KEEP-class line \"**Unique sources:** 7 Microsoft Learn articles\" sits in the SAME BULLET LIST as the three call lines. Any scope defined at list level - \"delete the MCP research summary block\" - takes a keep-class claim with it. The timestamp bullet raises the same category question as idx-26ac and idx-26u. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-engineering/references/mlops-genaiops/model-versioning-registry-management.md:579-584 (file ends at :590); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[]); docs sec. 9.16 point 2 (this file named as invisible to sweeps A and B); idx-26ac and idx-26u (timestamp category)",
"anchors": [
"### MCP research summary",
"- **Total searches:** 3 (Azure ML registry, AI Foundry, MLOps lifecycle)",
"- **Document fetches:** 2 (Full registry guide, MLflow management guide)",
"- **Code samples:** 1 (MLflow Python SDK examples)",
"- **Unique sources:** 7 Microsoft Learn articles"
]
},
{
"id": "idx-26ae",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-document-preprocessing.md",
"class": "replacement",
"status": "open",
"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**"
]
},
{
"id": "idx-26af",
"file": "skills/ms-ai-governance/references/responsible-ai/ai-act-annex-iii-checklist.md",
"class": "replacement",
"status": "open",
"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: BLOCK FORM plus flag provenance_mix. Bucket ikke_sjekkbar - no call total. MEASURED at :521-:525: the heading \"### MCP-søk utført 2026-02\" over three bullets - \"microsoft_docs_search: 2 queries\", \"WebSearch: 4 queries\", \"tavily_extract: 5 URLs\". TWO OF THE THREE TOOLS ARE NOT MCP microsoft-learn. The heading asserts MCP provenance for all three, so the overclaim is structural: it is carried by the heading, not by any single bullet, which is a shape idx-26w did not have (there the qualifier sat inside the counted line itself). A second, independent problem: the three bullets do not share a unit - \"queries\" twice and \"URLs\" once - so even summing them is not well defined, and no total is stated. 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. OPEN QUESTION #21 APPLIES: this is the idx-26w provenance class - the defect is not the number but what the number is asserted to be a count OF. STATE records that all four members of this class (idx-26w plus the three booked here) are to be decided TOGETHER, not one at a time. Neighbour \"**Total sources referenced:** 8 (5 primary, 3 secondary)\" is the KEEP class and is anchored so it cannot be removed with the block.",
"evidence": "skills/ms-ai-governance/references/responsible-ai/ai-act-annex-iii-checklist.md:521-527 (file ends at :534); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[provenance_mix]); idx-26w (same class, open); STATE open question #21",
"anchors": [
"### MCP-søk utført 2026-02",
"- `microsoft_docs_search`: 2 queries (EU AI Act compliance, Purview AI governance)",
"- `WebSearch`: 4 queries (Annex III categories, classification criteria, forvaltningsloven, Art. 6(3))",
"- `tavily_extract`: 5 URLs (Official Annex III text, Article 6, DPO guide, Hjort analysis)",
"**Total sources referenced:** 8 (5 primary, 3 secondary)"
]
},
{
"id": "idx-26ag",
"file": "skills/ms-ai-governance/references/responsible-ai/ai-act-compliance-guide.md",
"class": "replacement",
"status": "open",
"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: BLOCK FORM plus flag provenance_mix. Bucket ikke_sjekkbar - no call total. MEASURED at :732-:735: the label \"**MCP-søk utført 2026-02-03:**\" over three bullets - \"microsoft_docs_search: 3 queries\", \"WebSearch: 2 queries\", \"microsoft_docs_fetch: 3 URLs\". Same structural shape as idx-26af: the MCP claim is carried by the LABEL and one of the three tools (WebSearch) is not MCP microsoft-learn, so the overclaim cannot be repaired by editing any single bullet. Units again differ across bullets (queries, queries, URLs) with no stated total. 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. OPEN QUESTION #21 APPLIES: this is the idx-26w provenance class - the defect is not the number but what the number is asserted to be a count OF. STATE records that all four members of this class (idx-26w plus the three booked here) are to be decided TOGETHER, not one at a time. Neighbour \"**Total sources referenced:** 16 (7 primary, 9 secondary/tertiary)\" is the KEEP class and is anchored.",
"evidence": "skills/ms-ai-governance/references/responsible-ai/ai-act-compliance-guide.md:732-737 (file ends at :744); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[provenance_mix]); idx-26w and idx-26af (same class); STATE open question #21",
"anchors": [
"**MCP-søk utført 2026-02-03:**",
"- `microsoft_docs_search`: 3 queries (EU AI Act compliance, governance, risk classification)",
"- `WebSearch`: 2 queries (EU AI Act 2026 requirements, Norway implementation)",
"- `microsoft_docs_fetch`: 3 URLs (Compliance Manager, AI governance guides)",
"**Total sources referenced:** 16 (7 primary, 9 secondary/tertiary)"
]
},
{
"id": "idx-26ah",
"file": "skills/ms-ai-governance/references/responsible-ai/ai-impact-assessment-framework.md",
"class": "replacement",
"status": "open",
"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)"
]
},
{
"id": "idx-26ai",
"file": "skills/ms-ai-governance/references/responsible-ai/ai-risk-taxonomy-classification.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket ikke_sjekkbar - per-tool counts with no call total. MEASURED at :458-:463: a \"### MCP Calls Summary\" heading over four bullets - search 3 calls, fetch 2 calls, code_sample 1 call, then \"**Totalt unike URLer:** 6 verified Microsoft Learn articles\". 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. WHAT IS UNDECIDED: the KEEP-class bullet sits inside the same list under a heading that names the whole block \"MCP Calls Summary\", so a heading-level scope removes a source count that is not part of the class - the idx-26ad problem in a second file. NOTED FOR THE ADJUDICATOR, NOT BOOKED: that surviving bullet has a label/value tension of its own - it counts ARTICLES under a URL label, the same shape as idx-26ae and idx-26ah - so deleting the call bullets does not leave a clean block behind. That is a keep-side observation and is deliberately left out of this entry's scope rather than widened into it. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-governance/references/responsible-ai/ai-risk-taxonomy-classification.md:458-463 (file ends at :474); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[]); idx-26ad (same list-level scope problem); idx-26ae, idx-26ah (label/value shape)",
"anchors": [
"### MCP Calls Summary",
"- **microsoft_docs_search:** 3 calls (AI risk classification, AI Act levels, Azure framework)",
"- **microsoft_docs_fetch:** 2 calls (AI Risk Assessment, Code of Conduct)",
"- **microsoft_code_sample_search:** 1 call (AI risk assessment code examples)",
"- **Totalt unike URLer:** 6 verified Microsoft Learn articles"
]
},
{
"id": "idx-26aj",
"file": "skills/ms-ai-governance/references/responsible-ai/algorithmic-accountability-auditability.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket konsistent - MEASURED at :562-:565, \"### MCP Calls: 6\" followed by 3x microsoft_docs_search, 2x microsoft_docs_fetch, 1x microsoft_code_sample_search, and 3+2+1 = 6. Consistency does not save it: the ground is unverifiability. WHAT MAKES THIS ONE STRUCTURALLY DIFFERENT: THE TOTAL IS CARRIED BY A MARKDOWN HEADING, not by a body line. sec. 9.16 records the method finding that a digit-less heading can hide a whole structural form; this is its inverse - a heading that IS the count. \"Delete the line\" therefore means deleting a section heading, and the document structure around it is part of the edit: the block sits between \"---\" separators, and the very next heading \"### Unique Sources: 8 verified Microsoft Learn URLs\" is KEEP class at the SAME level and carries its own count in the same shape. Deleting one heading and keeping its structural twin is a judgement the ratified form does not make. 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. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-governance/references/responsible-ai/algorithmic-accountability-auditability.md:560-567 (file ends at :572); dataset r11-footer-class-2026-08-11.json (form=block, b=konsistent, flags=[]); docs sec. 9.16 (heading-borne count); idx-26ak (second heading-borne instance)",
"anchors": [
"### MCP Calls: 6",
"- 3x `microsoft_docs_search`",
"- 2x `microsoft_docs_fetch`",
"- 1x `microsoft_code_sample_search`",
"### Unique Sources: 8 verified Microsoft Learn URLs"
]
},
{
"id": "idx-26ak",
"file": "skills/ms-ai-governance/references/responsible-ai/model-monitoring-drift-detection.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket inkonsistent - MEASURED at :774-:777: \"### Total MCP Calls: 4\" followed by microsoft_docs_search 3 queries, microsoft_docs_fetch 2 deep reads, microsoft_code_sample_search 1 query. The components sum to 6 against a stated 4 - the class's dominant direction (stated < sum), which sec. 9.16 measured at 10 of 11. 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. STRUCTURALLY this is the second heading-borne total, the same shape as idx-26aj: the number lives in a \"###\" heading, so the edit is a structural one, and the following heading \"### Total Unique URLs: 9\" is KEEP class at the same level with its own two-bullet breakdown (Primary 4 + Secondary 5 = 9, internally consistent). Decide the two heading-borne cases together or the corpus ends up with two different treatments of one shape. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-governance/references/responsible-ai/model-monitoring-drift-detection.md:774-781 (file ends at :782); dataset r11-footer-class-2026-08-11.json (form=block, b=inkonsistent, flags=[]); idx-26aj (same heading-borne shape); docs sec. 9.16 point 5",
"anchors": [
"### Total MCP Calls: 4",
"- microsoft_docs_search: 3 queries",
"- microsoft_docs_fetch: 2 deep reads",
"- microsoft_code_sample_search: 1 query",
"### Total Unique URLs: 9"
]
},
{
"id": "idx-26al",
"file": "skills/ms-ai-governance/references/responsible-ai/responsible-ai-framework-overview.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM (two lines), no flags. Bucket ikke_sjekkbar - no total. MEASURED at :374-:375: \"**MCP-søk utført:** 3 søk (microsoft-learn)\" and \"**Sider hentet:** 3 full-fetch (concept-responsible-ai, assurance-ai, concept-responsible-ai-dashboard)\". Another fourth-escape-form instance (sec. 9.16 point 2): the count rides on the action word \"MCP-søk utført\", with no \"kall\" and no tool identifier next to 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. WHAT IS UNDECIDED: the block is two lines with no heading and no wrapper, so there is no structural boundary to delete to - \"the line\" is ambiguous between one line and both. The second line differs from every other member of the class in one respect worth recording: IT NAMES THE THREE FETCHED DOCUMENTS, so unlike a bare count it is in principle checkable against the file's own source list. THAT CHECK WAS NOT RUN - it is a different check from the class's internal cross-check, and asserting a verdict without running it is the exact failure this programme keeps catching. Whoever adjudicates should either run it or record that it was not run. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-governance/references/responsible-ai/responsible-ai-framework-overview.md:373-375 (file ends at :381); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[]); docs sec. 9.16 point 2",
"anchors": [
"**MCP-søk utført:** 3 søk (microsoft-learn)",
"**Sider hentet:** 3 full-fetch (concept-responsible-ai, assurance-ai, concept-responsible-ai-dashboard)"
]
},
{
"id": "idx-26am",
"file": "skills/ms-ai-security/references/ai-security-engineering/prompt-injection-defense-patterns.md",
"class": "multi-locator",
"status": "open",
"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: BLOCK FORM, no flags. Bucket ikke_sjekkbar - no total. MEASURED at :476-:480: a \"**Research Coverage:**\" label over four bullets - \"3 MCP microsoft-learn docs_search calls\", \"3 MCP microsoft-learn docs_fetch calls (full documentation)\", \"9 unique source URLs from Microsoft Learn\", and a \"Coverage:\" bullet that is prose with no number. 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. WHAT IS UNDECIDED: the label \"Research Coverage\" spans four different kinds of statement - two call counts (the class), one source count (KEEP), and one prose enumeration of topics. There is no scope that removes the class without either leaving a label over a changed referent or taking non-class content with it. This is the idx-26x problem in a second file, and the two should be decided under one rule rather than file by file. OPEN QUESTION #22 APPLIES: \"delete the line\" is undefined for block form. This entry is booked, not repaired, and no deletion should be applied to it until the block-form deletion scope is ratified.",
"evidence": "skills/ms-ai-security/references/ai-security-engineering/prompt-injection-defense-patterns.md:476-480 (file ends at :484); dataset r11-footer-class-2026-08-11.json (form=block, b=ikke_sjekkbar, flags=[]); idx-26x (same mixed-label shape)",
"anchors": [
"**Research Coverage:**",
"- 3 MCP microsoft-learn docs_search calls",
"- 3 MCP microsoft-learn docs_fetch calls (full documentation)",
"- 9 unique source URLs from Microsoft Learn",
"- Coverage: Prompt Shields, Security Benchmark (AI-2, AI-3), LLM Security Planning, Content Filtering"
]
}
]
}