open/ms-ai-architect
1
0
Fork 0
Code Issues Releases 1 Activity
ms-ai-architect/scripts/kb-eval/data/r11-pilot-classification.json
Kjell Tore Guttormsen 67a34975fd 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.
2026-08-12 20:16:46 +02:00

3009 lines
223 KiB
JSON
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

{
"_meta": {
"purpose": "R11 pilot measurement (§10): fix-operation classification over the densest not_grounded sample. Read-only — no KB file and no ledger record was written.",
"contract": "docs/r11-tiered-fix-design.md §3/§4/§10",
"classifier": "scripts/kb-eval/lib/fix-op.mjs (the O1 driver with writes disabled)",
"ledger": "scripts/kb-eval/data/judge-pass-manifest.json",
"ledger_records": 243,
"threshold": "not_grounded >= 7 (source_silent excluded, per §10)",
"generated_from": "ledger snapshot at run time — counts are re-derived, never read from a plan",
"disclaimer_two_202s": "This population is 202 flags. §3's '202 flags whose claim and quote contain a numeric token' is a DIFFERENT 202, measured over the full 712-flag population. Do not conflate them."
},
"population": {
"files": 24,
"flags": 202
},
"s4_as_written": {
"O1": 6,
"note": "What §4 exactly as written would admit on the NUMERIC path (§4b status swaps excluded — they never run through the context condition). NOT a source of proposals: on the >=7 pilot all 6 were hand-verified and 4 were wrong edits (unit crossing, metric crossing, two mutilated identifiers) — measured precision 2/6. Runs at other thresholds carry no hand-verification."
},
"status_synonym": {
"contract": "§4b — the closed synonym table, ratified 2026-08-03",
"class_total": 15,
"proven": 5,
"aborts": {
"FILE_ALREADY_MATCHES": 2,
"NO_SOURCE_STATUS": 5,
"NO_COMPLETE_FILE_LABEL": 2,
"SOURCE_STATUS_AMBIGUOUS": 1
},
"hand_verified": "All 5 hand-judged 2026-08-03 (docs/r11-pilot-results.md appendix B). Four carry the source phrasing on the row's OWN subject and are correct. One (security-copilot-integration.md:94) harvests a \"(Preview)\" marker that belongs to a DIFFERENT agent in an enumerated quote — the same provenance-without-referent defect that falsified §4. Its outcome is plausibly right; its proof is not.",
"applicability": "REVIEW-GRADE, NOT APPLY-GRADE. status is deliberately absent from o1_recommended: §4b binds the table, the completeness of the file label and the written value, and nothing about whether the source phrasing refers to the row's subject. A referent condition is an open operator decision."
},
"o1_by_type": {
"status": 5,
"iso_date": 2
},
"o1_recommended": {
"count": 2,
"types": [
"iso_date"
],
"note": "The only O1 class that survived hand-verification: iso_date, which in this corpus is always an api-version bump inside a URL or code sample. number/version proposals are NOT safe to apply — they mutilate product, model and certification identifiers."
},
"split": {
"O1": 7,
"O2": 0,
"O3": 195,
"O2_note": "O2 requires operator ratification (§5); until then every non-O1 item is O3 by design."
},
"abort_codes": {
"MULTI_PART_CLAIM": 96,
"STATUS_SYNONYM": 10,
"LOCATOR_AMBIGUOUS": 7,
"MULTI_VALUE_TOKEN": 29,
"NOT_VERBATIM": 15,
"CONTEXT_MISMATCH": 4,
"MULTI_REPLACEMENT": 6,
"NO_VALUE_TOKEN": 28
},
"locator_aborts": {
"count": 7,
"note": "fixable engineering gap — every other abort is intrinsic to the flag"
},
"by_rule": {
"(none)": 22,
"R8": 87,
"R3": 11,
"R2": 60,
"R4": 15,
"R1": 2,
"R7": 5
},
"r8": {
"total": 87,
"O1": 0,
"codes": {
"MULTI_PART_CLAIM": 46,
"STATUS_SYNONYM": 6,
"LOCATOR_AMBIGUOUS": 4,
"MULTI_VALUE_TOKEN": 12,
"MULTI_REPLACEMENT": 3,
"NO_VALUE_TOKEN": 12,
"NOT_VERBATIM": 4
}
},
"items": [
{
"id": "ms-ai-engineering/agent-orchestration/foundry-agent-service-ga.md#6",
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 68,
"rule": "(none)",
"claim": "Hosted agents (din egen kode/container) er i Public Preview.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/agents/how-to/migrate-hosted-agent-preview",
"evidence_quote": "As of `azure-ai-projects` 2.3.0 on the GA `v1` API, hosted agents are generally available and this header is no longer required.",
"reason": "Hosted agents are now generally available, contradicting the claimed Public Preview status.",
"op": "O1",
"code": "PROVEN",
"proposal": {
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 68,
"token": "Preview",
"replacement": "GA",
"type": "status",
"status_row_from": "PREVIEW",
"status_row_to": "GA",
"before": "| Hosted agents (din egen kode/container) | **Preview** |",
"after": "| Hosted agents (din egen kode/container) | **GA** |",
"evidence_url": "https://learn.microsoft.com/azure/foundry/agents/how-to/migrate-hosted-agent-preview",
"evidence_quote": "As of `azure-ai-projects` 2.3.0 on the GA `v1` API, hosted agents are generally available and this header is no longer required."
}
},
{
"id": "ms-ai-engineering/agent-orchestration/foundry-agent-service-ga.md#11",
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 72,
"rule": "(none)",
"claim": "Logic Apps-triggerintegrasjon er GA.",
"evidence_url": "https://learn.microsoft.com/azure/foundry-classic/agents/how-to/triggers",
"evidence_quote": "Trigger an agent by using Logic Apps (preview) (classic)",
"reason": "Logic Apps trigger integration is labeled preview (both classic and the new Logic Apps agent-action doc), not GA as claimed.",
"op": "O1",
"code": "PROVEN",
"proposal": {
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 72,
"token": "GA",
"replacement": "Preview",
"type": "status",
"status_row_from": "GA",
"status_row_to": "PREVIEW",
"before": "| Logic Apps-triggerintegrasjon | **GA** |",
"after": "| Logic Apps-triggerintegrasjon | **Preview** |",
"evidence_url": "https://learn.microsoft.com/azure/foundry-classic/agents/how-to/triggers",
"evidence_quote": "Trigger an agent by using Logic Apps (preview) (classic)"
}
},
{
"id": "ms-ai-engineering/agent-orchestration/foundry-agent-service-ga.md#13",
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 187,
"rule": "R8",
"claim": "Innebygde verktøy i Foundry Agent Service: Code Interpreter | File Search | Grounding with Bing Search | Bing Custom Search | SharePoint | Azure Functions | Azure Logic Apps | OpenAPI tool | MCP tool | Deep Research tool | Fabric Data Agent | Morningstar tool.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/agents/how-to/migrate",
"evidence_quote": "Deep Research | Yes (Public Preview) | No (Recommendation: Deep Research model with Web Search tool)",
"reason": "The list asserts current built-in tools but includes Deep Research (deprecated/removed in new Foundry) and Morningstar (absent from the current tool catalog), so it describes a superseded classic tool set.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/agent-orchestration/foundry-agent-service-ga.md#15",
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 193,
"rule": "R8",
"claim": "SharePoint-verktøyet er i Preview (de øvrige innebygde verktøyene er GA).",
"evidence_url": "https://learn.microsoft.com/azure/foundry/agents/concepts/tool-catalog",
"evidence_quote": "Image Generation (preview) ... Browser Automation (preview) ... Computer Use (preview) ... Microsoft Fabric (preview) ... SharePoint (preview)",
"reason": "SharePoint being preview is correct, but the claim's load-bearing 'other built-in tools are GA' is false since Image Generation, Browser Automation, Computer Use and Microsoft Fabric are also preview.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "FILE_ALREADY_MATCHES",
"row": "PREVIEW"
}
},
{
"id": "ms-ai-engineering/agent-orchestration/foundry-agent-service-ga.md#16",
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 197,
"rule": "R8",
"claim": "MCP tool (koble til remote MCP-servere) er GA (juni 2025).",
"evidence_url": "https://learn.microsoft.com/azure/foundry/agents/how-to/migrate",
"evidence_quote": "MCP | Yes (Public Preview) | Yes (GA)",
"reason": "MCP reached GA only in the new Foundry; in classic (its June 2025 debut) it was Public Preview, so 'GA (June 2025)' is wrong on the date part.",
"op": "O3",
"code": "LOCATOR_AMBIGUOUS",
"detail": {
"token": "2025",
"hits": 2,
"window": {
"start": 187,
"end": 200
}
}
},
{
"id": "ms-ai-engineering/agent-orchestration/foundry-agent-service-ga.md#17",
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 198,
"rule": "(none)",
"claim": "Deep Research tool (o3-deep-research + Bing) er GA (juni 2025).",
"evidence_url": "https://learn.microsoft.com/azure/foundry-classic/agents/how-to/tools-classic/deep-research",
"evidence_quote": "The Deep Research tool is deprecated.",
"reason": "The Deep Research tool was Public Preview and is now deprecated/'No' in new Foundry — never GA — contradicting the claimed GA status.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"3",
"2025"
]
}
},
{
"id": "ms-ai-engineering/agent-orchestration/foundry-agent-service-ga.md#22",
"file": "skills/ms-ai-engineering/references/agent-orchestration/foundry-agent-service-ga.md",
"line": 340,
"rule": "R8",
"claim": "Foundry Agent Service er tilgjengelig i 19 regioner totalt, inkludert Norway East | Sweden Central | West Europe | Germany West Central | France Central | Switzerland North | UK South | East US / East US 2.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/agents/concepts/limits-quotas-regions?view=foundry",
"evidence_quote": "West Central US | Yes | Yes | No",
"reason": "The Agents column is Yes for 30 regions; the named regions are all present but the stated total of 19 is contradicted by 30.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#2",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 30,
"rule": "R3",
"claim": "Prebuilt-modellene støtter 27 språk",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/language-support/prebuilt?view=doc-intel-4.0.0",
"evidence_quote": "The following tables list the available language and locale support by model and feature ... Model ID: prebuilt-bankStatement | English (United States) en-US",
"reason": "The canonical language-support page has no uniform 27-language figure and its per-model frame contradicts one: support ranges from en-US only (bankStatement, check, contract, tax, mortgage) to ~40 languages (invoice) to 100+ (thermal receipts), so 'prebuilt models support 27 languages' matches no current enumeration.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "27",
"type": "number"
}
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#4",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 40,
"rule": "R8",
"claim": "Financial Services prebuilt-modeller: prebuilt-invoice | prebuilt-receipt | prebuilt-bankStatement | prebuilt-creditCard | prebuilt-check | prebuilt-contract | prebuilt-payStub.us",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/model-overview?view=doc-intel-4.0.0",
"evidence_quote": "prebuilt-check.us | ✓ | ✓",
"reason": "Six of the seven model IDs match the page's model analysis features table, but the bank-check model ID is prebuilt-check.us, not prebuilt-check as claimed - a copy-into-code SKU string, so one load-bearing part is wrong.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#5",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 52,
"rule": "R8",
"claim": "Identity & Tax prebuilt-modeller: prebuilt-idDocument | prebuilt-healthInsuranceCard.us | prebuilt-marriageCertificate | prebuilt-tax.us.w2 | prebuilt-tax.us.1098 | prebuilt-tax.us.1099 | prebuilt-tax.us.1040 | prebuilt-tax.us (unified)",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/model-overview?view=doc-intel-4.0.0",
"evidence_quote": "prebuilt-marriageCertificate.us | ✓ | ✓",
"reason": "The tax and ID model IDs match (prebuilt-tax.us.w2/.1098/.1099/.1040 and prebuilt-tax.us all appear), but the marriage-certificate model ID is prebuilt-marriageCertificate.us, not prebuilt-marriageCertificate as claimed - one load-bearing SKU string is wrong.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#6",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 65,
"rule": "R8",
"claim": "US Mortgage prebuilt-modeller: prebuilt-mortgage.us.1003 | prebuilt-mortgage.us.1004 | prebuilt-mortgage.us.1005 | prebuilt-mortgage.us.1008 | prebuilt-mortgage.us.disclosure",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/model-overview?view=doc-intel-4.0.0",
"evidence_quote": "| Closing Disclosure | Extract closing, transaction costs, and loan details. | prebuilt-mortgage.us.closingDisclosure |",
"reason": "prebuilt-mortgage.us.1003/.1004/.1005/.1008 match the page, but the fifth model ID is prebuilt-mortgage.us.closingDisclosure, not prebuilt-mortgage.us.disclosure as claimed.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#7",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 77,
"rule": "R8",
"claim": "Grunnleggende modeller: prebuilt-read (OCR: tekst, linjer, ord, språkdeteksjon) | prebuilt-layout (tabeller, selection marks, seksjoner, valgfrie key-value pairs) | prebuilt-document (key-value pairs, tabeller, selection marks)",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/model-overview?view=doc-intel-4.0.0",
"evidence_quote": "All the capabilities for the general document model are available in the layout model. The general model is no longer supported.",
"reason": "prebuilt-read and prebuilt-layout descriptions are grounded, but the page states the general document model (prebuilt-document) is no longer supported in v4.0 (its 2024-11-30 column reads 'Supported in layout model'), so presenting it as a current basic model contradicts the source.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#8",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 234,
"rule": "R8",
"claim": "AI Builder bruker Document Intelligence v3.1 (ikke alltid v4.0); Premium-lisens påkrevd for AI Builder; 1M AI Builder credits inkludert i visse Power Apps/Automate-lisenser",
"evidence_url": "https://learn.microsoft.com/ai-builder/credit-management",
"evidence_quote": "AI Builder credits coming from the AI Builder add-on (1,000,000 credits per add-on) and AI Builder credits coming from licenses with seeded capacity (like Power Automate premium, which brings 5,000 credits) are gathered at the tenant level.",
"reason": "The premium-feature part holds, but the 1M-credits part is contradicted: 1,000,000 credits come only from the purchased AI Builder capacity add-on, while Power Apps/Automate licenses seed just 250-5,000 credits (and seeded credits are removed November 1, 2026); the DI v3.1 part was not confirmed either.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"3.1",
"4.0",
"1"
]
}
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#10",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 391,
"rule": "R8",
"claim": "Free (F0): 500 sider/måned, 2 sider per dokument, 20 calls/min | Standard (S0): 2 000 sider per dokument, 15 TPS",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/service-limits?view=doc-intel-4.0.0",
"evidence_quote": "Analyze transactions Per Second limit | 1 | 15 (default value) ... Max number of pages (Analysis) | 2 | 2000",
"reason": "F0 2 pages/document, S0 2,000 pages, and S0 15 TPS all match, but the F0 rate limit is stated as 1 analyze transaction per second, not '20 calls/min' as claimed (and the 500 pages/month allowance appears only on the JS-rendered pricing page), so one load-bearing part is contradicted.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#11",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 355,
"rule": "(none)",
"claim": "Standard 30-dagers oppbevaring av dokumenter i Document Intelligence (kan slettes umiddelbart etter prosessering)",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/faq?view=doc-intel-4.0.0",
"evidence_quote": "Your data is then deleted 24 hours from the time that you submit an analyze request. If you would like the data deleted sooner, you can call the delete analyze response.",
"reason": "The documented standard retention is 24 hours (deletable sooner via the Delete Analyze Result API), not 30 days as claimed - the immediate-deletion part holds but the retention value is contradicted.",
"op": "O3",
"code": "CONTEXT_MISMATCH",
"detail": {
"token": "30",
"replacement": "24",
"would_have_been": "4. **Data retention:** Standard 24-dagers oppbevaring av dokumenter i Document Intelligence (kan slettes umiddelbart etter prosessering)"
}
},
{
"id": "ms-ai-engineering/azure-ai-services/document-intelligence-prebuilt-models.md#12",
"file": "skills/ms-ai-engineering/references/azure-ai-services/document-intelligence-prebuilt-models.md",
"line": 488,
"rule": "(none)",
"claim": "Custom models utløper (slutter å virke) etter 12–24 måneder; krever retraining-schedule",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/document-intelligence/train/custom-lifecycle?view=doc-intel-4.0.0",
"evidence_quote": "The model is configured to expire two years after its creation for all requests utilizing a GA API to build it.",
"reason": "The retraining-need part holds, but the page states a uniform two-year expiration (also two years for preview-trained models), not the claimed '12-24 months' - no 12-month expiry exists, so the stated interval contradicts the documented value.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"12",
"24"
]
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#4",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 50,
"rule": "R2",
"claim": "For kategoriske features bruker Azure ML drift detection disse metrikkene: Pearson's Chi-Squared Test | Euclidean Distance (beregnet på empiriske fordelinger av kategoriske kolonner).",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/reference-yaml-monitor?view=azureml-api-2",
"evidence_quote": "Allowed categorical metric names: `jensen_shannon_distance`, `chi_squared_test`, `population_stability_index`",
"reason": "Pearson's Chi-Squared Test stemmer, men Euclidean Distance er fraværende fra den kanoniske kategoriske metrikklisten for v2 model monitoring — den er en v1 Dataset Monitor-metrikk (how-to-monitor-datasets), ikke en v2 drift detection-metrikk.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#9",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 112,
"rule": "R4",
"claim": "ServerlessSparkCompute for model monitoring konfigureres med runtime_version \"3.3\" (Spark runtime-versjon).",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-monitor-model-performance?view=azureml-api-2",
"evidence_quote": "spark_compute = ServerlessSparkCompute( instance_type=\"standard_e4s_v3\", runtime_version=\"3.4\" )",
"reason": "Den siterte siden bruker runtime_version \"3.4\" i samtlige SDK- og CLI-eksempler; 3.3 er en superseded verdi som kun henger igjen i skjematabellen på reference-yaml-monitor, hvis egne YAML-eksempler også er oppdatert til 3.4.",
"op": "O3",
"code": "MULTI_REPLACEMENT",
"detail": {
"candidates": [
"4",
"3",
"3.4"
]
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#10",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 144,
"rule": "R4",
"claim": "Azure ML Dataset Monitors (preview, v1 SDK) er deprecated, og migrering til Model Monitor (v2 SDK) er anbefalt.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-monitor-datasets?view=azureml-api-1",
"evidence_quote": "Data drift (preview) was retired on September 1, 2025. Migrate to Model Monitor for your data drift tasks.",
"reason": "Migreringsanbefalingen stemmer, men statusen er ikke «deprecated»: funksjonen er retired per 1. september 2025 — et senere og handlingsendrende stadium enn deprecation, og eksakt-verdi-regelen gjelder med særlig kraft for status.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"1",
"2"
]
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#14",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 191,
"rule": "R8",
"claim": "Data drift monitoring krever: Azure ML workspace (v2 API) | compute (serverless Spark eller managed compute cluster) | datastore for production inference data (Azure Blob Storage eller ADLS Gen2) | valgfritt Application Insights for custom metrics logging.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-monitor-model-performance?view=azureml-api-2",
"evidence_quote": "Schedule model monitoring jobs to run on serverless Spark compute pools.",
"reason": "Workspace (v2) og Blob Storage stemmer, men den bærende compute-delen er feil: både how-to-siden og skjemaet krever Spark pool («Description of compute resources for Spark pool to run monitoring job»), «managed compute cluster» er ikke et alternativ, og Application Insights står ikke i prerequisites for v2 model monitoring.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#18",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 218,
"rule": "R8",
"claim": "Microsoft Foundry har egen monitoring for generativ AI med generation quality metrics: groundedness | relevance | fluency, og støtter drift detection for grounding data i RAG-scenarier.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/concepts/observability",
"evidence_quote": "Scheduled evaluation: Scheduled quality and safety evaluation using test datasets to detect system drift",
"reason": "Første del holder (Foundry har egen gen-AI-monitorering med groundedness, relevance og fluency), men den kanoniske observability-siden lister kun Evaluation, Monitoring og Tracing og «system drift» via planlagt evaluering på testdatasett — drift detection for grounding data i RAG er fraværende.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#20",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 388,
"rule": "R2",
"claim": "Out-of-box model monitoring-signaler for online endpoints er: Data quality | Data drift | Prediction drift | Feature attribution drift | Custom signals (brukerdefinerte metrics via Python-skript).",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/reference-yaml-monitor?view=azureml-api-2",
"evidence_quote": "**Optional** for basic model monitoring that uses recent past production data as comparison baseline and has 3 monitoring signals: data drift, prediction drift, and data quality.",
"reason": "Den kanoniske opplistingen gir out-of-box nøyaktig tre signaler (data drift, prediction drift, data quality) — feature attribution drift og custom signals er fraværende og krever advanced- eller custom-oppsett, hvilket how-to-siden bekrefter.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#21",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 395,
"rule": "R8",
"claim": "Oppsettsalternativene for model monitoring er: Out-of-box (automatisk konfigurert for Azure ML online endpoints, ingen konfigurasjon påkrevd) | Advanced (custom monitoring for modeller deployet utenfor Azure ML, batch endpoints eller eksternt) | Azure Event Grid-integrasjon for ruting av alerts.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-monitor-model-performance?view=azureml-api-2",
"evidence_quote": "You can monitor models that you deploy to Azure Machine Learning batch endpoints or models you deploy outside Azure Machine Learning.",
"reason": "Siden definerer «advanced» som flere signaler, trenings-/valideringsdata som referanse og topp-N-features, mens modeller utenfor Azure ML og batch endpoints hører til et eget oppsett («Set up model monitoring for production data») — claimets kategorimapping av Advanced er dermed feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/data-drift-monitoring-detection.md#22",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/data-drift-monitoring-detection.md",
"line": 400,
"rule": "R8",
"claim": "Statistiske metoder brukt av Azure ML model monitoring: Jensen-Shannon divergence for kategoriske features | Wasserstein distance (Earth Mover's Distance) for numeriske features | Population Stability Index (PSI) for feature-stabilitet.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/reference-yaml-monitor?view=azureml-api-2",
"evidence_quote": "Allowed numerical metric names: `jensen_shannon_distance`, `normalized_wasserstein_distance`, `population_stability_index`, `two_sample_kolmogorov_smirnov_test`",
"reason": "jensen_shannon_distance er tillatt for både numeriske og kategoriske features (ikke kategoriske alene), metrikken heter «Jensen-Shannon Distance» og ikke divergence, og «Normalized Wasserstein Distance» er det korrekte navnet — mappingen claimet setter opp finnes ikke i dokumentasjonen og motsier claim #3 i samme fil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#3",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 93,
"rule": "R2",
"claim": "Foundry Agent Evaluation utfører evaluering med LLM judges for correctness | relevance | groundedness | safety.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/concepts/built-in-evaluators",
"evidence_quote": "General purpose evaluators | Coherence | Fluency ... RAG evaluators | Retrieval | Document Retrieval | Groundedness | Groundedness Pro (preview) | Relevance | Response Completeness (preview)",
"reason": "Jeg hentet den kanoniske enumereringen (Built-in evaluators reference) — relevance, groundedness og safety-evaluatorer finnes, men INGEN «Correctness»-evaluator er oppført i noen kategori (Correctness er en MLflow-scorer, ikke en Foundry-evaluator); fraværet er bevis mot eksistenspåstanden.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#6",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 176,
"rule": "R8",
"claim": "Azure ML model monitoring kan kjøre på ServerlessSparkCompute med instance_type standard_e4s_v3 og runtime_version 3.3.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/machine-learning/how-to-monitor-model-performance?view=azureml-api-2",
"evidence_quote": "spark_compute = ServerlessSparkCompute(instance_type=\"standard_e4s_v3\", runtime_version=\"3.4\")",
"reason": "instance_type standard_e4s_v3 stemmer, men siden angir runtime_version \"3.4\" i alle eksempler (YAML og SDK) — claimets 3.3 er en gammel verdi; én bærende del er feil.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"4",
"3",
"3.3"
]
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#7",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 241,
"rule": "R2",
"claim": "MLflow for GenAI definerer en 10-stegs kontinuerlig forbedringssyklus: Production App (traces) | User Feedback (thumbs up/down) | Monitor & Score (LLM judges) | Identify Issues (Trace UI) | Domain Expert Review (Review App) | Build Eval Dataset | Tune Scorers | Evaluate New Versions | Compare Results (MLflow evaluation runs) | Deploy or Iterate.",
"evidence_url": "https://learn.microsoft.com/azure/databricks/mlflow3/genai/",
"evidence_quote": "MLflow 3 for GenAI provides these key pieces for efficient development, deployment, and continuous improvement: Tracing ... Built-in and custom LLM judges and scorers ... Review apps for expert feedback ... Automated evaluation and monitoring ... App and prompt versioning",
"reason": "Den siterte kanoniske siden definerer forbedringssyklusen med FEM nøkkelelementer; ingen Learn-side definerer en «10-stegs» syklus med de ti navngitte stegene claimet lister.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#8",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 443,
"rule": "R8",
"claim": "Feedback loop-komponentene i Azure Machine Learning er: Data collection via inference tables på managed endpoints | Monitoring via Model Monitor | Alerting via Azure Monitor Alerts (e-post/webhook ved threshold breach) | Retraining via Azure ML Pipelines | A/B-testing via staging endpoints | Deployment via Managed Online Endpoints med blue-green deployment.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/concept-data-collection?view=azureml-api-2",
"evidence_quote": "Azure Machine Learning Data collector provides real-time logging of input and output data from models that are deployed to managed online endpoints or Kubernetes online endpoints. Azure Machine Learning stores the logged inference data in Azure blob storage.",
"reason": "Model Monitor, retraining via pipelines og blue-green på managed online endpoints holder, men datainnsamlingen skjer via Data collector til Azure Blob Storage — Azure ML har ingen «inference tables» (det er et Databricks-begrep); én bærende del er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#9",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 471,
"rule": "R8",
"claim": "Feedback loop-komponentene for GenAI i Microsoft Foundry er: MLflow Tracing (Databricks, span-nivå telemetri) | Review App (thumbs up/down og tekstlig feedback) | Agent Evaluation (LLM judges) | Microsoft Foundry Observability (dashboard for kvalitetstrender, latens, feil) | MLflow Datasets i Unity Catalog (versjonerte testsett) | AI Red Teaming Agent (adversarial testing for sikkerhet).",
"evidence_url": "https://learn.microsoft.com/azure/foundry/concepts/observability",
"evidence_quote": "Microsoft Foundry provides three core capabilities that work together to deliver comprehensive observability across the AI application lifecycle: Evaluation ... Monitoring ... Tracing",
"reason": "Foundry Observability-dashboardet og AI red teaming agent står på siden, men den kanoniske Foundry-enumereringen inneholder verken Review App, «Agent Evaluation» eller MLflow Datasets i Unity Catalog — det er Databricks-MLflow-komponenter, ikke Microsoft Foundry-komponenter.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#11",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 552,
"rule": "R8",
"claim": "Feedback loop-komponentene i Power Platform AI er: Power Automate (ruting av low-confidence predictions til human review) | Dataverse / SharePoint (lagring av feedback-data) | AI Builder Feedback Loop (legger reviewede samples automatisk til treningssettet) | AI Builder (manuell/planlagt retraining).",
"evidence_url": "https://learn.microsoft.com/en-us/ai-builder/feedback-loop",
"evidence_quote": "Select Feedback loop and add the documents that could improve your model. Tag these new documents and retrain the model.",
"reason": "Power Automate-ruting på confidence score og Dataverse-lagring stemmer, men feedback loop legger IKKE reviewede samples automatisk til treningssettet — dokumentene må velges, tagges og modellen retrenes manuelt; SharePoint nevnes heller ikke som lagring (siden sier eksplisitt Dataverse-tabellen «AI Builder Feedback Loop»).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#14",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 663,
"rule": "R2",
"claim": "Azure ML Enterprise er inkludert i Azure-subscription og faktureres som per-use compute pricing.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/concept-plan-manage-cost?view=azureml-api-2",
"evidence_quote": "When you create resources for an Azure Machine Learning workspace, you also create resources for other Azure services. They are: Azure Container Registry basic account, Azure Blob Storage (general purpose v2), Azure Key Vault, Azure Monitor ... Each VM is billed per hour that it runs.",
"reason": "Den kanoniske «full billing model»-siden beskriver hele faktureringsmodellen uten NOEN edition/tier — «Azure ML Enterprise» finnes ikke (Basic/Enterprise-editionene er utgått); per-use compute-delen stemmer, men den navngitte SKU-en gjør ikke det.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-engineering/mlops-genaiops/feedback-loops-continuous-improvement.md#16",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/feedback-loops-continuous-improvement.md",
"line": 670,
"rule": "R4",
"claim": "Power Platform-lisensene gir følgende AI Builder-kreditter og feedback loop-støtte: Per User = 500 kreditter/mnd med feedback loop-støtte | Per App = ingen kreditter inkludert og ingen feedback loop-støtte (krever Per User) | AI Builder add-on = kreditter kjøpes ekstra, med feedback loop-støtte.",
"evidence_url": "https://learn.microsoft.com/ai-builder/credit-management",
"evidence_quote": "| Power Apps per app | 250 | Maximum = 1,000,000 AI Builder credits per tenant. Per app licenses purchased before November 2022, don't include any credits. |",
"reason": "Per user = 500 stemmer (Power Apps Premium/per user), men gjeldende rad gir Per app 250 kreditter — claimets «ingen kreditter inkludert» matcher kun den tidsstemplede legacy-regelen for lisenser kjøpt før november 2022; feedback loop-støtte per lisenstype er dessuten ikke dokumentert.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/mlops-teams-collaboration-tools.md#5",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/mlops-teams-collaboration-tools.md",
"line": 153,
"rule": "R2",
"claim": "Azure Artifacts tilbyr pakkefeeds for NuGet | pip | conda til ML-biblioteker og delte komponenter.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/devops/artifacts/start-using-azure-artifacts",
"evidence_quote": "Azure Artifacts supports multiple package types, including NuGet, npm, Python, Maven, Cargo, and Universal Packages.",
"reason": "NuGet og pip (Python) stemmer, men conda finnes ikke i den kanoniske opplistingen av pakketyper — heller ikke i Feature availability-tabellen (NuGet, dotnet, npm, Maven, Gradle, Python, Cargo, Universal Packages), saa conda-feeden er ikke tilbudt.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/mlops-teams-collaboration-tools.md#6",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/mlops-teams-collaboration-tools.md",
"line": 155,
"rule": "R8",
"claim": "Azure DevOps MCP Server gir naturlig-språk-spørringer for prosjektstyring (f.eks. «Summarize sprint status», «List blocked work items», «Show pipeline success rates») og er en 2026-funksjon.",
"evidence_url": "https://learn.microsoft.com/azure/devops/release-notes/features-timeline-released#azure-devops-services",
"evidence_quote": "June 30 2025 | Azure DevOps MCP Server public preview",
"reason": "Naturlig-spraak-delen stemmer (what-is-azure-devops viser promptene Summarize the current sprint status, List all work items that are blocked og Show the success rate for all pipelines), men den baerende dateringen er feil: MCP-serveren gikk i public preview 30. juni 2025 (sprint 258) og ble GA i sprint 264 i 2025 — ikke en 2026-funksjon.",
"op": "O3",
"code": "MULTI_REPLACEMENT",
"detail": {
"candidates": [
"30",
"2025"
]
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/mlops-teams-collaboration-tools.md#10",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/mlops-teams-collaboration-tools.md",
"line": 204,
"rule": "R2",
"claim": "Azure Pipelines-oppgaven for modellutrulling til Azure Machine Learning er AzureMLModelDeploy@1.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-devops-machine-learning",
"evidence_quote": "- task: AzureMLJobWaitTask@1",
"reason": "Den kanoniske siden for Azure Pipelines mot Azure Machine Learning dokumenterer kun AzureMLJobWaitTask@1 fra Machine Learning-utvidelsen, og modellutrulling skjer via pipeline-YAML og az ml CLI (deploy-online-endpoint-pipeline.yml paa den siterte siden); ingen learn-side nevner en oppgave AzureMLModelDeploy@1.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "1",
"type": "number"
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/mlops-teams-collaboration-tools.md#12",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/mlops-teams-collaboration-tools.md",
"line": 262,
"rule": "R2",
"claim": "Azure Artifacts har kapabilitetene private Python-pakkefeeds | conda-pakkehosting | Docker image registry (Azure Container Registry) | dependency security scanning.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/devops/artifacts/start-using-azure-artifacts",
"evidence_quote": "Azure Artifacts supports multiple package types, including NuGet, npm, Python, Maven, Cargo, and Universal Packages.",
"reason": "Private Python-feeds stemmer, men conda-pakkehosting finnes ikke i pakketype-opplistingen, og Docker image registry (Azure Container Registry) samt dependency security scanning er ikke Azure Artifacts-kapabiliteter — kapabilitetslisten paa what-is-azure-devops er upstream sources, versjonering, tilgangskontroll, build-integrasjon og kodesoek.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/mlops-teams-collaboration-tools.md#17",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/mlops-teams-collaboration-tools.md",
"line": 425,
"rule": "R8",
"claim": "Anbefalt RBAC for Azure ML-workspaces: dev-workspace - data scientists har Contributor og data analysts har Reader; staging-workspace - model testers har Contributor og data scientists har Reader; produksjons-workspace - kun CI/CD-prosesser og platform support har Owner.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/architecture/ai-ml/guide/machine-learning-operations-v2",
"evidence_quote": "| Model tester | R | R, KVR | R | R | R | R | LAR | MR |",
"reason": "Owner-delen stemmer (kun platform technical support og CI/CD processes har O i produksjonstabellen), men RBAC-tabellene gir data scientists ADS (Machine Learning Data Scientist) — ikke Contributor — paa dev-workspacet, og model testers har Reader i preproduksjon og ingen workspace-rolle i produksjonsmiljoeene (som ifoelge siden inkluderer staging og test), aldri Contributor.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-engineering/mlops-genaiops/mlops-teams-collaboration-tools.md#19",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/mlops-teams-collaboration-tools.md",
"line": 494,
"rule": "R3",
"claim": "Delte prompt flows for team-samarbeid ligger i Azure AI Studio.",
"evidence_url": "https://learn.microsoft.com/azure/foundry-classic/concepts/prompt-flow",
"evidence_quote": "Prompt flow in Microsoft Foundry and Azure Machine Learning will be retired on April 20, 2027. Prompt flow is no longer recommended for new development.",
"reason": "Team-samarbeid rundt flows er bekreftet (Debug, share, and iterate your flows with ease through team collaboration), men rammen er superseded: dokumentasjonen plasserer prompt flow i Microsoft Foundry portal (classic) og Azure Machine Learning studio — produktnavnet Azure AI Studio finnes ikke lenger — og prompt flow er dessuten under utfasing med frist 2027-04-20.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-engineering/mlops-genaiops/mlops-teams-collaboration-tools.md#21",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/mlops-teams-collaboration-tools.md",
"line": 541,
"rule": "R8",
"claim": "Azure DevOps gratis tier gir opptil 5 brukere med Basic access og ubegrenset antall stakeholders med read-only-tilgang.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/devops/organizations/security/access-levels",
"evidence_quote": "Stakeholders use drag-and-drop to create and change work items, but they can only change the State field on cards.",
"reason": "Fem gratis brukere med Basic og ubegrenset antall gratis stakeholders stemmer, men den baerende delen read-only-tilgang er feil: den kanoniske access-levels-siden gir Stakeholder tilgang til View My Work Items med add and modify work items, samt oppretting og endring av kort paa boards.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "5",
"type": "number"
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#1",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 37,
"rule": "R3",
"claim": "AI-assisterte kvalitetsmetrikker i Microsoft Foundry omfatter Groundedness | Relevance | Coherence | Fluency | GPT similarity, krever en judge-modell (GPT-3.5+/GPT-4), og kun GPT similarity krever ground truth.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/concepts/built-in-evaluators",
"evidence_quote": "Similarity | AI-assisted textual similarity measurement.",
"reason": "Gjeldende Foundry-enumerasjon navngir metrikken «Similarity» (SDK-klasse SimilarityEvaluator, nøkkel «similarity»); «GPT similarity» er den utgåtte Prompt Flow-/gpt_-prefiks-benevnelsen, så claimets ramme er en omdøpt/erstattet metrikk.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#3",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 39,
"rule": "R8",
"claim": "Risk & Safety-metrikker i Microsoft Foundry omfatter Self-harm | Hateful content | Violence | Sexual content | Protected material | Indirect attack, krever ikke ground truth, og kjøres av en Foundry-hostet GPT-4.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/concepts/evaluation-evaluators/risk-safety-evaluators",
"evidence_quote": "Unlike LLM-as-judge evaluators such as coherence and fluency, these evaluators run against Microsoft's hosted safety models.",
"reason": "De seks risikonavnene og «krever ikke ground truth» (required inputs: query, response) holder, men den bærende delen «kjøres av en Foundry-hostet GPT-4» motsies: siden sier tjenesten «employs a set of language models» / «hosted safety models» og kontrasterer dem eksplisitt mot GPT-baserte LLM-as-judge-evaluatorer.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#4",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 43,
"rule": "(none)",
"claim": "Microsoft Foundry støtter tre evalueringsmål: Model | Agent | Dataset.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/how-to/evaluate-generative-ai-app?view=foundry-classic",
"evidence_quote": "**Traces** | Evaluates agent interactions already captured in Application Insights. Select the agent and time range, and the portal retrieves the matching traces for evaluation.",
"reason": "Steg 1-tabellen «Select evaluation target» lister fire mål — Agent, Model, Dataset OG Traces — så den eksakte påstanden om «tre evalueringsmål» er superseded.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#5",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 51,
"rule": "R8",
"claim": "Data mapping-krav for Foundry-evaluering: Groundedness og Relevance krever query + response + context; Coherence og Fluency krever query + response; GPT similarity krever query + response + ground truth; F1/BLEU/ROUGE/METEOR krever response + ground truth; safety-metrikker krever query + response.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/concepts/evaluation-evaluators/rag-evaluators",
"evidence_quote": "| Relevance | `query`, `response` | `deployment_name` |",
"reason": "Flere deler stemmer (Groundedness med context, Similarity med ground truth, NLP-metrikker med response+ground truth, safety med query+response), men den bærende delen «Relevance krever query + response + context» motsies — RAG-evaluatortabellen krever kun query og response for Relevance.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "1",
"type": "number"
}
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#8",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 129,
"rule": "R3",
"claim": "MLflow 3 har fem scorer-typer: Built-in judges | Guidelines judges | Custom LLM judges | Code-based scorers | Multi-turn judges.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/databricks/mlflow3/genai/eval-monitor/concepts/scorers",
"evidence_quote": "Built-in judges also include **Guidelines judges**, built-in judges that check whether responses pass or fail custom natural-language rules, such as style or factuality guidelines.",
"reason": "Siden strukturerer scorers i FIRE tilnærminger (Built-in judges, Custom judges, Code-based scorers, Third-party scorers); Guidelines judges og multi-turn judges er undertyper av built-in judges, og Third-party scorers mangler helt i claimet — claimets fem-type-ramme er erstattet.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#9",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 133,
"rule": "R2",
"claim": "MLflow 3 built-in judges omfatter Correctness | RetrievalGroundedness | Safety | RelevanceToQuery | Fluency | Equivalence.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/databricks/mlflow3/genai/eval-monitor/concepts/judges/",
"evidence_quote": "RelevanceToQuery, RetrievalRelevance, Safety, RetrievalGroundedness, Correctness, RetrievalSufficiency, Guidelines, ExpectationsGuidelines, ToolCallCorrectness, ToolCallEfficiency",
"reason": "«Available judges»-tabellen på den kanoniske Built-in LLM judges-siden lister disse ti; verken «Fluency» eller «Equivalence» finnes der — fraværet i den enumererende siden er bevis mot eksistens-påstanden.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#18",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 384,
"rule": "R2",
"claim": "AI-assisterte safety-metrikker er kun hostet i regionene East US 2 | France Central | UK South | Sweden Central.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/concepts/evaluation-regions-limits-virtual-network",
"evidence_quote": "These regions support the following safety evaluators: Hate and unfairness, Sexual, Violent, Self-harm, Indirect attack, Code vulnerabilities, and Ungrounded attributes.",
"reason": "Den kanoniske regionlisten for risk- og safety-evaluatorer er East US 2, North Central US, France Central, Sweden Central, Switzerland West og Australia East — UK South står ikke der, og claimets «kun»-liste utelater tre faktiske regioner.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/mlops-genaiops/model-evaluation-frameworks.md#21",
"file": "skills/ms-ai-engineering/references/mlops-genaiops/model-evaluation-frameworks.md",
"line": 430,
"rule": "R2",
"claim": "MLflow 3 er inkludert i Databricks-abonnement på Premium- eller Enterprise-tier.",
"evidence_url": "https://learn.microsoft.com/azure/databricks/admin/account-settings/account",
"evidence_quote": "Azure Databricks is available with two pricing options, Standard and Premium, which offer features for different types of workloads.",
"reason": "Azure Databricks har prisnivåene Standard, Premium (og Trial ved opprettelse) — noe «Enterprise-tier» finnes ikke i den kanoniske enumerasjonen, så tier-påstanden er feil.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "3",
"type": "number"
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#1",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 4,
"rule": "R8",
"claim": "Filen angir status GA for Azure OpenAI og Azure AI Search, og Preview for Multilingual E5 og Custom embeddings.",
"evidence_url": "https://learn.microsoft.com/azure/search/vector-search-how-to-generate-embeddings",
"evidence_quote": "One step involves selecting an embedding model to vectorize your plain text content. The following models are supported: 1. text-embedding-3-small 2. text-embedding-3-large 3. text-embedding-ada-002 4. Cohere-embed-v3-english 5. Cohere-embed-v3-multilingual",
"reason": "GA-delene for Azure OpenAI og Azure AI Search holder, men den kanoniske listen over embedding-modeller Azure tilbyr inneholder verken Multilingual E5 eller Custom embeddings, så Preview-statusen for disse to bærende delene er ikke dekket.",
"op": "O3",
"code": "MULTI_REPLACEMENT",
"detail": {
"candidates": [
"1",
"3",
"2",
"002",
"4"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#8",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 43,
"rule": "R2",
"claim": "multilingual-e5-small og multilingual-e5-large har status Preview og tilbys via Azure AI.",
"evidence_url": "https://learn.microsoft.com/azure/search/vector-search-how-to-generate-embeddings",
"evidence_quote": "The following models are supported: 1. text-embedding-3-small 2. text-embedding-3-large 3. text-embedding-ada-002 4. Cohere-embed-v3-english 5. Cohere-embed-v3-multilingual",
"reason": "Jeg hentet de kanoniske listene over embedding-modeller Azure tilbyr (Azure AI Search/Foundry-veiviseren og Foundry Models sold by Azure) — multilingual-e5-small og multilingual-e5-large er fraværende i begge, og fraværet er bevis mot at de tilbys via Azure AI med Preview-status.",
"op": "O3",
"code": "LOCATOR_AMBIGUOUS",
"detail": {
"token": "5",
"hits": 8,
"window": {
"start": 38,
"end": 45
}
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#10",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 45,
"rule": "R2",
"claim": "Custom embeddings for domene-spesifikk fine-tuning har status Announced, med variabel dimensjonalitet og variabel maks token-kapasitet.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure",
"evidence_quote": "The following models are supported for fine-tuning: gpt-4o-mini (2024-07-18), gpt-4o (2024-08-06), gpt-4.1 (2025-04-14), gpt-4.1-mini (2025-04-14), gpt-4.1-nano (2025-04-14), o4-mini (2025-04-16), gpt-5 (2025-08-07), Ministral-3B (2411), Qwen-32B, Llama-3.3-70B-Instruct, gpt-oss-20b",
"reason": "Jeg sjekket den kanoniske listen over modeller som kan finjusteres — den inneholder ingen embedding-modell, så Custom embeddings for domenespesifikk fine-tuning finnes ikke som et Azure-tilbud med status Announced.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#13",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 60,
"rule": "R4",
"claim": "Azure OpenAI embeddings-kall i eksempelet bruker api_version 2024-02-01.",
"evidence_url": "https://learn.microsoft.com/azure/ai-foundry/openai/api-version-lifecycle",
"evidence_quote": "Azure OpenAI API version 2024-10-21 is currently the latest GA API release. This API version is the replacement for the previous 2024-06-01 GA API release.",
"reason": "2024-02-01 er en tidsstemplet, avløst rad: Learn oppgir at 2024-06-01 erstattet 2024-02-01, at 2024-10-21 nå er siste GA, og at v1-API-et anbefales — eksempelets api_version peker på en versjon som siden er skiftet ut.",
"op": "O3",
"code": "MULTI_REPLACEMENT",
"detail": {
"candidates": [
"2024-10-21",
"2024-06-01"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#15",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 247,
"rule": "R8",
"claim": "Copilot Studio kan kobles til Azure AI Search som knowledge source; embeddings genereres automatisk ved opplasting, og standardmodellen (typisk ada-002 eller text-embedding-3-small) kan ikke endres direkte i UI.",
"evidence_url": "https://learn.microsoft.com/microsoft-copilot-studio/knowledge-azure-ai-search",
"evidence_quote": "Copilot Studio supports vectorized indexes using integrated vectorization. Prepare your data and choose an embedded model, then use Import and vectorize data in Azure AI Search to create vector indexes.",
"reason": "Koblingen til Azure AI Search som knowledge source stemmer, men Learn krever at du selv velger embedding-modell og vektoriserer indeksen i Azure AI Search før tilkobling — embeddings genereres altså ikke automatisk ved opplasting, og modellvalget er ikke låst slik claimet påstår.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"002",
"3"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#18",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 289,
"rule": "R8",
"claim": "Azure AI Search støtter sletting av dokumenter, men ikke selektiv sletting av embeddings uten reindeksering.",
"evidence_url": "https://learn.microsoft.com/azure/search/search-howto-reindex",
"evidence_quote": "delete | Removes the entire document from the index. If you want to remove an individual field, use merge instead, setting the field in question to null.",
"reason": "Første del stemmer, men Learn dokumenterer at et enkeltfelt — inkludert et vektorfelt — fjernes selektivt med merge og null uten reindeksering, så den bærende begrensningen claimet påstår er motsagt.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#19",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 360,
"rule": "R1",
"claim": "Et batch-kall til embeddings-API-et kan sende ca. 100 dokumenter per API-kall, med maks 8191 tokens totalt.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/how-to/embeddings",
"evidence_quote": "The maximum input length for the current embedding models is 8,192 tokens. ... If you send an array of inputs in a single embedding request, the maximum array size is 2,048. ... Each /embeddings request has a 300,000-token aggregate limit across all inputs.",
"reason": "Taket claimet setter (maks 8191 tokens totalt per batch-kall) er avløst: 8 192 gjelder per input, mens Learns gjeldende aggregerte grense per forespørsel er 300 000 tokens, og maks array-størrelse er 2 048 — det oppgitte taket binder ikke lenger.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"100",
"8191"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#22",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 449,
"rule": "R2",
"claim": "Microsoft Foundry støtter fine-tuning av embedding-modeller via Custom Models, som er i preview.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure",
"evidence_quote": "The following models are supported for fine-tuning: gpt-4o-mini (2024-07-18), gpt-4o (2024-08-06), gpt-4.1 (2025-04-14), gpt-4.1-mini (2025-04-14), gpt-4.1-nano (2025-04-14), o4-mini (2025-04-16), gpt-5 (2025-08-07), Ministral-3B (2411), Qwen-32B, Llama-3.3-70B-Instruct, gpt-oss-20b",
"reason": "Jeg hentet den kanoniske listen over modeller som støtter fine-tuning i Foundry — den inneholder kun chat- og tekstmodeller, ingen embedding-modell, og ingen Custom Models-preview for embeddings.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "NO_SOURCE_STATUS"
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#23",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 453,
"rule": "R2",
"claim": "Fine-tuning av embeddings i Microsoft Foundry støtter modellene text-embedding-3-small | text-embedding-3-large, med treningsdata i JSON Lines-format bestående av query-document pairs.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure",
"evidence_quote": "The following models are supported for fine-tuning: gpt-4o-mini (2024-07-18), gpt-4o (2024-08-06), gpt-4.1 (2025-04-14), gpt-4.1-mini (2025-04-14), gpt-4.1-nano (2025-04-14), o4-mini (2025-04-16), gpt-5 (2025-08-07), Ministral-3B (2411), Qwen-32B, Llama-3.3-70B-Instruct, gpt-oss-20b",
"reason": "text-embedding-3-small og text-embedding-3-large står ikke i Learns liste over finjusterbare modeller, og verken JSON Lines-formatet eller query-document-par for embedding-finjustering er dokumentert.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#24",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 455,
"rule": "R2",
"claim": "Fine-tuning av embedding-modeller krever minimum 100 positive query-document-par, med 1000+ som anbefalt antall.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure",
"evidence_quote": "The following models are supported for fine-tuning: gpt-4o-mini (2024-07-18), gpt-4o (2024-08-06), gpt-4.1 (2025-04-14), gpt-4.1-mini (2025-04-14), gpt-4.1-nano (2025-04-14), o4-mini (2025-04-16), gpt-5 (2025-08-07), Ministral-3B (2411), Qwen-32B, Llama-3.3-70B-Instruct, gpt-oss-20b",
"reason": "Embedding-finjustering finnes ikke i den kanoniske listen over finjusterbare modeller, så kravet om minimum 100 positive query-document-par og anbefalingen om 1000+ har ingen dekning i Microsoft-dokumentasjonen.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"100",
"1000"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/embedding-models-selection.md#25",
"file": "skills/ms-ai-engineering/references/rag-architecture/embedding-models-selection.md",
"line": 457,
"rule": "R2",
"claim": "Evaluering av fine-tunede embedding-modeller i Microsoft Foundry skjer med metrikkene Recall@k | NDCG | MRR mot valideringssett.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure",
"evidence_quote": "The following models are supported for fine-tuning: gpt-4o-mini (2024-07-18), gpt-4o (2024-08-06), gpt-4.1 (2025-04-14), gpt-4.1-mini (2025-04-14), gpt-4.1-nano (2025-04-14), o4-mini (2025-04-16), gpt-5 (2025-08-07), Ministral-3B (2411), Qwen-32B, Llama-3.3-70B-Instruct, gpt-oss-20b",
"reason": "Foundry tilbyr ikke finjustering av embedding-modeller ifølge den kanoniske listen, og metrikkene Recall@k, NDCG og MRR er ikke dokumentert som evalueringsmetrikker for slike jobber noe sted på learn.microsoft.com.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#1",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 27,
"rule": "R2",
"claim": "Multi-layer caching i RAG dekker fire nivåer: result caching (hele LLM-responser) | retrieval caching (knowledge fragments fra vektorsøk) | embedding caching (forhåndsberegnede vektorrepresentasjoner) | semantic caching (semantisk like prompts).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/well-architected/ai/application-design#implement-multi-layer-caching-strategies",
"evidence_quote": "Result and answer caching: Use this approach to reuse responses for identical or semantically similar queries ... Retrieval and grounding snippet caching: Cache frequently retrieved knowledge fragments and grounding data ... Model output caching: Cache intermediate model outputs that can be reused across requests.",
"reason": "Den kanoniske WAF-siden ramser opp nøyaktig TRE cache-lag (result/answer, retrieval/grounding, model output); embedding caching er fraværende som eget nivå, semantisk caching er innbakt i result/answer-laget, og model output caching mangler i påstanden — fire-nivå-taksonomien stemmer ikke med kilden.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#2",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 29,
"rule": "R8",
"claim": "Microsoft-stakken tilbyr disse cache-tjenestene for AI-workloads: Azure Cache for Redis (tradisjonell og semantisk caching) | Azure Cosmos DB (semantisk cache med vektorsøk) | Azure AI Search (innebygd caching av søkeresultater) | Azure API Management (semantisk caching for LLM-API-er).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/search-pagination-page-layout",
"evidence_quote": "each query is independent and operates on the current view of the data as it exists in the index at query time (in other words, there's no caching or snapshot of results, such as those found in a general purpose database)",
"reason": "Redis-, Cosmos DB- og APIM-delene er dekket av Learn, men den bærende delen «Azure AI Search (innebygd caching av søkeresultater)» motsies direkte: Azure AI Search cacher ikke søkeresultater — den eneste dokumenterte cachen er enrichment cache (preview) for skillset-output i Azure Storage.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#3",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 37,
"rule": "R2",
"claim": "Multi-layer caching-tabellen lister fire cache-lag: Result caching (cache hele LLM-responser) | Retrieval caching (cache knowledge fragments fra vector search) | Embedding caching (cache forhåndsberegnede embeddings) | Model output caching (cache intermediate model outputs).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/well-architected/ai/application-design#implement-multi-layer-caching-strategies",
"evidence_quote": "Result and answer caching: ... Retrieval and grounding snippet caching: Cache frequently retrieved knowledge fragments and grounding data to avoid repeated database and search queries or data API operations. Model output caching: Cache intermediate model outputs that can be reused across requests.",
"reason": "Kilden lister tre cache-lag, ikke fire; «Embedding caching» finnes ikke i den kanoniske oppramsingen, så påstanden om fire lag er ikke dekket.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#5",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 68,
"rule": "R2",
"claim": "Cache invalidation utløses av fire triggere: data updates (webhook-triggered ved endring i kildedata) | model changes (ved model deployment/retraining) | prompt modifications (ved endring av prompt-template) | manual purge (admin-utløst for compliance eller testing).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/well-architected/ai/application-design#implement-multi-layer-caching-strategies",
"evidence_quote": "Invalidation hooks: Implement cache invalidation triggers for data updates, model changes, and prompt modifications.",
"reason": "Den kanoniske oppramsingen har nøyaktig tre invalideringstriggere; «manual purge (admin-utløst for compliance eller testing)» er fraværende, og søk fant ingen Learn-side som lister manuell purge som en fjerde trigger for AI-caching.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#7",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 186,
"rule": "R8",
"claim": "Azure Cache for Redis-tiers: Premium tier (99,9 % SLA, opptil 120 GB per shard) | Enterprise tier (99,99 % SLA, active-active geo-replikering, Flash-storage-støtte) | Enterprise Flash tier (opptil 13 TB cache-størrelse, 20 % RAM + 80 % NVMe Flash).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/azure-cache-for-redis/cache-overview",
"evidence_quote": "Memory: The Basic and Standard tiers offer 250 MB – 53 GB; the Premium tier 6 GB - 1.2 TB; the Enterprise tier 1 GB - 2 TB, and the Enterprise Flash tier 300 GB - 4.5 TB.",
"reason": "SLA-nivåene (99,9 % Premium / 99,99 % Enterprise) og Premium-kapasiteten holder, men den bærende delen «Enterprise Flash opptil 13 TB» motsies: siden oppgir 300 GB – 4,5 TB for Enterprise Flash.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#11",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 237,
"rule": "R7",
"claim": "APIM-policyene for semantisk cache er azure-openai-semantic-cache-lookup (inbound, med attributtene score-threshold, embeddings-backend-id, embeddings-backend-auth, ignore-system-messages, max-message-count og underelementet vary-by) og azure-openai-semantic-cache-store (outbound, med attributtet duration).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/api-management/llm-semantic-cache-lookup-policy",
"evidence_quote": "Use the llm-semantic-cache-lookup policy to perform cache lookup of responses to large language model (LLM) API requests from a configured external cache, based on vector proximity of the prompt to previous requests and a specified score threshold.",
"reason": "Attributtene og vary-by-underelementet stemmer, men policy-navnene er superseded: azure-openai-semantic-cache-lookup/-store heter nå llm-semantic-cache-lookup og llm-semantic-cache-store (den gamle URL-en redirigerer til llm-siden, og eksempel-XML-en på how-to-siden bruker llm-navnene). Navnet ER påstanden (R7s load-bearing-carve-out) — en leser kopierer strengen inn i policy-XML.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#12",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 253,
"rule": "R8",
"claim": "APIM-attributtet score-threshold er en distanse der lavere verdi gir strengere matching og krever høyere semantisk likhet: 0,1–0,2 = strict matching | 0,3–0,5 = balanced | 0,6–0,8 = liberal matching.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/api-management/llm-semantic-cache-lookup-policy",
"evidence_quote": "Score threshold above 0.2 may lead to cache mismatch. Consider using lower value for sensitive use cases.",
"reason": "Retningsdelen er grunngitt («Lower values require higher semantic similarity for a match»), men den bærende tre-bånds-rubrikken er ikke dokumentert og strider mot Learn-veiledningen: doc-en anbefaler å starte på 0,05 og advarer om cache-mismatch over 0,2, mens påstanden kaller 0,3–0,5 «balanced» og 0,6–0,8 «liberal».",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#14",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 296,
"rule": "R8",
"claim": "Azure Cosmos DB som semantisk cache gir: global distribusjon med multi-region writes | automatisk indeksering av vektorer | 99,999 % SLA med multi-region-oppsett | innebygd TTL-støtte.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/cosmos-db/how-to-dotnet-vector-index-query",
"evidence_quote": "After you decide on the vector embedding paths, you must add vector indexes to the indexing policy. ... The vector path is added to the excludedPaths section of the indexing policy to ensure optimized performance for insertion.",
"reason": "Multi-region writes, 99,999 % og TTL er grunngitt, men den bærende delen «automatisk indeksering av vektorer» motsies: vektorindeks må defineres eksplisitt i indekseringspolicyen (kun ved container-opprettelse), og vektorstien legges i excludedPaths — altså holdes vektorer utenfor den automatiske indekseringen.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#18",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 361,
"rule": "(none)",
"claim": "Microsoft Entra ID-autentisering for Redis er i preview.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/redis/entra-for-authentication",
"evidence_quote": "Azure Managed Redis offers a password-free authentication mechanism by integrating with Microsoft Entra ID. Azure Managed Redis caches use Microsoft Entra ID by default. When you create a new cache, managed identity is enabled.",
"reason": "Status-påstanden er utdatert: Entra ID-autentisering er ikke merket preview noe sted på den kanoniske siden (som derimot merker underfunksjonen «Configure custom data access permissions (preview)»), den siterte quickstart-en sier «Microsoft Entra ID is enabled by default», og What's New oppgir at Entra ID for autentisering og RBAC ble tilgjengelig i alle regioner i juni 2023.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "NO_SOURCE_STATUS"
}
},
{
"id": "ms-ai-engineering/rag-architecture/rag-caching-optimization.md#19",
"file": "skills/ms-ai-engineering/references/rag-architecture/rag-caching-optimization.md",
"line": 373,
"rule": "R8",
"claim": "Azure Cache for Redis SKU-er med kapasitet: Basic C0 (250 MB, ingen SLA) | Standard C1 (1 GB, 2 replicas, 99,9 % SLA) | Premium P1 (6 GB, clustering, geo-replikering) | Enterprise E10 (12 GB, active-active, 99,99 % SLA) | Enterprise Flash F300 (345 GB, 20 % RAM + 80 % Flash).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/azure-cache-for-redis/cache-best-practices-performance",
"evidence_quote": "| F300 | 384 GB | 8 | 3,200 | 500,000 | 390,000 |",
"reason": "E10 = 12 GB stemmer, men den bærende størrelsen for Enterprise Flash er feil: F300 er 384 GB, ikke 345 GB. I tillegg beskriver Learn Standard-tieren som «An OSS Redis cache running on two VMs in a replicated configuration» (primary + én replica), ikke 2 replicas.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#1",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 4,
"rule": "R8",
"claim": "Hybrid search i Azure AI Search er GA, mens scalar quantization er i Preview.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-index-size",
"evidence_quote": "Vector optimization techniques are generally available. Use capabilities like narrow data types, scalar and binary quantization, and elimination of redundant storage to reduce your vector quota and storage quota consumption.",
"reason": "Hybrid search som GA holder, men den andre lastbærende delen svikter: Learn oppgir scalar quantization som generelt tilgjengelig (GA), ikke preview.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "NO_COMPLETE_FILE_LABEL"
}
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#2",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 37,
"rule": "R2",
"claim": "Azure AI Search tilbyr vektoralgoritmene Hierarchical NSW (HNSW, approximate nearest neighbor) | Exhaustive KNN (exact nearest neighbor) | Flat indexing (linear scan).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-create-index",
"evidence_quote": "vectorSearch.algorithms is either hnsw or exhaustiveKnn. These are the Approximate Nearest Neighbors (ANN) algorithms used to organize vector content during indexing.",
"reason": "Den kanoniske opplistingen av vektoralgoritmer nevner kun hnsw og exhaustiveKnn; «Flat indexing (linear scan)» finnes ikke som algoritme i Azure AI Search.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#5",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 71,
"rule": "R2",
"claim": "Vekting av hybrid scores i Azure AI Search styres av parametrene alpha (balanse mellom vector 1.0 og BM25 0.0, standard 0.5, range 0.0–1.0) | k (antall vektorer fra vector search, standard 50, range 1–1000) | top (totale resultater etter merge, standard 10, range 1–1000).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/hybrid-search-ranking",
"evidence_quote": "If you add vector weighting, the initial scores are subject to a weighting multiplier that increases or decreases the score. The default is 1.0, which means no weighting and the initial score is used as-is in RRF scoring.",
"reason": "Vekting i hybrid search skjer med RRF og parameteren weight (standard 1.0) — noen alpha-parameter finnes ikke i den kanoniske dokumentasjonen, og top har standardverdi 50, ikke 10 («$top ... This defaults to 50» i Search Documents REST).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#8",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 168,
"rule": "R2",
"claim": "Azure AI Search har en rate limit på 3000 requests per sekund per replika.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/search-limits-quotas-capacity",
"evidence_quote": "| Search queries (POST /indexes/{index}/docs/search) | Varies by SU count and query complexity | 50 queries/sec (aggregate read throttle per index) |",
"reason": "Den kanoniske throttling-tabellen oppgir ingen grense på 3000 requests per sekund per replika — søkespørringer varierer med antall search units og spørringskompleksitet.",
"op": "O3",
"code": "CONTEXT_MISMATCH",
"detail": {
"token": "3000",
"replacement": "50",
"would_have_been": "**Viktig:** Azure AI Search har rate limits (50 requests/sekund per replika). Høy-volum indexing krever skalering av replicas eller bruk av push-pattern via indexer."
}
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#10",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 235,
"rule": "(none)",
"claim": "Scalar quantization i Azure AI Search komprimerer vektorer fra 1536 float32 (6 KB) til 384 int8 (384 bytes).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-quantization",
"evidence_quote": "Scalar quantization compresses float values into narrower data types. AI Search currently supports int8, which is 8 bits, reducing vector index size fourfold.",
"reason": "Scalar quantization reduserer datatypen (float32 til int8), ikke antall dimensjoner: 1536 float32 (6 144 byte) blir 1536 int8 (én firedel), ikke 384 int8 / 384 byte — dimensjonsreduksjon er en separat MRL-funksjon (truncationDimension).",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"1536",
"32",
"6",
"384",
"8"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#11",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 237,
"rule": "(none)",
"claim": "Scalar quantization er en preview-funksjon i Azure AI Search (2026).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-index-size",
"evidence_quote": "Vector optimization techniques are generally available. Use capabilities like narrow data types, scalar and binary quantization, and elimination of redundant storage to reduce your vector quota and storage quota consumption.",
"reason": "Scalar quantization er GA (og står ikke i Learn sin liste over preview-funksjoner i Azure AI Search), så statuspåstanden «preview» er motsagt.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "2026",
"type": "number"
}
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#12",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 318,
"rule": "R8",
"claim": "Copilot Studio konfigurerer Azure AI Search som kunnskapskilde for Generative answers via Security & Data → Knowledge sources → Add Azure AI Search, uten kontroll over HNSW-parametere (managed service).",
"evidence_url": "https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-azure-ai-search",
"evidence_quote": "Select Add knowledge from either the Overview or Knowledge pages, or the Properties of a generative answers node. From the Add knowledge dialog, select Featured. Select Azure AI Search.",
"reason": "Den dokumenterte navigasjonen er Add knowledge → Featured → Azure AI Search (alternativt Data sources → Azure AI Search), ikke «Security & Data → Knowledge sources → Add Azure AI Search»; den lastbærende stien er feil.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#13",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 324,
"rule": "R2",
"claim": "AI Builder i Power Automate støtter semantic search via Azure AI Search-connectoren med handlingen «Search documents (semantic)», som krever en index med contentVector-felt.",
"evidence_url": "https://learn.microsoft.com/en-us/connectors/azureaisearch/",
"evidence_quote": "| Agentic Search (Preview) | Delete a document | Delete multiple documents | Get index statistics | Get search indexes | Index a document | Index multiple documents | Merge a document | Search vectors | Search vectors with natural language | Semantic Hybrid Search |",
"reason": "Den kanoniske opplistingen av handlinger i Azure AI Search-connectoren inneholder ingen handling som heter «Search documents (semantic)» (nærmeste er «Semantic Hybrid Search»), og connectoren stiller ingen krav om et felt ved navn contentVector.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#16",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 368,
"rule": "R4",
"claim": "Azure AI Search tilbys i tierne Basic (2 GB storage, 3 queries/sek, 1 replica) | S1 (25 GB, 15 queries/sek, 3 replicas) | S2 (100 GB, 60 queries/sek, 6 replicas) | S3 (200 GB, 60 queries/sek, 12 replicas).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/search-limits-quotas-capacity",
"evidence_quote": "| After May 17, 2024 ^2^ | 15 | 160 | 512 | 1,024 | 2,048 | 4,096 | N/A |",
"reason": "Lagringstallene 2/25/100/200 GB matcher kun den daterte legacy-raden «Before April 3, 2024»; gjeldende grenser er 15/160/512/1024 GB, og replika-tallene er også feil (Basic 3, S1–S3 12, ikke 1/3/6/12).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#17",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 384,
"rule": "R8",
"claim": "Semantic Ranker er inkludert i S1 og høyere tiers, med et gratis usage cap på 50 000 queries per måned.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/semantic-how-to-enable-disable",
"evidence_quote": "| Free (default) | Provides a monthly free request allowance. After the free allowance is consumed, semantic ranker requests return a billing error. | Available on all pricing tiers. |",
"reason": "Gratisplanen for semantic ranker er tilgjengelig på alle pristiers (standardplanen krever Basic eller høyere), ikke bare S1 og oppover, og ingen Learn-side oppgir et gratistak på 50 000 spørringer per måned.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"1",
"50",
"000"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#18",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 385,
"rule": "(none)",
"claim": "Integrasjon med Copilot Studio krever Copilot Studio-lisens med 20 000 meldinger per måned.",
"evidence_url": "https://learn.microsoft.com/en-us/power-platform/admin/powerapps-flow-licensing-faq",
"evidence_quote": "Licensed by tenant, Microsoft Copilot Studio entitles a tenant to 25,000 messages per month.",
"reason": "Lisensen gir 25 000 meldinger per tenant per måned, ikke 20 000.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"20",
"000"
]
}
},
{
"id": "ms-ai-engineering/rag-architecture/vector-indexing-techniques.md#19",
"file": "skills/ms-ai-engineering/references/rag-architecture/vector-indexing-techniques.md",
"line": 433,
"rule": "R7",
"claim": "I hybrid search-indekser er HNSW standard ANN-algoritme med efSearch og maxConnections som tunbare parametere, mens eKNN (exhaustive K-Nearest Neighbors) gir fullstendig søk og aktiveres med \"exhaustive\": true i spørringen.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-create-index",
"evidence_quote": "vectorSearch.algorithms.m is the bi-directional link count. Default is 4. The range is 4 to 10.",
"reason": "HNSW-parameteren for koblinger heter m, ikke maxConnections — navnet er selve påstanden (lastbærende API-identifikator), så R7 sin leniens for illustrerende tall gjelder ikke, selv om efSearch og \"exhaustive\": true stemmer.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/bias-detection-mitigation-strategies.md#10",
"file": "skills/ms-ai-governance/references/responsible-ai/bias-detection-mitigation-strategies.md",
"line": 121,
"rule": "R8",
"claim": "Default-konfigurasjonen i Azure AI Content Safety er Medium severity threshold (blokkerer medium og høyere) for både text models og image models — samme terskel for alle fire kategorier.",
"evidence_url": "https://learn.microsoft.com/azure/foundry-classic/concepts/model-catalog-content-safety",
"evidence_quote": "For image models, the default content filtering configuration is set at the low configuration threshold, filtering at this level or higher.",
"reason": "Medium-terskelen gjelder tekstmodeller, men claimens andre bærende del - samme Medium-terskel for image models - motsies: bildemodeller har Low som default.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/bias-detection-mitigation-strategies.md#14",
"file": "skills/ms-ai-governance/references/responsible-ai/bias-detection-mitigation-strategies.md",
"line": 194,
"rule": "R2",
"claim": "Azure ML-komponentregisteret inneholder RAI-fairness-komponenten `microsoft_azureml_rai_tabular_fairness`, som hentes med label «latest».",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-responsible-ai-insights-sdk-cli?view=azureml-api-2",
"evidence_quote": "microsoft_azureml_rai_tabular_insight_constructor ... microsoft_azureml_rai_tabular_causal ... microsoft_azureml_rai_tabular_counterfactual ... microsoft_azureml_rai_tabular_erroranalysis ... microsoft_azureml_rai_tabular_explanation ... microsoft_azureml_rai_tabular_insight_gather",
"reason": "Den kanoniske komponentsiden lister alle RAI-komponentene i azureml-registeret, og microsoft_azureml_rai_tabular_fairness er ikke blant dem - fairness beregnes av selve konstruktør-/model overview-komponenten.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/bias-detection-mitigation-strategies.md#17",
"file": "skills/ms-ai-governance/references/responsible-ai/bias-detection-mitigation-strategies.md",
"line": 303,
"rule": "R7",
"claim": "Databricks tilbyr DataQualityMonitor.create for fairness-/bias-overvåkning av inference-logger med parametrene table_name, inference_log, problem_type og slicing_exprs, og beregner automatisk metrikkene predictive_parity, predictive_equality og equal_opportunity.",
"evidence_url": "https://learn.microsoft.com/azure/databricks/data-governance/unity-catalog/data-quality-monitoring/data-profiling/create-monitor-api",
"evidence_quote": "For information about the deprecated quality_monitors API, see Create a data profile using the quality_monitors API (deprecated).",
"reason": "Claimens bærende SDK-identifikator DataQualityMonitor.create finnes ikke: gjeldende API er w.data_quality (create-monitor), og den gamle var quality_monitors - metrikkdelen stemmer, men navnet gjør at en leser ville kopiert feil kall.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/bias-detection-mitigation-strategies.md#18",
"file": "skills/ms-ai-governance/references/responsible-ai/bias-detection-mitigation-strategies.md",
"line": 409,
"rule": "R2",
"claim": "Responsible AI Dashboard-komponentene har SDK-metodene RAIInsights.from_model() (oppretter dashboard) | add_fairness() | add_error_analysis() | add_explainer() | add_counterfactual() | add_causal().",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-responsible-ai-insights-sdk-cli?view=azureml-api-2",
"evidence_quote": "rai_causal_component = ml_client_registry.components.get(name=microsoft_azureml_rai_tabular_causal, label=latest)",
"reason": "Den kanoniske SDK-siden dokumenterer komponentkall via ml_client_registry.components.get, og nevner RAIInsights kun som objektet i dashboard-porten; ingen Learn-side navngir metodene RAIInsights.from_model, add_fairness, add_error_analysis, add_explainer, add_counterfactual eller add_causal.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/bias-detection-mitigation-strategies.md#20",
"file": "skills/ms-ai-governance/references/responsible-ai/bias-detection-mitigation-strategies.md",
"line": 503,
"rule": "R8",
"claim": "Custom categories i Azure AI Content Safety er i preview og trenes via Content Safety Studio.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview",
"evidence_quote": "In Content Safety Studio, the following Azure AI Content Safety features are available: Moderate Text Content ... Moderate Image Content ... Monitor Online Activity",
"reason": "Preview-statusen stemmer, men den andre bærende delen gjør det ikke: Content Safety Studios dokumenterte funksjonsliste inneholder ikke custom categories, og kvikkstarten trener dem via REST-API eller Azure AI Foundry sin Guardrails + controls-fane.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "NO_SOURCE_STATUS"
}
},
{
"id": "ms-ai-governance/responsible-ai/bias-detection-mitigation-strategies.md#22",
"file": "skills/ms-ai-governance/references/responsible-ai/bias-detection-mitigation-strategies.md",
"line": 575,
"rule": "R3",
"claim": "AI Builder-modelltypene omfatter Form Processing | Text Classification | Prediction | Object Detection.",
"evidence_url": "https://learn.microsoft.com/ai-builder/custom-overview",
"evidence_quote": "Document processing: Extract custom information from documents. Category classification: Classify texts into custom categories. Entity extraction ... Prediction ... Object detection ... Azure machine learning models",
"reason": "To av de fire navnene er erstattet i gjeldende taksonomi: Form Processing heter nå Document processing og Text Classification heter Category classification.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/bias-detection-mitigation-strategies.md#27",
"file": "skills/ms-ai-governance/references/responsible-ai/bias-detection-mitigation-strategies.md",
"line": 939,
"rule": "(none)",
"claim": "Responsible AI dashboard (Azure Machine Learning) har status GA.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-responsible-ai-insights-ui?view=azureml-api-2",
"evidence_quote": "Select Create Responsible AI dashboard (preview).",
"reason": "Gjeldende dokumentasjon merker Responsible AI dashboard som public preview (også i AutoML-artikkelen), så statusen GA motsies.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "NO_COMPLETE_FILE_LABEL"
}
},
{
"id": "ms-ai-governance/responsible-ai/content-safety-implementation.md#1",
"file": "skills/ms-ai-governance/references/responsible-ai/content-safety-implementation.md",
"line": 37,
"rule": "R8",
"claim": "Content Safety feature statuses: Analyze Text GA | Analyze Image GA | Prompt Shields GA | Groundedness Detection Preview | Protected Material Text GA | Protected Material Code GA | Custom Categories (Standard) Preview | Custom Categories (Rapid) Preview | Blocklists GA",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/content-safety/whats-new",
"evidence_quote": "Protected material detection for code (preview)",
"reason": "The claim marks Protected Material Code as GA, but What's new and the current quickstart both title it '(preview)' with no GA announcement (the Aug 2024 GA covered only Prompt Shields and Protected Material for text), so one load-bearing status in the bundle is wrong while the other statuses match.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/content-safety-implementation.md#2",
"file": "skills/ms-ai-governance/references/responsible-ai/content-safety-implementation.md",
"line": 37,
"rule": "R8",
"claim": "Content Safety input limits: Analyze Text max 10K chars | Analyze Image JPEG/PNG/GIF/BMP/TIFF/WEBP max 4MB | Prompt Shields text max 10K chars | Groundedness query+sources max 55K chars | Protected Material Text min 110 chars",
"evidence_url": "https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview",
"evidence_quote": "Maximum length for grounding sources: 55,000 characters (per API call). Maximum text and query length: 7,500 characters.",
"reason": "The claim states 'Groundedness query+sources max 55K chars', but the live page gives 55,000 characters to grounding sources only and caps text/query separately at 7,500 characters, so that load-bearing limit is misstated even though the other limits (Analyze Text 10K, image 4MB JPEG/PNG/GIF/BMP/TIFF/WEBP, Prompt Shields 10K, PM-Text min 110) match.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/content-safety-implementation.md#6",
"file": "skills/ms-ai-governance/references/responsible-ai/content-safety-implementation.md",
"line": 239,
"rule": "R8",
"claim": "Copilot Studio cannot (per feb 2026) configure severity levels per category - uses Azure OpenAI deployment settings; custom blocklists can be enabled in Agent Settings",
"evidence_url": "https://learn.microsoft.com/microsoft-copilot-studio/knowledge-copilot-studio",
"evidence_quote": "The moderation levels range from **Lowest** to **Highest**. The lowest level generates the most answers, but they might contain harmful content. The highest level of content moderation generates fewer answers, and applies a stricter filter to restrict harmful content. The default moderation level is **High**.",
"reason": "The claim says moderation 'uses Azure OpenAI deployment settings' and that custom blocklists can be enabled in Agent Settings, but the live docs show content moderation is configured inside Copilot Studio itself (agent-, topic-, and prompt-level Lowest-Highest levels) and the Agent settings enumeration has a Moderation level dropdown with no custom-blocklist option, so two load-bearing parts are wrong (only the no-per-category-severity part holds).",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "2026",
"type": "number"
}
},
{
"id": "ms-ai-governance/responsible-ai/content-safety-implementation.md#7",
"file": "skills/ms-ai-governance/references/responsible-ai/content-safety-implementation.md",
"line": 243,
"rule": "R3",
"claim": "AI Builder Text generation uses Azure OpenAI with content filtering enabled by default and no configuration options (default Medium+High block)",
"evidence_url": "https://learn.microsoft.com/ai-builder/prebuilt-azure-openai",
"evidence_quote": "This feature is deprecated and isn't visible anymore.",
"reason": "AI Builder's Text generation model is deprecated and replaced by prompt builder, where per the prompts FAQ 'Makers can configure the content moderation level for harmful content only' - so the claim's frame (no configuration options, implicit Medium+High default) describes a replaced world and its 'no configuration' part is contradicted for the current mechanism.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/content-safety-implementation.md#8",
"file": "skills/ms-ai-governance/references/responsible-ai/content-safety-implementation.md",
"line": 259,
"rule": "R8",
"claim": "Microsoft 365 Copilot has its own content filtering policies that are not customer-configurable (severity levels cannot be adjusted; Microsoft-managed)",
"evidence_url": "https://learn.microsoft.com/microsoft-365/copilot/harmful-content-protection-copilot-chat",
"evidence_quote": "For these special use cases, your organization's Microsoft 365 Apps administrator can enable a specific group of users to adjust harmful content protection settings in their Copilot Chat experiences.",
"reason": "The claim's load-bearing 'not customer-configurable' is contradicted by the live page: admins can policy-enable users to toggle harmful content protection off in Copilot Chat (and the privacy doc says Microsoft 'offers certain content filtering controls for admins and users'), even though severity-level sliders indeed do not exist.",
"op": "O3",
"code": "LOCATOR_AMBIGUOUS",
"detail": {
"token": "365",
"hits": 2,
"window": {
"start": 258,
"end": 261
}
}
},
{
"id": "ms-ai-governance/responsible-ai/content-safety-implementation.md#9",
"file": "skills/ms-ai-governance/references/responsible-ai/content-safety-implementation.md",
"line": 275,
"rule": "R2",
"claim": "Azure AI Content Safety available in West Europe and Norway East (via Azure OpenAI); models run in EU (no data transfer to USA)",
"evidence_url": "https://learn.microsoft.com/en-us/azure/ai-services/content-safety/overview",
"evidence_quote": "To use the Content Safety APIs, you must create your Azure AI Content Safety resource in a supported region. Currently, the Content Safety features are available in the following Azure regions with different API versions:",
"reason": "The canonical region-availability table on the cited page lists West Europe but contains no Norway East row for any Content Safety API, and no Learn page found states Content Safety availability in Norway East (even via Azure OpenAI), so the Norway East part fails the enumeration check; the in-region/no-US-transfer part alone is supported by the data-privacy page.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/content-safety-implementation.md#10",
"file": "skills/ms-ai-governance/references/responsible-ai/content-safety-implementation.md",
"line": 270,
"rule": "R8",
"claim": "Azure AI Content Safety has PII detection for completions (names, addresses; Norwegian national ID format not officially supported), configurable to block or mask PII in LLM output",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/language-service/personally-identifiable-information/concepts/entity-categories",
"evidence_quote": "To retrieve this entity type, specify **NOIdentityNumber** in the **piiCategories** request parameter. If detected, the entity appears in the **PII** response payload.",
"reason": "The PII filter page links this entity-categories list as 'the complete list of supported personal data entity types' and lists 'National ID numbers (50+ countries)' including a dedicated Norway Identity Number (NOIdentityNumber) type, contradicting the claim's load-bearing assertion that the Norwegian national ID format is not officially supported (the detection-in-completions and block/redacted_text parts do hold).",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#1",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 48,
"rule": "R8",
"claim": "Multistage- og AI-approvals i Power Automate/Copilot Studio er i Preview, og det finnes en ny 'Human in the loop'-kobling samt conditions mellom stages for dynamisk routing.",
"evidence_url": "https://learn.microsoft.com/en-us/microsoft-copilot-studio/flows-advanced-approvals",
"evidence_quote": "In your flow, you can add *Run a multistage approval* as an action via the new *Human review* connector.",
"reason": "Preview-statusen og conditions mellom stages stemmer, men den nye koblingen heter *Human review*, ikke 'Human in the loop' — en lastbærende, navngitt streng leseren ville søkt etter i handlingslisten.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "NO_SOURCE_STATUS"
}
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#2",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 48,
"rule": "R8",
"claim": "AI-stage i Power Automate-approvals bruker GPT-o3 til å gjøre Approve/Reject med begrunnelse.",
"evidence_url": "https://learn.microsoft.com/en-us/microsoft-copilot-studio/faqs-ai-approvals",
"evidence_quote": "AI approvals support GPT 4.1 mini, GPT 4.o, GPT 4.1, and o3 models, which are hosted on Azure OpenAI Service.",
"reason": "Approve/Reject med begrunnelse stemmer, men modellen er ikke låst til GPT-o3: fire modeller støttes via en model picker, og FAQ-en sier GPT-4.1 er typisk ideell — den lastbærende versjonsdelen holder ikke.",
"op": "O3",
"code": "MULTI_REPLACEMENT",
"detail": {
"candidates": [
"4.1",
"4"
]
}
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#5",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 55,
"rule": "R2",
"claim": "Godkjenningstypene i approval-flyter er: First to respond | Everyone must approve | Conditional approvals | Multistage.",
"evidence_url": "https://learn.microsoft.com/en-us/power-automate/all-assigned-must-approve",
"evidence_quote": "Approve/Reject - Everyone must approve | Approve/Reject - First to respond | Custom responses - Wait for all responses | Custom responses - Wait for one response",
"reason": "Den kanoniske tabellen 'Approval types and their behaviors' lister fire typer, og verken 'Conditional approvals' eller 'Multistage' står der — de er egenskaper ved multistage-flyten, ikke godkjenningstyper; de to Custom responses-typene mangler i claimet.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#7",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 90,
"rule": "R7",
"claim": "Microsoft Agent Framework har approval modes for funksjoner: `never` (default) | `always_require` | `confidence_based`.",
"evidence_url": "https://learn.microsoft.com/agent-framework/integrations/ag-ui/human-in-the-loop",
"evidence_quote": "1. always_require: Always request approval before execution 2. never_require: Never request approval (default behavior) 3. conditional: Request approval based on certain conditions (custom logic)",
"reason": "Enum-verdiene er lastbærende strenger: default-modusen heter never_require (ikke 'never'), og den tredje modusen heter conditional — 'confidence_based' finnes ikke i API-et (Literal['always_require', 'never_require']).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#8",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 128,
"rule": "R2",
"claim": "Sikkerhetskravet i AI-5.1 krever kryptering av review-systemer med TLS 1.2 eller nyere.",
"evidence_url": "https://learn.microsoft.com/en-us/security/benchmark/azure/mcsb-v2-artificial-intelligence-security#ai-5-ensure-human-in-the-loop",
"evidence_quote": "Secure HITL interfaces: Protect review systems with encryption, implement strict access controls using Microsoft Entra ID, and deploy anomaly detection to prevent tampering or unauthorized access to approval processes.",
"reason": "AI-5.1s sikkerhetskrav sier 'encryption' generelt; ingen TLS-versjon er angitt noe sted i kontrollen — kravet om 'TLS 1.2 eller nyere' er lagt til av claimet.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"5.1",
"1.2"
]
}
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#9",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 310,
"rule": "R2",
"claim": "Azure AI Security Benchmark AI-5 definerer fem kriterier som alltid krever HITL: External data transfers | Processing of confidential information | Decisions impacting financial outcomes | Safety-related commands | Compliance-critical processes.",
"evidence_url": "https://learn.microsoft.com/en-us/security/benchmark/azure/mcsb-v2-artificial-intelligence-security#ai-5-ensure-human-in-the-loop",
"evidence_quote": "Define critical actions: Identify high-risk AI operations requiring human review such as external data transfers, processing of confidential information, or decisions impacting financial or operational outcomes, using risk assessments to prioritize review pathways.",
"reason": "AI-5 definerer ikke fem kriterier som 'alltid krever HITL', men gir tre eksempler innledet med 'such as'; 'Safety-related commands' og 'Compliance-critical processes' står ikke i oppregningen, og 'operational outcomes' er utelatt.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#10",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 349,
"rule": "R2",
"claim": "AI-5.1 sine opplæringskrav for reviewere omfatter fem punkter: AI system behavior | Potential vulnerabilities | Domain-specific risks | Decision-support tools | Escalation procedures.",
"evidence_url": "https://learn.microsoft.com/en-us/security/benchmark/azure/mcsb-v2-artificial-intelligence-security#ai-5-ensure-human-in-the-loop",
"evidence_quote": "Train reviewers: Equip personnel with training on AI system behavior, potential vulnerabilities (e.g., adversarial inputs), and domain-specific risks, providing access to contextual data and decision-support tools to enable informed validation.",
"reason": "Opplæringskravet omfatter tre temaer pluss tilgang til kontekstdata og beslutningsstøtteverktøy — 'Escalation procedures' finnes ikke i AI-5.1, så den femte oppføringen er uten dekning.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#11",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 356,
"rule": "R2",
"claim": "For å forhindre reviewer fatigue skal en person ikke review mer enn 50 AI-beslutninger per dag.",
"evidence_url": "https://learn.microsoft.com/en-us/security/benchmark/azure/mcsb-v2-artificial-intelligence-security#ai-5-ensure-human-in-the-loop",
"evidence_quote": "Optimize review processes: Implement selective HITL reviewing only low-confidence AI outputs or high-impact decisions to balance security with operational efficiency, regularly assessing workflows to prevent reviewer fatigue and maintain effectiveness.",
"reason": "Kontrollen som faktisk adresserer reviewer fatigue foreskriver selektiv review og jevnlig vurdering av arbeidsflyten, ikke noe tak; grensen på 50 beslutninger per dag står ingen steder.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "50",
"type": "number"
}
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#13",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 396,
"rule": "R2",
"claim": "Copilot Studio sine HITL-features er: Human Handoff Topic som overfører samtalen til Live Agent (Omnichannel, Dynamics 365) | Escalation Rate Tracking i analytics dashboard | Rationale Generation | Approval Topics som pauser for menneskelig input.",
"evidence_url": "https://learn.microsoft.com/en-us/microsoft-copilot-studio/guidance/deflection-topic-escalation-analysis",
"evidence_quote": "The direct way to initiate an escalation to a human representative is through the **Escalate** system topic. ... Another way to trigger this escalation is through the **Transfer Conversation** node in the authoring canvas.",
"reason": "Escalate-systemtopic mot Omnichannel og 'Escalation Rate Drivers' i analytics stemmer, men den kanoniske gjennomgangen kjenner verken 'Approval Topics' eller 'Rationale Generation' — godkjenninger i Copilot Studio er stages i agent flows via Human review-koblingen, ikke topics.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#15",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 421,
"rule": "R8",
"claim": "FAQ for AI Approvals anbefaler GPT-4.1 for komplekse approval-scenarioer, og o3 for advanced reasoning men da tregere.",
"evidence_url": "https://learn.microsoft.com/en-us/microsoft-copilot-studio/faqs-ai-approvals",
"evidence_quote": "GPT-4.1 is typically ideal for most approval scenarios. Advanced reasoning models (like O3) might handle complex logic better but are slower.",
"reason": "O3-delen stemmer, men FAQ-en anbefaler GPT-4.1 for de fleste scenarioene og tilskriver nettopp kompleks logikk til O3 — claimet snur dette og gir GPT-4.1 de komplekse approval-scenarioene.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"4.1",
"3"
]
}
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#21",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 604,
"rule": "R3",
"claim": "AI Approvals er inkludert i Copilot Studio-lisensen (per-user-lisens + AI credits) og forbruker AI credits ved bruk.",
"evidence_url": "https://learn.microsoft.com/ai-builder/administer-licensing",
"evidence_quote": "Agents and agent flows only consume Copilot Credits.",
"reason": "Valutaen ble byttet fra messages til Copilot Credits 1. september 2025, og AI Builder credits er en egen valuta som agent flows nettopp ikke bruker — enheten claimet bygger på er erstattet.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#22",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 628,
"rule": "(none)",
"claim": "Power Automate Premium kreves for approvals.",
"evidence_url": "https://learn.microsoft.com/power-automate/get-started-approvals",
"evidence_quote": "Because the approvals connector is a standard connector, any license that grants access to Power Automate and the ability to use standard connectors is sufficient to create approval flows.",
"reason": "Kilden sier det motsatte: approvals er en standardkobling, og Office 365- eller Dynamics 365-lisenser holder — Power Automate Premium er ikke et krav for approvals.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/human-in-the-loop-oversight.md#23",
"file": "skills/ms-ai-governance/references/responsible-ai/human-in-the-loop-oversight.md",
"line": 637,
"rule": "R8",
"claim": "Microsoft Agent Framework har ingen direkte lisenskostnad (open source), men krever Azure OpenAI eller Microsoft Foundry for modellene.",
"evidence_url": "https://learn.microsoft.com/agent-framework/overview/",
"evidence_quote": "Supports Microsoft Foundry, Anthropic, Azure OpenAI, OpenAI, Ollama, and more.",
"reason": "Open source-delen står seg, men den lastbærende påstanden om at Azure OpenAI eller Foundry kreves motsies: rammeverket støtter også OpenAI direkte, Anthropic og lokal Ollama-kjøring.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/stakeholder-communication-ai-decisions.md#2",
"file": "skills/ms-ai-governance/references/responsible-ai/stakeholder-communication-ai-decisions.md",
"line": 91,
"rule": "R8",
"claim": "Model Interpretability gir tre nivåer av forklaring med tilhørende Microsoft-verktøy: Global explanations (Azure ML Interpretability component) | Local explanations (Counterfactual What-If) | Cohort explanations (Responsible AI Dashboard)",
"evidence_url": "https://learn.microsoft.com/en-us/azure/machine-learning/how-to-machine-learning-interpretability?view=azureml-api-2",
"evidence_quote": "It provides multiple views into a model's behavior: Global explanations: For example, what features affect the overall behavior of a loan allocation model? Local explanations: For example, why was a customer's loan application approved or rejected? You can also observe model explanations for a selected cohort as a subgroup of data points.",
"reason": "De tre nivåene stemmer, men verktøy-tilordningen er feil: siden legger BÅDE lokale og kohort-forklaringer til selve interpretability-komponenten, mens counterfactual what-if er en separat DiCE-basert komponent for feature-perturbasjoner — to bærende deler er feilkoblet.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/stakeholder-communication-ai-decisions.md#4",
"file": "skills/ms-ai-governance/references/responsible-ai/stakeholder-communication-ai-decisions.md",
"line": 170,
"rule": "R2",
"claim": "Microsofts governance-struktur for AI i organisasjoner består av et AI Governance Board (executive sponsorship) over tre enheter: AI Center of Excellence | Cross-Functional Governance Team | Incident Response Team",
"evidence_url": "https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/ai-agents/responsible-ai-across-organization",
"evidence_quote": "Empower a cross-functional governance team. Establish an AI Center of Excellence or ethics committee that includes representatives from legal, security, product, and engineering teams. This group defines standards and provides consultative support rather than acting as gatekeepers. Give this team executive sponsorship and clear authority to enforce policies when necessary.",
"reason": "Den kanoniske siden beskriver ÉN enhet — AI Center of Excellence = cross-functional governance team, med executive sponsorship — ikke et AI Governance Board over tre underenheter; verken AI Governance Board eller Incident Response Team finnes som navngitte enheter (incident response er en prosedyre, ikke et organ).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/stakeholder-communication-ai-decisions.md#6",
"file": "skills/ms-ai-governance/references/responsible-ai/stakeholder-communication-ai-decisions.md",
"line": 374,
"rule": "R8",
"claim": "Azure Machine Learning tilbyr disse Responsible AI-verktøyene: Responsible AI Dashboard (end-to-end model assessment) | Responsible AI Scorecard (PDF-eksport av dashboard-innsikt) | Model Interpretability | Fairness Assessment (group-based performance metrics) | Error Analysis (cohort-based error distribution) | Causal Inference (what-if-analyse for counterfactuals)",
"evidence_url": "https://learn.microsoft.com/en-us/azure/machine-learning/concept-responsible-ai?view=azureml-api-2",
"evidence_quote": "Data-driven insights, which help stakeholders understand causal treatment effects on outcomes using historical data only... These insights come from the causal inference component of the Responsible AI dashboard. Model-driven insights, which answer user questions (such as What can I do to get a different outcome from your AI next time?) so they can take action. These insights are provided through the counterfactual what-if component of the Responsible AI dashboard.",
"reason": "Alle seks verktøyene finnes, men den bærende glossen er feil: counterfactuals hører til den separate counterfactual what-if-komponenten (DiCE), mens causal inference (EconML) handler om kausale behandlingseffekter — claimet slår dem sammen og utelater counterfactual what-if som eget verktøy.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/stakeholder-communication-ai-decisions.md#9",
"file": "skills/ms-ai-governance/references/responsible-ai/stakeholder-communication-ai-decisions.md",
"line": 437,
"rule": "R8",
"claim": "Copilot Studio har disse stakeholder-kommunikasjonsfunksjonene: agent observability med unik identitet per agent (owner, version, lifecycle status) | centralized logging av nøkkelhendelser til Azure Log Analytics | cost tracking av token consumption og compute usage per agent | user disclosure der agenter identifiserer seg som AI og ikke menneske",
"evidence_url": "https://learn.microsoft.com/en-us/azure/cloud-adoption-framework/ai-agents/governance-security-across-organization",
"evidence_quote": "Require every AI agent to be recorded in a single organizational inventory. Track ownership, purpose, platform, and access scope. Treat agents as managed organizational resources. Agent 365 provides an Agent Registry when adopted... Use Microsoft Entra Agent ID to assign identity, permissions, and lifecycle controls.",
"reason": "Siden tilskriver unik agent-identitet til Microsoft Entra Agent ID og registeret til Agent 365 — ikke til Copilot Studio — og sporer ownership/purpose/platform/access scope, ikke owner/version/lifecycle status; Copilot Studios egen fasilitering er Application Insights-telemetri og Copilot Credits-analytics, ikke Log Analytics-hendelseslogg med token/compute per agent.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/stakeholder-communication-ai-decisions.md#11",
"file": "skills/ms-ai-governance/references/responsible-ai/stakeholder-communication-ai-decisions.md",
"line": 471,
"rule": "R2",
"claim": "Power Platform CoE Starter Kit gir governance-kapabiliteter: inventory av alle AI Builder-modeller i tenant | compliance checks (er modellen i prod uten review?) | automatiske varsler til governance team ved high-risk deployments",
"evidence_url": "https://learn.microsoft.com/power-platform/guidance/coe/governance-components",
"evidence_quote": "Compliance detail request emails are sent for apps and chatbots. This flow sends an email to users who have apps in the tenant that aren't compliant with the following thresholds...",
"reason": "Den kanoniske opplistingen dekker apper, flows, chatbots, custom connectors og miljøer — AI Builder-modeller er fraværende fra inventaret — og compliance-varslene går til eierne/makerne, ikke til et governance-team ved high-risk deployments.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/stakeholder-communication-ai-decisions.md#15",
"file": "skills/ms-ai-governance/references/responsible-ai/stakeholder-communication-ai-decisions.md",
"line": 617,
"rule": "R2",
"claim": "Agent observability (Microsoft Agent 365) er inkludert i Copilot Studio-lisensen",
"evidence_url": "https://learn.microsoft.com/office365/servicedescriptions/microsoft-agent-365/microsoft-agent-365",
"evidence_quote": "Microsoft Agent 365 is available as a stand-alone subscription for use with eligible Microsoft 365 subscriptions and is also included with Microsoft 365 E7.",
"reason": "Agent 365 lisensieres per bruker som egen subscription eller via Microsoft 365 E7 (med E5 som forutsetning for nye kjøp fra 2026-06-01); Copilot Studio finnes ikke i opplistingen over lisenser som inkluderer Agent 365.",
"op": "O3",
"code": "CONTEXT_MISMATCH",
"detail": {
"token": "365",
"replacement": "7",
"would_have_been": "- Agent observability (Microsoft Agent 7): Inkludert i Copilot Studio-lisens"
}
},
{
"id": "ms-ai-governance/responsible-ai/stakeholder-communication-ai-decisions.md#21",
"file": "skills/ms-ai-governance/references/responsible-ai/stakeholder-communication-ai-decisions.md",
"line": 797,
"rule": "R8",
"claim": "Responsible AI Scorecard er dokumentert som preview — status GA med public preview for enkelte funksjoner (verifisert 2026-02)",
"evidence_url": "https://learn.microsoft.com/en-us/azure/machine-learning/concept-responsible-ai-scorecard?view=azureml-api-2",
"evidence_quote": "This feature is currently in public preview. This preview version is provided without a service-level agreement, and we don't recommend it for production workloads. Certain features might not be supported or might have constrained capabilities.",
"reason": "Hele scorecardet er i public preview — også tittelen bærer (preview) — så den bærende delen status GA med public preview kun for enkelte funksjoner er direkte motsagt.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"2026",
"02"
]
}
},
{
"id": "ms-ai-governance/responsible-ai/transparency-documentation-standards.md#3",
"file": "skills/ms-ai-governance/references/responsible-ai/transparency-documentation-standards.md",
"line": 82,
"rule": "R8",
"claim": "Microsofts model card-implementasjon omfatter Microsoft Foundry model catalog med innebygde model cards for pretrained modeller | Hugging Face-integrasjon der model cards synkroniseres automatisk | mal for å generere egne model cards for custom modeller.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/concepts/foundry-models-overview",
"evidence_quote": "Some of the details available in the model card are: Quick facts: Key information about the model at a quick glance",
"reason": "Kun første del holder (model catalog har innebygde model cards); den kanoniske model card-oppregningen nevner verken automatisk synkronisering av Hugging Face model cards eller noen mal for å generere egne model cards for custom modeller — to angitte, bærende deler mangler dekning.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/transparency-documentation-standards.md#4",
"file": "skills/ms-ai-governance/references/responsible-ai/transparency-documentation-standards.md",
"line": 114,
"rule": "R8",
"claim": "Responsible AI Scorecard inneholder komponentene Model overview | Fairness assessment | Model interpretability | Error analysis | Counterfactual analysis | Causal inference | Data quality.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/machine-learning/how-to-responsible-ai-scorecard",
"evidence_quote": "The Responsible AI scorecard is a PDF summary of key insights from your Responsible AI dashboard. The first summary segment of the scorecard gives you an overview of the machine learning model and the key target values you set",
"reason": "Den kanoniske gjennomgangen av scorecard-segmentene lister model overview, data analysis, model performance, cohorts, top important factors, fairness insights og causal insights — verken Error analysis eller Counterfactual analysis er scorecard-komponenter, så to angitte bærende deler er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/transparency-documentation-standards.md#8",
"file": "skills/ms-ai-governance/references/responsible-ai/transparency-documentation-standards.md",
"line": 386,
"rule": "R2",
"claim": "Azure OpenAI Service har egne transparency notes per modell, blant annet GPT-4 | DALL-E 3 | Whisper | o1.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/transparency-note",
"evidence_quote": "Model group | Text / code | Vision | Audio / Speech ... GPT-image-1 series ... GPT-image-2 ... Transcribe ... GPT-4 Turbo with Vision ... o1 series ... o3/o3-pro",
"reason": "Jeg hentet den kanoniske siden som ville oppregnet dem: Azure OpenAI har ÉN samlet transparency note med faner for tekst/vision/audio, ikke egne notiser per modell — og modelltabellen lister verken DALL-E 3 eller Whisper (erstattet av GPT-image-serien og Transcribe).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/transparency-documentation-standards.md#11",
"file": "skills/ms-ai-governance/references/responsible-ai/transparency-documentation-standards.md",
"line": 407,
"rule": "R1",
"claim": "Microsoft Foundry model catalog inneholder over 1500 pretrained modeller med model cards.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/concepts/foundry-models-overview",
"evidence_quote": "The catalog includes over 10,000 models, with approximately 50 new models published each month.",
"reason": "Nedre grense over 1500 overskrides grovt av den faktiske verdien over 10 000 (~7x) — bounden er beslutningsendrende misvisende etter lower-bound-policyen.",
"op": "O3",
"code": "MULTI_REPLACEMENT",
"detail": {
"candidates": [
"10,000",
"50"
]
}
},
{
"id": "ms-ai-governance/responsible-ai/transparency-documentation-standards.md#12",
"file": "skills/ms-ai-governance/references/responsible-ai/transparency-documentation-standards.md",
"line": 426,
"rule": "R8",
"claim": "Microsoft Copilot Studio har innebygde disclosures: «Powered by AI»-merke i chat-vinduet | attribusjonslenker til kildedokumenter for generative answers | bekreftelsesdialoger før sensitive plugin-handlinger | lenke til personvernerklæring i bot-innstillingene.",
"evidence_url": "https://learn.microsoft.com/microsoft-copilot-studio/system-service-card-copilot-studio",
"evidence_quote": "Human Oversight Guidance: Users are advised to review AI-generated outputs and automated actions before applying them in high-stakes scenarios. This practice mitigates risks of overreliance and ensures accountability.",
"reason": "Citations for generative answers og lenke til personvernerklæring er dokumentert, men den kanoniske oppregningen av innebygde sikkerhetskomponenter i Copilot Studio nevner verken et «Powered by AI»-merke i chat-vinduet eller bekreftelsesdialoger før sensitive plugin-handlinger — to angitte bærende deler mangler.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-governance/responsible-ai/transparency-documentation-standards.md#14",
"file": "skills/ms-ai-governance/references/responsible-ai/transparency-documentation-standards.md",
"line": 497,
"rule": "(none)",
"claim": "Microsofts Transparency Notes er tilgjengelige kun på engelsk (kan oversettes av kunden).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/transparency-note",
"evidence_quote": "Non-English translations are provided for convenience only. Please consult the EN-US version of this document for the definitive version.",
"reason": "Siden slår fast at Microsoft selv leverer ikke-engelske oversettelser (engelsk er kun den autoritative versjonen), noe som motsier at notisene er tilgjengelige kun på engelsk.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-governance/responsible-ai/transparency-documentation-standards.md#20",
"file": "skills/ms-ai-governance/references/responsible-ai/transparency-documentation-standards.md",
"line": 717,
"rule": "R4",
"claim": "Transparency note for Azure OpenAI er sist oppdatert med o3/o4-mini og Deep Research system cards.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/transparency-note",
"evidence_quote": "see the OpenAI o1 System Card, o3-mini System Card, o3/o4-mini System Card, Deep Research System Card, and GPT-5 System Card.",
"reason": "Den gjeldende versjonen av notisen er oppdatert langt forbi o3/o4-mini og Deep Research — den siterer GPT-5 System Card og dekker GPT-5-serien, GPT-5.1-Codex-Max, GPT-image-2 og GPT-Realtime-2, så «sist oppdatert med» treffer en foreldet tilstand.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"3",
"4"
]
}
},
{
"id": "ms-ai-infrastructure/bcdr/monitoring-alerting-failover-detection.md#3",
"file": "skills/ms-ai-infrastructure/references/bcdr/monitoring-alerting-failover-detection.md",
"line": 147,
"rule": "R2",
"claim": "Availability tests for health check-endepunkter opprettes med Azure CLI-kommandoen az monitor app-insights web-test create, med parametrene --resource-group | --app-insights | --web-test-name | --location | --defined-web-test-name | --url | --expected-status-code | --frequency | --timeout | --enabled.",
"evidence_url": "https://learn.microsoft.com/en-us/cli/azure/monitor/app-insights/web-test",
"evidence_quote": "--request-url\n\nUrl location to test.",
"reason": "Parameterlisten for az monitor app-insights web-test create inneholder verken --app-insights eller --url; URL-parameteren heter --request-url, og komponenten oppgis ikke med --app-insights i create.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-infrastructure/bcdr/monitoring-alerting-failover-detection.md#5",
"file": "skills/ms-ai-infrastructure/references/bcdr/monitoring-alerting-failover-detection.md",
"line": 180,
"rule": "R8",
"claim": "Azure OpenAI-diagnostikklogger ligger i AzureDiagnostics med ResourceProvider MICROSOFT.COGNITIVESERVICES og Category RequestResponse, med feltene properties_s.modelDeploymentName | duration_s | resultCode_d.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry-classic/openai/how-to/monitor-openai",
"evidence_quote": "| project TimeGenerated, _ResourceId, Category, OperationName, DurationMs, ResultSignature, properties_s",
"reason": "Kolonnene som dokumenteres for Azure OpenAI i AzureDiagnostics er DurationMs og ResultSignature, ikke duration_s og resultCode_d — én bærende del av claimen (feltnavnene) er motsagt selv om ResourceProvider og properties_s stemmer.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-infrastructure/bcdr/monitoring-alerting-failover-detection.md#6",
"file": "skills/ms-ai-infrastructure/references/bcdr/monitoring-alerting-failover-detection.md",
"line": 200,
"rule": "R2",
"claim": "Azure AI Search-diagnostikklogger ligger i AzureDiagnostics med ResourceProvider MICROSOFT.SEARCH og OperationName Query.Search, med feltene DurationMs | Properties.ResultCount | resultSignature_d.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/monitor-azure-cognitive-search-data-reference",
"evidence_quote": "| ResultSignature | Int | An HTTP result code. For example: `200` |",
"reason": "Properties-skjemaet på den kanoniske referansesiden lister kun Description_s, Documents_d, IndexName_s og Query_s — Properties.ResultCount finnes ikke — og resultatfeltet heter ResultSignature, ikke resultSignature_d (Microsoft.Search og Query.Search stemmer).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-infrastructure/bcdr/monitoring-alerting-failover-detection.md#7",
"file": "skills/ms-ai-infrastructure/references/bcdr/monitoring-alerting-failover-detection.md",
"line": 238,
"rule": "R3",
"claim": "Custom metrics til Application Insights sendes fra Python via opencensus.ext.azure.log_exporter.AzureLogHandler og applicationinsights-pakkens TelemetryClient initialisert med instrumentation_key, og emitteres med track_metric.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/azure-monitor/app/migrate-to-opentelemetry",
"evidence_quote": "Remove all instances of the OpenCensus SDK and the Azure Monitor OpenCensus exporter from your code.",
"reason": "Rammeverket claimen beskriver er erstattet: OpenCensus Python SDK er retired og `from opencensus.ext.azure.log_exporter import AzureLogHandler` står i listen over importer som skal fjernes; gjeldende løsning er Azure Monitor OpenTelemetry Distro med connection string, ikke instrumentation_key.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-infrastructure/bcdr/monitoring-alerting-failover-detection.md#8",
"file": "skills/ms-ai-infrastructure/references/bcdr/monitoring-alerting-failover-detection.md",
"line": 304,
"rule": "R2",
"claim": "Metric alert-regler opprettes med az monitor metrics alert create med parametrene --condition | --window-size | --evaluation-frequency | --severity | --action-group, der severity 0 er høyeste alvorlighetsgrad.",
"evidence_url": "https://learn.microsoft.com/en-us/cli/azure/monitor/metrics/alert",
"evidence_quote": "--action -a\n\nAdd an action group and optional webhook properties to fire when the alert is triggered.",
"reason": "az monitor metrics alert create har ingen --action-group-parameter; handlingsgruppe angis med --action. Severity 0 som høyeste stemmer («Severity of the alert from 0 (critical) to 4 (verbose)»), men det oppgitte flagget finnes ikke.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-infrastructure/bcdr/monitoring-alerting-failover-detection.md#9",
"file": "skills/ms-ai-infrastructure/references/bcdr/monitoring-alerting-failover-detection.md",
"line": 316,
"rule": "R7",
"claim": "Log-baserte alerts (scheduled query rules) opprettes med az monitor scheduled-query create med parametrene --condition | --condition-query | --evaluation-frequency | --window-size | --severity | --action-group.",
"evidence_url": "https://learn.microsoft.com/en-us/cli/azure/monitor/scheduled-query",
"evidence_quote": "--action-groups\n\nAction Group resource Ids to invoke when the alert fires.",
"reason": "Fem av seks parametere stemmer, men siden oppgir --action-groups (flertall); --action-group står ikke i parameterlisten, og flaggnavnet er selve påstanden.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-infrastructure/bcdr/monitoring-alerting-failover-detection.md#12",
"file": "skills/ms-ai-infrastructure/references/bcdr/monitoring-alerting-failover-detection.md",
"line": 437,
"rule": "R2",
"claim": "Agent details view i Application Insights omfatter funksjonene Unified agent view | End-to-end transaction details | Live metrics | Availability tests.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/azure-monitor/app/agents-view",
"evidence_quote": "The **Agent details** view in Application Insights provides a unified experience for monitoring AI agents across multiple sources, including Microsoft Foundry, Copilot Studio, and third-party agents.",
"reason": "Den kanoniske Agent details-siden omfatter unified view, traceundersøkelse, End-to-end transaction details og Grafana — Live metrics og Availability tests er egne Application Insights-visninger under Investigate og er fraværende fra Agent details-oppregningen.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#1",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 39,
"rule": "R2",
"claim": "Azure-verktøyene som knyttes til AI-spesifikke deteksjonstriggere er Azure AI Anomaly Detector | Microsoft Purview | Azure API Management analytics | Microsoft Sentinel | Azure AI Content Safety | Azure Monitor Log Analytics.",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/anomaly-detector/how-to/create-resource",
"evidence_quote": "Starting 20 September 2023 you won't be able to create new Anomaly Detector resources. The Anomaly Detector service is being retired 1 October 2026.",
"reason": "Den siterte CAF-siden («Detect AI security threats») navngir kun AI security posture management i Microsoft Defender for Cloud, ikke de seks verktøyene i påstanden, og Azure AI Anomaly Detector står i tillegg under utfasing — nye ressurser kan ikke opprettes og tjenesten pensjoneres 2026-10-01, så verktøylisten er både usourced og delvis utdatert.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#5",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 72,
"rule": "R2",
"claim": "Azure AI Content Safety har en «strict mode» som kan slås på for forsterket input-filtrering ved prompt injection-hendelser.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter-configurability",
"evidence_quote": "Low, medium, high | Yes | Yes | Strictest filtering configuration. Content detected at severity levels low, medium, and high is filtered.",
"reason": "Den kanoniske konfigurerbarhets-siden enumererer alvorlighetsterskler (strengeste konfigurasjon = filtrering på low/medium/high), Prompt Shields og blokkeringslister — ingen navngitt «strict mode» finnes å slå på, så entiteten er fraværende fra den siden som ville listet den.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#6",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 103,
"rule": "R8",
"claim": "Azure PowerShell-cmdleten Get-AzNetworkSecurityGroup støtter parameteren -ResourceId, og Add-AzNetworkSecurityRuleConfig støtter parameterne -Name | -Priority | -Access | -Protocol | -Direction | -SourceAddressPrefix | -DestinationAddressPrefix.",
"evidence_url": "https://learn.microsoft.com/en-us/powershell/module/az.network/get-aznetworksecuritygroup",
"evidence_quote": "Get-AzNetworkSecurityGroup [-Name <String>] [-ResourceGroupName <String>] [-DefaultProfile <IAzureContextContainer>] [<CommonParameters>]",
"reason": "Den andre halvdelen holder — Add-AzNetworkSecurityRuleConfig har alle sju parameterne — men Get-AzNetworkSecurityGroup har ingen -ResourceId-parameter i noen parametersett (kun -Name, -ResourceGroupName, -ExpandResource, -DefaultProfile), så én lastbærende del er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#7",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 111,
"rule": "R8",
"claim": "Azure PowerShell-cmdleten New-AzSnapshot støtter parameterne -SnapshotName og -Disk for å ta et forensisk øyeblikksbilde av en disk.",
"evidence_url": "https://learn.microsoft.com/en-us/powershell/module/az.compute/new-azsnapshot",
"evidence_quote": "New-AzSnapshot [-ResourceGroupName] <String> [-SnapshotName] <String> [-Snapshot] <PSSnapshot> [-AsJob] [-DefaultProfile <IAzureContextContainer>] [-WhatIf] [-Confirm] [<CommonParameters>]",
"reason": "-SnapshotName finnes, men -Disk gjør ikke det; cmdleten tar et lokalt snapshot-objekt via -Snapshot (PSSnapshot), så den lastbærende parameternavn-påstanden er feil og en leser ville kopiert et ikke-eksisterende flagg inn i koden.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#8",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 132,
"rule": "R8",
"claim": "Azure Storage sin immutabilityPolicy for blob-lagring består av feltene immutabilityPeriodSinceCreationInDays | allowProtectedAppendWrites | state (f.eks. «Locked»), med legalHold som eget objekt med tags og enabled.",
"evidence_url": "https://learn.microsoft.com/dotnet/api/microsoft.azure.management.storage.models.legalhold.haslegalhold",
"evidence_quote": "Gets the hasLegalHold public property is set to true by SRP if there are at least one existing tag. The hasLegalHold public property is set to false by SRP if all existing legal hold tags are cleared out.",
"reason": "immutabilityPolicy-feltene (immutabilityPeriodSinceCreationInDays, allowProtectedAppendWrites, state med verdien Locked) stemmer mot API-modellen, men legalHold-objektet har feltene tags og hasLegalHold — ikke «enabled» — så én lastbærende, bokstavelig feltnavn-påstand er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#15",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 461,
"rule": "(none)",
"claim": "Azure Monitor Log Analytics har de første 5 GB per dag gratis, deretter betaling per GB.",
"evidence_url": "https://learn.microsoft.com/azure/azure-local/manage/monitor-single-23h2",
"evidence_quote": "Azure Monitor has pay-as-you-go pricing, and the first 5 GB per billing account per month is free.",
"reason": "Kilden oppgir 5 GB per faktureringskonto per MÅNED som gratis, mens påstanden sier 5 GB per DAG — en enhet som er over 30 ganger for gunstig, og eksakt-verdi-regelen gjør avviket til en motstrid.",
"op": "O3",
"code": "LOCATOR_AMBIGUOUS",
"detail": {
"token": "5",
"hits": 14,
"window": {
"start": 457,
"end": 464
}
}
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#16",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 472,
"rule": "R3",
"claim": "Microsoft Sentinel lisensieres enten frittstående (standalone) eller via Microsoft 365 E5 Security.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/sentinel/billing",
"evidence_quote": "There are two ways to pay for the analytics tier: pay-as-you-go and commitment tiers.",
"reason": "Sentinel har ingen lisensmodell «frittstående eller via Microsoft 365 E5 Security» — betalingsmodellen er forbruksbasert (pay-as-you-go eller commitment tiers) på ingestert/analysert datavolum, og E5 gir kun en data-benefit-måler, så påstandens organiserende ramme er erstattet.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"365",
"5"
]
}
},
{
"id": "ms-ai-security/ai-security-engineering/ai-incident-response-procedures.md#17",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-incident-response-procedures.md",
"line": 474,
"rule": "R8",
"claim": "Microsoft Defender XDR krever Microsoft 365 E5 Security eller E5, og inkluderer Defender for Endpoint | Defender for Identity | Defender for Microsoft 365.",
"evidence_url": "https://learn.microsoft.com/defender-xdr/prerequisites",
"evidence_quote": "Any of these licenses give you access to Microsoft Defender XDR features via the Microsoft Defender portal without any additional cost: Microsoft 365 E5 or A5; Microsoft 365 E3 with the Microsoft Defender Suite add-on; Microsoft 365 E3 with the Enterprise Mobility + Security E5 add-on",
"reason": "To lastbærende deler svikter: Defender XDR krever ikke E5/E5 Security — en lang rekke lisenser gir tilgang uten ekstra kostnad — og komponenten heter Microsoft Defender for Office 365, ikke «Defender for Microsoft 365», et bokstavelig produktnavn en leser ville handlet på.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#2",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 37,
"rule": "R8",
"claim": "STRIDE-kategorien Spoofing dekker AI-truslene Neural Net Reprogramming | Malicious ML Providers, med alvorlighetsgrad Important-Critical.",
"evidence_url": "https://learn.microsoft.com/en-us/security/engineering/threat-modeling-aiml",
"evidence_quote": "#8 Malicious ML providers who can recover training data ... Traditional Parallels: Targeted information disclosure ... Severity: Important if data is PII, Moderate otherwise",
"reason": "Kilden plasserer «Malicious ML Providers» under information disclosure — ikke Spoofing — og gir den alvorlighetsgrad Important/Moderate, mens Neural Net Reprogramming beskrives som «an abuse scenario»; to lastbærende deler (STRIDE-kategorien og alvorlighetsbåndet) motsies.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#3",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 38,
"rule": "R8",
"claim": "STRIDE-kategorien Tampering dekker AI-truslene Data Poisoning (målrettet/vilkårlig) | Backdoored Models, med alvorlighetsgrad Critical.",
"evidence_url": "https://learn.microsoft.com/en-us/security/engineering/threat-modeling-aiml",
"evidence_quote": "#2b Indiscriminate Data Poisoning ... Traditional Parallels: Authenticated Denial of service against a high-value asset ... Severity: Important",
"reason": "Claimen sier begge poisoning-variantene er Critical og hører under Tampering, men kilden gir vilkårlig (indiscriminate) datapoisoning alvorlighetsgrad Important og parallellen Denial of service — én lastbærende del er direkte motsagt selv om targeted poisoning og Backdoored Models faktisk er Critical.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#4",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 39,
"rule": "R2",
"claim": "STRIDE-kategorien Repudiation dekker AI-truslene manipulasjon av modelloutput | tap av lineage for treningsdata, med alvorlighetsgrad Moderate.",
"evidence_url": "https://learn.microsoft.com/en-us/security/engineering/threat-modeling-aiml",
"evidence_quote": "#1 Adversarial Perturbation ... #2a Targeted Data Poisoning ... #2b Indiscriminate Data Poisoning ... #3 Model Inversion Attacks ... #4 Membership Inference Attack ... #5 Model Stealing ... #6 Neural Net Reprogramming ... #7 Adversarial Example in the Physical domain ... #8 Malicious ML providers who can recover training data ... #9 Attacking the ML Supply Chain ... #10 Backdoor Machine Learning ... #11 Exploit software dependencies of the ML system",
"reason": "Jeg hentet den kanoniske trusseloppregningen (11 trusler) og verken «manipulasjon av modelloutput» eller «tap av lineage for treningsdata» finnes som trusler der, ingen Repudiation-kategori er definert, og ingen slik alvorlighetsgrad Moderate er oppgitt — fraværet i den enumererende siden er bevis mot eksistenspåstanden.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#5",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 40,
"rule": "R8",
"claim": "STRIDE-kategorien Information Disclosure dekker AI-truslene Model Inversion | Membership Inference | Model Stealing, med alvorlighetsgrad Important-Critical.",
"evidence_url": "https://learn.microsoft.com/en-us/security/engineering/threat-modeling-aiml",
"evidence_quote": "#4 Membership Inference Attack ... Traditional Parallels: Data Privacy. Inferences are being made about a data point's inclusion in the training set but the training data itself is not being disclosed ... Severity: This is a privacy issue, not a security issue.",
"reason": "Model Inversion og Model Stealing knyttes riktignok til information disclosure i kilden, men Membership Inference plasseres eksplisitt under Data Privacy med presiseringen at treningsdata IKKE avsløres og uten sikkerhetsalvorlighet, og Model Stealing er «Important in security-sensitive models, Moderate otherwise» — båndet Important-Critical holder ikke for alle tre.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#7",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 42,
"rule": "R2",
"claim": "STRIDE-kategorien Elevation of Privilege dekker AI-truslene Adversarial Perturbation | Excessive Agency | Physical Domain Attacks, med alvorlighetsgrad Critical.",
"evidence_url": "https://learn.microsoft.com/en-us/security/engineering/threat-modeling-aiml",
"evidence_quote": "#1 Adversarial Perturbation ... #6 Neural Net Reprogramming ... #7 Adversarial Example in the Physical domain (bits->atoms) ... #11 Exploit software dependencies of the ML system",
"reason": "Adversarial Perturbation («Remote Elevation of Privilege», Critical) og Physical Domain Attacks («Elevation of Privilege, remote code execution», Critical) stemmer, men «Excessive Agency» finnes ikke i dokumentets kanoniske oppregning av de 11 truslene — det er et OWASP LLM-begrep, og fraværet i den enumererende siden gjør eksistenspåstanden feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#11",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 211,
"rule": "R8",
"claim": "Microsoft Defender for Cloud AI Security Posture Management omfatter automatisk oppdagelse av AI-arbeidsbelastninger på tvers av Azure-abonnementer via Azure Resource Graph | automatisert deteksjon og utbedring av risiko i generativ AI | sikkerhetsanbefalinger for AI-modeller, datalagre og nettverksisolasjon | integrasjon med Purview for dataklassifisering, DLP og Insider Risk Management for prompt-basert dataeksfiltrering.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/defender-for-cloud/ai-security-posture",
"evidence_quote": "Defender for Cloud automatically and continuously discovers deployed AI workloads across the following services: Azure OpenAI Service, Azure AI Foundry, Azure Machine Learning, Amazon Bedrock, Google Vertex AI",
"reason": "Den midterste delen («automate detection and remediation of generative AI risks») er korrekt, men den kanoniske AISPM-siden oppgir ikke Azure Resource Graph som oppdagelsesmekanisme og nevner ikke Purview-integrasjon i det hele tatt — CAF-siden lister Azure Resource Graph og Purview som SEPARATE verktøy («Use Azure Resource Graph to discover AI resources across subscriptions. Use Microsoft Defender for Cloud to identify generative AI workloads.»), så to lastbærende deler er feilattribuert til AISPM.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#12",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 227,
"rule": "R2",
"claim": "Microsoft Threat Modeling Tool har AI-spesifikke maler: ML Training Pipeline | Model API | LLM Agent.",
"evidence_url": "https://learn.microsoft.com/azure/security/develop/threat-modeling-tool-getting-started",
"evidence_quote": "You must select which template to use before creating a model. Our main template is the Azure Threat Model Template, which contains Azure-specific stencils, threats and mitigations. For generic models, select the SDL TM Knowledge Base from the drop-down menu.",
"reason": "Den kanoniske siden som enumererer malene i Threat Modeling Tool lister Azure Threat Model Template, SDL TM Knowledge Base og et medisinsk utstyr-stensilsett fra fellesskapet — ingen «ML Training Pipeline», «Model API» eller «LLM Agent»-mal finnes, så fraværet i den enumererende siden motbeviser eksistenspåstanden.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#13",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 280,
"rule": "R8",
"claim": "Microsoft Threat Modeling Tool er gratis nedlasting (ingen lisenskostnad) og gir STRIDE-automatisering, AI-spesifikke maler og trusselrapporter.",
"evidence_url": "https://learn.microsoft.com/azure/security/develop/threat-modeling-tool-getting-started",
"evidence_quote": "Our main template is the Azure Threat Model Template, which contains Azure-specific stencils, threats and mitigations. For generic models, select the SDL TM Knowledge Base from the drop-down menu.",
"reason": "STRIDE-automatisering («STRIDE per Element: Guided analysis of threats and mitigations») og trusselrapporter («Reports: Create HTML reports to share with others») er forankret, men den lastbærende delen «AI-spesifikke maler» finnes ikke i malopplistingen — én stated, lastbærende del er feil, så hele claimen faller.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#14",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 281,
"rule": "R8",
"claim": "AI-kapabilitetene i Microsoft Defender for Cloud (oppdagelse av AI-arbeidsbelastninger, posture management, trusseldeteksjon) krever standard tier, lisensiert per server.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/defender-for-cloud/ai-security-posture",
"evidence_quote": "The Defender Cloud Security Posture Management (CSPM) plan in Microsoft Defender for Cloud secures enterprise-built, multicloud, or hybrid cloud environments.",
"reason": "AI-kapabilitetene krever Defender CSPM-planen (posture/oppdagelse) og Defender for AI Services-planen (trusseldeteksjon) — ikke en «standard tier»; og faktureringen er ikke per server, men «Defender CSPM billing is based on specific resources» og for AI Services per skannede tokens («capped at 75 billion tokens scanned»).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/ai-threat-modeling-stride.md#18",
"file": "skills/ms-ai-security/references/ai-security-engineering/ai-threat-modeling-stride.md",
"line": 359,
"rule": "R8",
"claim": "Microsoft Learn-dokumentet «Security Planning for LLM-based Applications» beskriver 11 LLM-spesifikke trusler kartlagt mot STRIDE, med mitigeringsmønstre for Azure OpenAI.",
"evidence_url": "https://learn.microsoft.com/en-us/ai/playbook/technology-guidance/generative-ai/mlops-in-openai/security/security-plan-llm-application",
"evidence_quote": "Security principles: Confidentiality refers to the objective of keeping data private or secret. ... Integrity is about ensuring that data has not been tampered with ... Availability means that networks, systems, and applications are up and running. ... Privacy relates to activities that focus on individual users' rights.",
"reason": "Antallet stemmer (Threat #1 til #11) og mitigeringene refererer Azure OpenAI/Azure AI Content Safety, men dokumentet kartlegger hver trussel mot CIA-prinsippene pluss Privacy («Principle: Confidentiality, integrity, availability») — ikke mot STRIDE; den lastbærende rammepåstanden «kartlagt mot STRIDE» er erstattet av et annet rammeverk.",
"op": "O3",
"code": "LOCATOR_AMBIGUOUS",
"detail": {
"token": "11",
"hits": 2,
"window": {
"start": 356,
"end": 363
}
}
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#9",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 134,
"rule": "(none)",
"claim": "Sensitivity label-basert blokkering for Copilot omfatter kun filer lagret i SharePoint Online og OneDrive for Business.",
"evidence_url": "https://learn.microsoft.com/en-us/purview/dlp-microsoft365-copilot-location-learn-about",
"evidence_quote": "Detects when a file or an email in Exchange has a chosen sensitivity label.",
"reason": "Påstanden om at kun filer lagret i SharePoint Online og OneDrive dekkes, motsies av at dekningen omfatter både File items, which are stored and items that are actively open og e-post i Exchange.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#16",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 367,
"rule": "R2",
"claim": "Microsoft Purview Endpoint DLP støtter handlingene: block paste | block upload | warn with override.",
"evidence_url": "https://learn.microsoft.com/purview/dlp-configure-endpoint-settings",
"evidence_quote": "When creating rules for endpoint devices, choose the Audit or restrict activities on devices option, and select one of these options: 1. Audit only 2. Block with override 3. Block",
"reason": "Den kanoniske handlingslisten for Endpoint DLP er Audit only, Block with override og Block — warn with override finnes ikke i oppregningen, selv om blokkering av paste og upload er støttet.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#17",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 384,
"rule": "(none)",
"claim": "Endpoint DLP mot tredjeparts generative AI-nettsteder støttes kun på Windows-maskiner med Endpoint DLP-agent installert.",
"evidence_url": "https://learn.microsoft.com/purview/endpoint-dlp-learn-about",
"evidence_quote": "Endpoint data loss prevention (Endpoint DLP) extends the activity monitoring and protection capabilities of DLP to Windows 10/11, macOS (the three latest released major versions) devices, and Windows certain server versions.",
"reason": "macOS er støttet for Sensitive service domain groups (som inneholder gruppen Generative AI websites) og for opplasting til begrensede skytjenestedomener, så begrensningen til kun Windows-maskiner er feil.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#18",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 396,
"rule": "R8",
"claim": "Purview Insider Risk Management har policy-malene: DSPM for AI - Detect risky AI usage | DSPM for AI - Unethical behavior in AI apps | DSPM for AI - Protect sensitive data from Copilot processing.",
"evidence_url": "https://learn.microsoft.com/purview/dspm-for-ai-considerations",
"evidence_quote": "Communication Compliance: DSPM for AI - Unethical behavior in AI apps",
"reason": "Bare DSPM for AI - Detect risky AI usage er en Insider Risk Management-policy; Unethical behavior in AI apps er en Communication Compliance-policy og Protect sensitive data from Copilot processing er en DLP-policy, så to av tre lastbærende deler er feil tilordnet.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#19",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 419,
"rule": "R8",
"claim": "Azure OpenAI støtter ikke persistent prompt caching på tvers av brukere; hvert API-kall er stateless med mindre samtalehistorikk sendes eksplisitt i requesten.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/how-to/prompt-caching",
"evidence_quote": "Extended prompt cache retention keeps cached prefixes active for longer, up to a maximum of 24 hours.",
"reason": "Azure OpenAI dokumenterer nå vedvarende prompt-caching i inntil 24 timer, som er standard for nyere modeller, og kilden garanterer bare at cacher ikke deles mellom Azure-abonnementer — ikke at hvert kall er stateless.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#22",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 565,
"rule": "R8",
"claim": "Retention for Microsoft Purview Audit er konfigurerbar fra 90 dager til 10 år.",
"evidence_url": "https://learn.microsoft.com/purview/audit-log-retention-policies",
"evidence_quote": "The available options are 7 Days, 30 Days, 6 Months, 9 Months, 1 Year, 3 Years, 5 Years, and 7 Years. Users with the 10-year Audit Log Retention add-on license can select a 10 Years option.",
"reason": "Øvre grense på 10 år stemmer, men det finnes ingen 90-dagers valgmulighet — varighetene starter på 7 dager, og standardretention ble dessuten endret fra 90 til 180 dager, så nedre grense i påstanden er feil.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"90",
"10"
]
}
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#24",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 589,
"rule": "R8",
"claim": "DSPM er i preview som ny versjon med utvidet AI activities-fane.",
"evidence_url": "https://learn.microsoft.com/purview/whats-new#may-2026",
"evidence_quote": "General availability (GA): The new version of Data Security Posture Management is now generally available. Partner solutions for non-Microsoft data sources remain in preview, as does the Data Security Posture Agent.",
"reason": "AI activities-fanen finnes i den nye versjonen, men den nye DSPM-versjonen er GA og dokumenteres som current version — ikke preview — så statusdelen av påstanden er motsagt.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "SOURCE_STATUS_AMBIGUOUS",
"rows": [
"GA",
"PREVIEW"
]
}
},
{
"id": "ms-ai-security/ai-security-engineering/data-leakage-prevention-ai.md#25",
"file": "skills/ms-ai-security/references/ai-security-engineering/data-leakage-prevention-ai.md",
"line": 606,
"rule": "R8",
"claim": "Sentrale DLP-cmdlets i ExchangePowerShell er: New-DlpCompliancePolicy | New-DlpComplianceRule | Get-DlpCompliancePolicy | Set-DlpPolicy | Get-Label.",
"evidence_url": "https://learn.microsoft.com/powershell/module/exchangepowershell/set-dlppolicy?view=exchange-ps",
"evidence_quote": "Note: This cmdlet is retired from the cloud-based service. This cmdlet is functional only in on-premises Exchange. Use the Set-DlpCompliancePolicy and Set-DlpComplianceRule cmdlets instead.",
"reason": "Fire av de fem cmdletene er gjeldende, men Set-DlpPolicy er pensjonert fra skytjenesten og virker kun i lokal Exchange, så listen over sentrale DLP-cmdlets har én lastbærende feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#1",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 37,
"rule": "R8",
"claim": "Security Copilot er inkludert i lisensen for Microsoft 365 E5/E7 inclusion-kunder og auto-provisjoneres etter en 7-dagers forhåndsvarsling fra Microsoft; ingen SCU-kjøp er nødvendig for grunnfunksjonalitet, og aktivering skjer per tenant.",
"evidence_url": "https://learn.microsoft.com/copilot/security/security-copilot-inclusion",
"evidence_quote": "Eligible Microsoft 365 E5 and E7 customers who were notified of their eligibility will automatically receive Security Copilot through zero click activation. For eligible customers, Security Copilot is provisioned automatically as part of their subscription. This means no Azure setup is needed or capacity provisioning required.",
"reason": "Inklusjon i E5/E7 og auto-provisjonering uten SCU-kjøp stemmer, men den lastbærende delen om en 7-dagers forhåndsvarsling finnes ikke på noen Learn-side (auto-provisioning-siden lister hva som provisjoneres uten noe varslingsvindu, og eneste dokumenterte frist er 30 dager for fremtidig pay-as-you-go).",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"365",
"5",
"7"
]
}
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#4",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 83,
"rule": "R4",
"claim": "Phishing Triage Agent i Defender XDR er i Public Preview.",
"evidence_url": "https://learn.microsoft.com/defender-xdr/security-alert-triage-agent",
"evidence_quote": "Email and collaboration alert triage capabilities are already generally available (GA).",
"reason": "Public preview-statusen stammer fra en tidsstemplet what's-new-post fra juli 2025; gjeldende dokumentasjon sier at phishing-/e-posttriagen er GA, og bare de utvidede sky- og identitetsalarmene er i preview.",
"op": "O1",
"code": "PROVEN",
"proposal": {
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 83,
"token": "Public Preview",
"replacement": "GA",
"type": "status",
"status_row_from": "PREVIEW",
"status_row_to": "GA",
"before": "| **Phishing Triage Agent** | Defender XDR | Autonomt triage og klassifisering av brukerrapporterte phishing-hendelser. Semantisk analyse av e-post, URLer og filer. Lærer av analytikerfeedback. | Public Preview |",
"after": "| **Phishing Triage Agent** | Defender XDR | Autonomt triage og klassifisering av brukerrapporterte phishing-hendelser. Semantisk analyse av e-post, URLer og filer. Lærer av analytikerfeedback. | GA |",
"evidence_url": "https://learn.microsoft.com/defender-xdr/security-alert-triage-agent",
"evidence_quote": "Email and collaboration alert triage capabilities are already generally available (GA)."
}
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#7",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 91,
"rule": "R4",
"claim": "Threat Intelligence Briefing Agent i standalone-portalen er i Public Preview.",
"evidence_url": "https://learn.microsoft.com/defender-xdr/security-copilot-agents-defender",
"evidence_quote": "1. Security Alert Triage Agent (Preview) 2. Threat Intelligence Briefing Agent 3. Threat Hunting Agent 4. Security Analyst Agent 5. Dynamic Threat Detection Agent 6. Data Security Triage Agent in Data Loss Prevention",
"reason": "«Public Preview» står kun i en tidsstemplet what's-new-post fra juli 2025; de gjeldende kanoniske sidene merker preview-agenter eksplisitt, og Threat Intelligence Briefing Agent står umerket både her og i standalone-artikkelen, som heller ikke har noe preview-varsel.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "FILE_ALREADY_MATCHES",
"row": "PREVIEW"
}
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#9",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 93,
"rule": "(none)",
"claim": "Vulnerability Remediation Agent i Microsoft Intune er GA.",
"evidence_url": "https://learn.microsoft.com/intune/copilot/agents/vulnerability-remediation-agent",
"evidence_quote": "This feature is in public preview. For more information, see Public preview in Microsoft Intune.",
"reason": "Agentsiden merker Vulnerability Remediation Agent som public preview (og admin-senteret viser «Vulnerability Remediation Agent (preview)»), mens påstanden sier GA.",
"op": "O1",
"code": "PROVEN",
"proposal": {
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 93,
"token": "GA",
"replacement": "Preview",
"type": "status",
"status_row_from": "GA",
"status_row_to": "PREVIEW",
"before": "| **Vulnerability Remediation Agent** | Microsoft Intune | Identifiserer topp-CVE-er, bruker Defender-data, gir trinnvis remediering via Intune | GA |",
"after": "| **Vulnerability Remediation Agent** | Microsoft Intune | Identifiserer topp-CVE-er, bruker Defender-data, gir trinnvis remediering via Intune | Preview |",
"evidence_url": "https://learn.microsoft.com/intune/copilot/agents/vulnerability-remediation-agent",
"evidence_quote": "This feature is in public preview. For more information, see Public preview in Microsoft Intune."
}
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#10",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 94,
"rule": "R2",
"claim": "Access Review Agent i Microsoft Entra + Teams er GA.",
"evidence_url": "https://learn.microsoft.com/entra/security-copilot/entra-agents",
"evidence_quote": "The following agents are currently available for Microsoft Entra. ... Conditional Access Optimization Agent ... Identity Risk Management Agent (Preview)",
"reason": "Access Review Agent er fraværende i den kanoniske Entra-agentlisten, og eneste statusangivelse på Learn er what's-new-posten «Microsoft and partner agents - Public preview» med Access Review Agent i Microsoft Entra — ingen kilde sier GA.",
"op": "O1",
"code": "PROVEN",
"proposal": {
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 94,
"token": "GA",
"replacement": "Preview",
"type": "status",
"status_row_from": "GA",
"status_row_to": "PREVIEW",
"before": "| **Access Review Agent** | Microsoft Entra + Teams | Leverer innsikt og anbefalinger for tilgangsgjennomgang direkte i Teams | GA |",
"after": "| **Access Review Agent** | Microsoft Entra + Teams | Leverer innsikt og anbefalinger for tilgangsgjennomgang direkte i Teams | Preview |",
"evidence_url": "https://learn.microsoft.com/entra/security-copilot/entra-agents",
"evidence_quote": "The following agents are currently available for Microsoft Entra. ... Conditional Access Optimization Agent ... Identity Risk Management Agent (Preview)"
}
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#11",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 100,
"rule": "(none)",
"claim": "Security Copilot-agentene for endepunktadministrasjon i Intune består av: Change Review Agent | Device Offboarding Agent | Policy Configuration Agent.",
"evidence_url": "https://learn.microsoft.com/intune/agents/",
"evidence_quote": "Microsoft Intune includes specialized Security Copilot agents, each designed for a specific security scenario. The following agents are available: Change Review Agent ... Device Offboarding Agent ... Policy Configuration Agent ... Vulnerability Remediation Agent",
"reason": "Påstanden sier at agentene «består av» tre navngitte agenter, mens den kanoniske Intune-siden lister fire — Vulnerability Remediation Agent mangler i den lukkede oppregningen.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#17",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 120,
"rule": "R8",
"claim": "Kunder mottar 30-dagers forhåndsvarsel, deretter auto-provisjoneres Security Copilot uten Azure-oppsett eller manuell SCU-tildeling (zero-click activation).",
"evidence_url": "https://learn.microsoft.com/copilot/security/security-copilot-inclusion",
"evidence_quote": "Customers will get a 30-day advanced notification when this option is available.",
"reason": "Zero-click-delen stemmer, men den lastbærende koblingen mellom et 30-dagers forhåndsvarsel og auto-provisjoneringen er feil — kilden knytter 30-dagersvarselet til den fremtidige pay-as-you-go-opsjonen, ikke til provisjoneringen.",
"op": "O3",
"code": "LOCATOR_AMBIGUOUS",
"detail": {
"token": "30",
"hits": 2,
"window": {
"start": 116,
"end": 122
}
}
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#27",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 235,
"rule": "(none)",
"claim": "Tilpassede Security Copilot-plugins finnes i typene: API-plugin (OpenAPI-spec-wrapper rundt REST API) | KQL-plugin (egendefinerte KQL-spørringer mot Sentinel/Defender) | OpenAI-format (ChatGPT-kompatibelt plugin-format) | Egendefinert agent.",
"evidence_url": "https://learn.microsoft.com/copilot/security/plugin-overview",
"evidence_quote": "A plugin is a collection of related tools that users can enable in Security Copilot when relevant to creating new prompting, promptbook, or agentic capabilities. ... For agents, plugins extend what an agent can do by giving it access to resources outside of an agent and LLM.",
"reason": "Learn kategoriserer egendefinerte plugins som API-, KQL- og GPT-plugins (egne artikler plugin-api, plugin-kql, plugin-gpt) pluss MCP, mens påstandens lukkede liste mangler GPT-plugin og fører opp «egendefinert agent», som ikke er en plugintype — plugins er noe som utvider agenter.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/security-copilot-integration.md#33",
"file": "skills/ms-ai-security/references/ai-security-engineering/security-copilot-integration.md",
"line": 272,
"rule": "R2",
"claim": "Tredjepartspluginer tilgjengelige via Security Store inkluderer: AbuseIPDB | Censys | CrowdSec CTI | CyberArk | Cybersixgill | Red Canary | Jamf.",
"evidence_url": "https://learn.microsoft.com/copilot/security/plugin-other",
"evidence_quote": "AbuseIPDB ... Aviatrix ... Censys ... CheckPhish ... CIRCL Hash Lookup ... CrowdSec ... CyberArk Privilege Cloud ... Cybersixgill ... Cyware Intel Exchange ... Darktrace ... Jamf ... Netskope ... Quest Security Guardian ... ReversingLabs ... Saviynt ... ServiceNow SIR ... Shodan ... Splunk ... Tanium ... UrlScan",
"reason": "Den kanoniske oversikten over ikke-Microsoft-plugins inneholder seks av de syv navnene, men Red Canary er ikke oppført i det hele tatt.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#4",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 126,
"rule": "R2",
"claim": "Azure Pipelines-oppgaven for avhengighetsskanning heter AdvancedSecurity-Dependency-Scanning@1 (versjon 1) og tar inputene scanMode og ecosystem (f.eks. «pip»).",
"evidence_url": "https://learn.microsoft.com/azure/devops/pipelines/tasks/reference/advanced-security-dependency-scanning-v1?view=azure-pipelines",
"evidence_quote": "Inputs: `directoryExclusionList` - Directory exclusion list `string`. List of relative directory paths to ignore as a set of semi-colon separated values.",
"reason": "Oppgavenavnet og versjon 1 stemmer, men den kanoniske oppgavereferansen enumererer kun inputen directoryExclusionList — verken scanMode eller ecosystem finnes i input-listen, og fraværet i den autoritative enumereringen er bevis mot påstanden.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "1",
"type": "number"
}
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#5",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 134,
"rule": "R8",
"claim": "Dependency scanning genererer alerts for tre kategorier: Direct vulnerabilities (pakker i requirements.txt) | Transitive vulnerabilities (pakker som direkte dependencies bruker) | CVE severity mapping.",
"evidence_url": "https://learn.microsoft.com/azure/devops/repos/security/github-advanced-security-dependency-scanning?view=azure-devops",
"evidence_quote": "Dependency scanning generates an alert for any open-source component, direct or transitive, found to be vulnerable that your code depends upon.",
"reason": "Kilden beskriver to alert-kategorier (direkte og transitive sårbare komponenter); CVSS-til-alvorlighet er en egenskap ved en alert, ikke en tredje alert-kategori, så den lastbærende «tre kategorier»-delen er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#8",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 157,
"rule": "R8",
"claim": "Microsoft Defender for Containers gjør tre ting: genererer vulnerability assessments automatisk når et image pushes til Azure Container Registry | blokkerer deployment av images med kritiske sårbarheter (konfigurerbart via Azure Policy) | integrerer med Azure Monitor for alerting.",
"evidence_url": "https://learn.microsoft.com/azure/defender-for-cloud/runtime-gated-overview",
"evidence_quote": "Gated deployment is a Microsoft Defender for Containers capability that uses an admission controller to evaluate container images before they're admitted into a Kubernetes cluster. ... Deny blocks deployments that match the rule conditions.",
"reason": "Skanning ved push stemmer, men blokkering av images med kritiske sårbarheter konfigureres via Defender for Containers' gated deployment-sikkerhetsregler i Defender for Cloud, ikke via Azure Policy — den relevante Azure Policy-definisjonen for sårbare running images har kun effektene AuditIfNotExists og Disabled, så den lastbærende Azure Policy-delen er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#9",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 202,
"rule": "R8",
"claim": "Microsoft tilbyr verifiserte modeller via to kanaler: Azure Machine Learning Model Catalog (kuraterte modeller med security attestation) | HuggingFace Registry i Azure (integrert med Azure ML, med provenance tracking).",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/how-to-deploy-models-from-huggingface?view=azureml-api-2#frequently-asked-questions",
"evidence_quote": "`HuggingFace` is a community registry and Microsoft support doesn't cover it. ... The model weights aren't hosted on Azure.",
"reason": "HuggingFace-registeret beskrives som et community registry som Microsoft-support ikke dekker, og modellvektene lastes ned direkte fra Hugging Face hub — det er derfor ikke en Microsoft-kanal for verifiserte modeller med provenance tracking, så den ene lastbærende halvdelen av påstanden er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#10",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 209,
"rule": "R4",
"claim": "Modellen gpt-35-turbo finnes i azureml-registeret med modellversjon 0301, referert som azureml://registries/azureml/models/gpt-35-turbo/versions/0301.",
"evidence_url": "https://learn.microsoft.com/azure/ai-foundry/openai/concepts/legacy-models",
"evidence_quote": "These models are no longer available for new deployments. ... `gpt-35-turbo` - 0301 ... Retirement date: February 13, 2025",
"reason": "gpt-35-turbo versjon 0301 står i tabellen over pensjonerte modeller med pensjoneringsdato 13. februar 2025 og er ikke lenger tilgjengelig for nye deployeringer; i tillegg unntar Learn eksplisitt Azure OpenAI-modeller fra registry-modell-ID-mønsteret påstanden bruker, så påstanden treffer kun en utdatert rad.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"35",
"0301"
]
}
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#12",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 255,
"rule": "R8",
"claim": "Azure Policy-definisjonen «[Preview]: Azure Machine Learning Deployments should only use approved Registry Models» er i preview og støtter effect «Deny» med parameterne allowedPublishers og approvedAssetIds.",
"evidence_url": "https://learn.microsoft.com/azure/foundry-classic/how-to/built-in-policy-model-deployment#enable-the-policy",
"evidence_quote": "Effect: Set to Deny. ... Allowed Models Publishers: Enter a list of publisher names enclosed in quotation marks, separated by commas. ... Allowed Asset Ids: Enter a list of model asset IDs enclosed in quotation marks, separated by commas.",
"reason": "Policyen finnes i preview (1.0.0-preview) og støtter Deny, men de dokumenterte parameterne heter «Allowed Models Publishers» og «Allowed Asset Ids» (allowedPublishers/allowedAssetIds i CLI-eksempelet) — approvedAssetIds forekommer ingen steder, og det er en lastbærende streng en leser ville kopiert inn i en parameterfil.",
"op": "O3",
"code": "STATUS_SYNONYM",
"detail": {
"reason": "NO_SOURCE_STATUS"
}
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#14",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 276,
"rule": "R8",
"claim": "Azure AI Anomaly Detector er tilgjengelig som tjeneste og kan deployes for å identifisere data poisoning i treningsdata (klienten AnomalyDetectorClient med detect_entire_series).",
"evidence_url": "https://learn.microsoft.com/azure/ai-services/anomaly-detector/overview",
"evidence_quote": "Starting 20 September 2023 you won't be able to create new Anomaly Detector resources. The Anomaly Detector service is being retired 1 October 2026.",
"reason": "Kilden sier at nye Anomaly Detector-ressurser ikke kan opprettes siden 20. september 2023 og at tjenesten pensjoneres 1. oktober 2026, så den lastbærende delen om at tjenesten er tilgjengelig og kan deployes, er direkte motsagt.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/ai-security-engineering/supply-chain-security-ai-models.md#15",
"file": "skills/ms-ai-security/references/ai-security-engineering/supply-chain-security-ai-models.md",
"line": 345,
"rule": "R2",
"claim": "Azure ML gir delvis SBOM-funksjonalitet via tre mekanismer: Model Registry Metadata (navn, versjon, tags, properties, koblet treningsjobb) | Environment Registry (conda-avhengigheter, pip-pakker, Docker base image, kryptografisk hash av miljødefinisjonen) | Dataset Versioning (Azure ML Data Assets med versjonering og lineage tracking).",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/reference-yaml-environment?view=azureml-api-2",
"evidence_quote": "YAML syntax: `$schema`, `name`, `version`, `description`, `tags`, `image`, `conda_file`, `build`, `build.path`, `build.dockerfile_path`, `os_type`, `inference_config`",
"reason": "Det kanoniske environment-skjemaet enumererer alle felt en environment-definisjon bærer, og ingen kryptografisk hash av miljødefinisjonen finnes blant dem; fraværet i den autoritative enumereringen gjør den lastbærende SBOM-mekanismen ugrunnet.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/gpt5-gpt41-pricing-models.md#1",
"file": "skills/ms-ai-security/references/cost-optimization/gpt5-gpt41-pricing-models.md",
"line": 56,
"rule": "(none)",
"claim": "gpt-4.1 kontekstvindu: 1047576 tokens (full), 128000 tokens (standard og provisioned deployments), 300000 tokens (batch deployments)",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure?view=foundry-classic",
"evidence_quote": "- 1,047,576 - 300,000 (standard deployments) - 128,000 (provisioned managed and batch deployments)",
"reason": "The page states 300,000 for standard deployments and 128,000 for provisioned managed AND batch deployments, while the claim swaps them (128K for standard/provisioned, 300K for batch).",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"4.1",
"1047576",
"128000",
"300000"
]
}
},
{
"id": "ms-ai-security/cost-optimization/gpt5-gpt41-pricing-models.md#5",
"file": "skills/ms-ai-security/references/cost-optimization/gpt5-gpt41-pricing-models.md",
"line": 83,
"rule": "R4",
"claim": "gpt-5 og gpt-5-codex krever registrering og godkjenning; gpt-5-mini, gpt-5-nano, gpt-5-chat har ingen registreringskrav",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/how-to/reasoning",
"evidence_quote": "Access is no longer restricted for this model.",
"reason": "The live availability table states this for both gpt-5 and gpt-5-codex, so the claimed registration/approval requirement is superseded (the mini/nano/chat 'No access request needed' half still matches, but the load-bearing requirement part is now wrong).",
"op": "O3",
"code": "LOCATOR_AMBIGUOUS",
"detail": {
"token": "5",
"hits": 5,
"window": {
"start": 82,
"end": 85
}
}
},
{
"id": "ms-ai-security/cost-optimization/gpt5-gpt41-pricing-models.md#9",
"file": "skills/ms-ai-security/references/cost-optimization/gpt5-gpt41-pricing-models.md",
"line": 203,
"rule": "R8",
"claim": "Copilot Credits-takstnivåer: gpt-4.1-mini Basic | gpt-4.1 Standard | gpt-5-chat (preview) Standard | gpt-5-reasoning (preview) Premium | o3 Premium | Claude Sonnet 4.5 (experimental) Standard | Claude Opus 4.5 (experimental) Premium",
"evidence_url": "https://learn.microsoft.com/microsoft-copilot-studio/prompt-model-settings",
"evidence_quote": "| Claude Sonnet 4.6 | Standard rate | External model from Anthropic. Context allowed up to 200K tokens. | General |",
"reason": "The current model table lists Claude Sonnet 4.6 (Standard) and Claude Opus 4.6 (Premium) instead of the claimed 4.5 versions and contains no o3 row at all, so multiple load-bearing entries in the claimed rate lineup are superseded or absent.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/gpt5-gpt41-pricing-models.md#11",
"file": "skills/ms-ai-security/references/cost-optimization/gpt5-gpt41-pricing-models.md",
"line": 217,
"rule": "R2",
"claim": "Claude Sonnet 4.5 og Claude Opus 4.5 er tilgjengelig i Copilot Studio (experimental, 200K kontekstvindu)",
"evidence_url": "https://learn.microsoft.com/microsoft-copilot-studio/prompt-model-settings",
"evidence_quote": "| Claude Sonnet 4.6 | Standard rate | External model from Anthropic. Context allowed up to 200K tokens. | General | | Claude Opus 4.6 | Premium rate | External model from Anthropic. Context allowed up to 200K tokens. | Deep |",
"reason": "The canonical model enumeration lists Claude Sonnet 4.6 and Claude Opus 4.6 (200K); the claimed Sonnet 4.5/Opus 4.5 are absent, i.e. superseded by 4.6 versions.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"4.5",
"200"
]
}
},
{
"id": "ms-ai-security/cost-optimization/gpt5-gpt41-pricing-models.md#14",
"file": "skills/ms-ai-security/references/cost-optimization/gpt5-gpt41-pricing-models.md",
"line": 393,
"rule": "R8",
"claim": "Norway East: kun gpt-4o/gpt-4o-mini tilgjengelig som Norge-resident (Standard/Regional PTU); gpt-4.1, o-serien (o3/o4-mini/o3-mini/o1) og GPT-5-familien finnes ikke som Regional/PTU i Norway East; eneste GPT-5 med EU-residens i Norway East er gpt-5.5 via Data Zone Standard",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure-region-availability?pivots=standard",
"evidence_quote": "| gpt-5.4 | 2026-03-05 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |",
"reason": "The Data Zone Standard Europe table now shows gpt-5.4, gpt-5.5, and gpt-5.6-sol/terra/luna all available in norwayeast, so the load-bearing 'eneste GPT-5 ... er gpt-5.5' part is superseded, even though the no-Regional/PTU parts (gpt-4.1, o-series, GPT-5 family absent in Norway East) still verify.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"4",
"4.1",
"3",
"1",
"5",
"5.5"
]
}
},
{
"id": "ms-ai-security/cost-optimization/gpt5-gpt41-pricing-models.md#16",
"file": "skills/ms-ai-security/references/cost-optimization/gpt5-gpt41-pricing-models.md",
"line": 468,
"rule": "R8",
"claim": "GPT-5-tilgjengelighet: gpt-5 GA (begrenset, krever godkjenning aka.ms/oai/gpt5access) | gpt-5-mini GA | gpt-5-nano GA | gpt-5-chat Preview (2 versjoner) | gpt-5-codex GA (begrenset) | gpt-5-pro GA (begrenset, kun MCA-E/Default-abonnementer)",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/how-to/reasoning",
"evidence_quote": "Access is no longer restricted for this model.",
"reason": "The live availability table states this for gpt-5, gpt-5-codex, and gpt-5-pro, contradicting the claimed aka.ms/oai/gpt5access approval requirement and the MCA-E/Default-only restriction for gpt-5-pro; the gpt-5-chat 'Preview (2 versions)' part matches, but the load-bearing access-status parts are superseded.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/gpt5-gpt41-pricing-models.md#17",
"file": "skills/ms-ai-security/references/cost-optimization/gpt5-gpt41-pricing-models.md",
"line": 443,
"rule": "R8",
"claim": "AI Builder: prompt builder credits inkludert i premium Power Platform-planer (500 credits/bruker/mnd); default modell gpt-4.1-mini (Basic rate)",
"evidence_url": "https://learn.microsoft.com/ai-builder/credit-management",
"evidence_quote": "| Power Apps Premium | 500 | Maximum = 1,000,000 AI Builder credits per tenant. | ... | Power Automate Premium | 5,000 | Maximum = 1,000,000 AI Builder credits per tenant. |",
"reason": "The 500-credit figure applies only to Power Apps Premium; other premium Power Platform plans seed 250 to 5,000 AI Builder credits, and the page adds that these seeded credits are removed on November 1, 2026, so the blanket '500 credits/bruker/mnd i premium-planer' is contradicted even though the gpt-4.1-mini Basic default is correct.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"500",
"4.1"
]
}
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#3",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 53,
"rule": "R2",
"claim": "Vector-database-alternativer for semantic caching på Azure er Azure Managed Redis (RediSearch) | Azure Cache for Redis Enterprise | Azure AI Search.",
"evidence_url": "https://learn.microsoft.com/azure/azure-cache-for-redis/retirement-faq",
"evidence_quote": "All instances of Azure Cache for Redis Enterprise and Enterprise Flash tiers will be retired on March 31, 2027.",
"reason": "Claimet lister Azure Cache for Redis Enterprise som et gjeldende vektor-DB-alternativ, men den tieren er under utfasing, og den kanoniske opplistingen «What are my other options for storing and searching vectors?» i Azure Managed Redis-dokumentasjonen nevner den ikke (kun Azure Managed Redis, Azure AI Search, Azure Cosmos DB og Azure Database for PostgreSQL).",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#5",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 82,
"rule": "R2",
"claim": "APIM-policyene llm-semantic-cache-lookup og azure-openai-semantic-cache-lookup bruker score-threshold som en semantisk avstand (prompts med score over terskelen bruker ikke cachen), og Microsofts eget eksempel bruker score-threshold=\"0.15\".",
"evidence_url": "https://learn.microsoft.com/azure/api-management/api-management-policies#ai-gateway",
"evidence_quote": "| [Get cached responses of large language model API requests](https://learn.microsoft.com/azure/api-management/llm-semantic-cache-lookup-policy) | Performs lookup in large language model API cache using semantic search and returns a valid cached response when available. | Yes | Yes | Yes | Yes | No |",
"reason": "Semantikken (score over terskel gir ikke cache-treff) og eksempelet score-threshold=\"0.15\" stemmer, men den kanoniske policy-opplistingen (både AI gateway- og Caching-tabellen) inneholder kun llm-semantic-cache-lookup/store — azure-openai-semantic-cache-lookup er fraværende, og URL-en til den policy-siden omdirigerer nå til llm-semantic-cache-lookup, så det navnet er superseded.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "0.15",
"type": "number"
}
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#10",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 271,
"rule": "R4",
"claim": "text-embedding-ada-002 er legacy og bør unngås for nye prosjekter.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/concepts/model-retirement-schedule#foundry-models-sold-by-azure",
"evidence_quote": "| text-embedding-ada-002 | 2 | GA | 2028-02-09 | — |",
"reason": "Den gjeldende livssyklus-raden gir text-embedding-ada-002 status GA (ikke Legacy eller Deprecated) med retirement 2028-02-09 og uten anbefalt erstatning, så påstanden om legacy-status er ikke i samsvar med dagens rad.",
"op": "O3",
"code": "CONTEXT_MISMATCH",
"detail": {
"token": "002",
"replacement": "2",
"would_have_been": "| `text-embedding-ada-2` | 1536 | ~80 NOK | Legacy, unngå for nye prosjekter |"
}
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#11",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 305,
"rule": "R3",
"claim": "Azure Managed Redis for semantic caching opprettes med SKU Enterprise_E10 og modulen RediSearch (az redis create --sku Enterprise_E10 --redis-module RediSearch).",
"evidence_url": "https://learn.microsoft.com/azure/redis/scripts/create-manage-cache",
"evidence_quote": "Azure Managed Redis uses the Azure CLI [az redisenterprise](/en-us/cli/azure/redisenterprise) commands.",
"reason": "Azure Managed Redis opprettes med az redisenterprise create og SKU-er av typen Balanced_B1 / Memory Optimized / Compute Optimized; az redis create gjelder Azure Cache for Redis Basic/Standard/Premium, og Enterprise_E10 er en Azure Cache for Redis Enterprise-SKU — hele kommando- og SKU-rammen i claimet er erstattet.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "10",
"type": "number"
}
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#14",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 436,
"rule": "R8",
"claim": "Azure Managed Redis og Azure OpenAI er tilgjengelig i regionene Norway East | Norway West, slik at data kan forbli i Norge/EU.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure-region-availability#standard-regional",
"evidence_quote": "| **Model** | **Version** | **francecentral** | **germanywestcentral** | **norwayeast** | **polandcentral** | **spaincentral** | **swedencentral** | **switzerlandnorth** | **uksouth** | **westeurope**|",
"reason": "Azure Managed Redis er tilgjengelig i både Norway East og Norway West, men den kanoniske regionstabellen for Foundry-modeller solgt av Azure har ingen norwaywest-kolonne i noen deployment-type — Azure OpenAI finnes kun i Norway East, så den ene lastbærende delen av claimet er feil.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#15",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 492,
"rule": "R3",
"claim": "Azure Managed Redis-tiers for semantic caching er Memory Optimized 1GB | Memory Optimized 10GB | Memory Optimized 50GB | Compute Optimized.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/redis/overview",
"evidence_quote": "| Size (GB) | 12 - 1920 | 0.5 - 960 | 3 - 720 | 250 - 4500 |",
"reason": "Azure Managed Redis har fire tiers (Memory Optimized, Balanced, Compute Optimized, Flash Optimized), og Memory Optimized starter på 12 GB — «Memory Optimized 1GB | 10GB | 50GB» finnes ikke, og Balanced/Flash Optimized mangler, så tier-rammen claimet forutsetter samsvarer ikke med gjeldende side.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#16",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 503,
"rule": "R8",
"claim": "Azure OpenAI lisensieres pay-per-token som PTU eller Consumption, uten ekstra lisenser for caching.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/concepts/provisioned-throughput-billing#how-ptu-billing-works",
"evidence_quote": "you're billed hourly based on the number of Provisioned Throughput Units (PTUs) you deploy, rather than the number of tokens consumed.",
"reason": "PTU er nettopp ikke pay-per-token, og Azure OpenAI-deployment-typene heter Standard (pay-per-call) og Provisioned — «Consumption» er ikke en Azure OpenAI-faktureringsmodell (det er en APIM-tier), så to lastbærende deler av claimet motsies.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#17",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 504,
"rule": "(none)",
"claim": "Azure API Management inkluderer semantic cache-policyene fra tier Basic v2 og oppover.",
"evidence_url": "https://learn.microsoft.com/azure/api-management/api-management-policies#ai-gateway",
"evidence_quote": "| [Get cached responses of large language model API requests](https://learn.microsoft.com/azure/api-management/llm-semantic-cache-lookup-policy) | ... | Yes | Yes | Yes | Yes | No |",
"reason": "Policyene er tilgjengelige i Classic, V2, Consumption og self-hosted gateways, og enable-artikkelen sier «APPLIES TO: All API Management tiers» — terskelen «fra Basic v2 og oppover» er dermed feil (bl.a. Consumption og de klassiske tierne er inkludert).",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "2",
"type": "number"
}
},
{
"id": "ms-ai-security/cost-optimization/semantic-caching-patterns.md#18",
"file": "skills/ms-ai-security/references/cost-optimization/semantic-caching-patterns.md",
"line": 505,
"rule": "R8",
"claim": "Azure Managed Redis faktureres pay-per-hour per tier, og RediSearch er inkludert i Enterprise-tier.",
"evidence_url": "https://learn.microsoft.com/azure/redis/tutorial-semantic-cache",
"evidence_quote": "| Azure Managed Redis tier | RediSearch support | | --- | --- | | Memory Optimized | Yes | | Balanced | Yes | | Compute Optimized | Yes | | Flash Optimized | No |",
"reason": "Azure Managed Redis har ingen «Enterprise-tier» — RediSearch følger tierne Memory Optimized/Balanced/Compute Optimized (Enterprise er en clustering policy, ikke en tier), så den lastbærende delen om Enterprise-tier er feil; pris-per-time-delen er dessuten kun oppgitt på den JS-rendrede pris-siden og ikke bekreftet på Learn.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/cost-optimization/small-language-models-economics.md#2",
"file": "skills/ms-ai-security/references/cost-optimization/small-language-models-economics.md",
"line": 49,
"rule": "R2",
"claim": "Microsofts Phi-serie av små språkmodeller består av Phi-4-mini | Phi-4-multimodal | Phi-3-small | Phi-3-medium | Phi-2.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/concepts/retired-models",
"evidence_quote": "Phi-3-small-8k-instruct | August 30, 2025 | Phi-4-mini-instruct",
"reason": "Den kanoniske Phi-oversikten i Foundry lister i dag Phi-4, Phi-4-mini-instruct, Phi-4-mini-reasoning, Phi-4-multimodal-instruct og Phi-4-reasoning; Phi-3-small, Phi-3-medium og Phi-2 er fraværende, og hele Phi-3-familien er ført opp som pensjonert 30. august 2025.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/small-language-models-economics.md#7",
"file": "skills/ms-ai-security/references/cost-optimization/small-language-models-economics.md",
"line": 53,
"rule": "R4",
"claim": "Phi-3-small har 7 milliarder parametere og støtter 128 000 tokens input-lengde.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/concepts/retired-models",
"evidence_quote": "Phi-3-small-128k-instruct | August 30, 2025 | Phi-4-mini-instruct",
"reason": "Parametertallet 7 milliarder står på AKS-konseptsiden, men 128 000-tokens-varianten (Phi-3-small-128k-instruct) er pensjonert 30. august 2025 og ingen gjeldende Learn-side oppgir 128 000 tokens for Phi-3-small — påstanden matcher kun en historisk rad.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"3",
"7",
"128",
"000"
]
}
},
{
"id": "ms-ai-security/cost-optimization/small-language-models-economics.md#8",
"file": "skills/ms-ai-security/references/cost-optimization/small-language-models-economics.md",
"line": 53,
"rule": "R2",
"claim": "Phi-3-small, Phi-3-medium og Phi-2 er alle GA i Azure.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/concepts/retired-models",
"evidence_quote": "Phi-3-medium-4k-instruct | August 30, 2025 | Phi-4",
"reason": "Den kanoniske livssyklusoversikten viser at Phi-3-medium og Phi-3-small er pensjonert 30. august 2025, og Phi-2 finnes ikke i noen gjeldende Foundry-katalogtabell — ingen av de tre er GA i Azure i dag.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"3",
"2"
]
}
},
{
"id": "ms-ai-security/cost-optimization/small-language-models-economics.md#9",
"file": "skills/ms-ai-security/references/cost-optimization/small-language-models-economics.md",
"line": 54,
"rule": "R4",
"claim": "Phi-3-medium har 14 milliarder parametere og støtter 128 000 tokens input-lengde.",
"evidence_url": "https://learn.microsoft.com/azure/foundry/openai/concepts/retired-models",
"evidence_quote": "Phi-3-medium-128k-instruct | August 30, 2025 | Phi-4",
"reason": "14 milliarder parametere står på AKS-konseptsiden, men 128k-varianten av Phi-3-medium er pensjonert 30. august 2025, og ingen gjeldende side oppgir 128 000 tokens input for Phi-3-medium.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"3",
"14",
"128",
"000"
]
}
},
{
"id": "ms-ai-security/cost-optimization/small-language-models-economics.md#15",
"file": "skills/ms-ai-security/references/cost-optimization/small-language-models-economics.md",
"line": 244,
"rule": "(none)",
"claim": "Azure App Service Phi-4 sidecar er GA.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/app-service/tutorial-ai-slm-dotnet",
"evidence_quote": "In the sidecar extension options, select **AI: phi-4-q4-gguf (Experimental)**.",
"reason": "Den kanoniske tutorialen merker sidecar-utvidelsen «(Experimental)» i portalvalget, altså det motsatte av GA-status.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "4",
"type": "number"
}
},
{
"id": "ms-ai-security/cost-optimization/small-language-models-economics.md#17",
"file": "skills/ms-ai-security/references/cost-optimization/small-language-models-economics.md",
"line": 285,
"rule": "R2",
"claim": "Microsoft Foundry tilbyr deployment-typene Serverless API | Managed Online Endpoints | Global Standard, der Global Standard gir fungibel kvote på tvers av regioner.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/deployment-types",
"evidence_quote": "Global Standard | GlobalStandard | Global Provisioned | GlobalProvisionedManaged | Global Batch | GlobalBatch | Data Zone Standard | DataZoneStandard | Data Zone Provisioned | DataZoneProvisionedManaged | Data Zone Batch | DataZoneBatch | Standard | Standard | Regional Provisioned | ProvisionedManaged | Developer | DeveloperTier",
"reason": "Den kanoniske enumereringen av deployment-typer i Foundry inneholder verken «Serverless API» eller «Managed Online Endpoints»; i tillegg beskriver Learn fungibel kvote som fungibel på tvers av modeller (fungible provisioned throughput), ikke på tvers av regioner.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/small-language-models-economics.md#18",
"file": "skills/ms-ai-security/references/cost-optimization/small-language-models-economics.md",
"line": 286,
"rule": "R2",
"claim": "Managed Online Endpoints i Microsoft Foundry krever dedikert VM av typen Standard_DS3_v2 eller bedre.",
"evidence_url": "https://learn.microsoft.com/azure/machine-learning/reference-managed-online-endpoints-vm-sku-list",
"evidence_quote": "standardDSv2Family | STANDARD_DS1_V2 | - | Cpu | 0 | 1 | -",
"reason": "Den kanoniske SKU-listen for managed online endpoints støtter flere SKU-er mindre enn Standard_DS3_v2 (blant annet STANDARD_DS1_V2 og STANDARD_DS2_V2), så «krever Standard_DS3_v2 eller bedre» er ikke et dokumentert minstekrav — Learn advarer bare at små SKU-er «may be too small for bigger models».",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"3",
"2"
]
}
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#6",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 55,
"rule": "R8",
"claim": "Azure AI Search-benchmark: baseline float32 gir 21.36 MB storage og 4.83 MB vector index; scalar quantization gir 17.76 MB storage og 1.22 MB vector index (75 % reduksjon); binary quantization gir 4.92 MB storage og 1.22 MB vector index (77 % total reduksjon).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-configure-compression-storage",
"evidence_quote": "| compressiontest-baseline | 21.3613 MB | 4.8277 MB | | compressiontest-scalar-compression | 17.7604 MB | 1.2242 MB | | compressiontest-narrow | 16.5567 MB | 2.4254 MB | | compressiontest-no-stored | 10.9224 MB | 4.8277 MB | | compressiontest-all-options | 4.9192 MB | 1.2242 MB |",
"reason": "Baseline- og scalar-tallene stemmer, men 4.92 MB / 1.22 MB er raden compressiontest-all-options — tabellen har ingen rad for binary quantization, så den bærende merkelappen på tredje del av påstanden er feil.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"32",
"21.36",
"4.83",
"17.76",
"1.22",
"75 %",
"4.92",
"77 %"
]
}
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#7",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 58,
"rule": "R8",
"claim": "Alle komprimeringsteknikker kombinert gir 4.92 MB storage og 1.22 MB vector index, tilsvarende 92,5 % reduksjon i vector index-størrelse.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-configure-compression-storage",
"evidence_quote": "| compressiontest-baseline | 21.3613 MB | 4.8277 MB | ... | compressiontest-all-options | 4.9192 MB | 1.2242 MB |",
"reason": "MB-tallene stemmer med compressiontest-all-options, men tabellens egne tall gir 4.8277 → 1.2242 MB, altså ca. 74,6 % reduksjon i vector index (og 77 % i total storage) — ikke 92,5 %; 92,5 % opptrer kun som et «up to»-tall i en bloggtittel om vector costs.",
"op": "O3",
"code": "MULTI_VALUE_TOKEN",
"detail": {
"candidates": [
"4.92",
"1.22",
"92,5 %"
]
}
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#11",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 73,
"rule": "R8",
"claim": "Azure AI Search lagrer vektorer i to kopier: index copy (i minne, brukt til query execution) | stored copy (på disk, brukt til retrieval i query response).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-storage-options",
"evidence_quote": "Azure AI Search stores multiple copies of vector fields that are used in specific workloads. ... How vector fields are stored: Vectors in the HNSW graph ... (Essential) | Source vectors received during document indexing (JSON data) ... Set `stored` property to false. | Original full-precision vectors (binary data) ... Set `rescoringOptions.rescoreStorageMethod` property to `discardOriginals`.",
"reason": "Kilden lister tre instanser (query-indeksen, source-/stored-kopien og de originale full-precision-vektorene for rescoring), ikke to; antallet er bærende i en kostnadssammenheng fordi den tredje kopien øker lagringsbehovet med omtrent størrelsen på den komprimerte indeksen.",
"op": "O3",
"code": "MULTI_PART_CLAIM"
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#21",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 266,
"rule": "R4",
"claim": "Azure AI Search REST-API for oppretting av indeks og vektorsøk bruker `api-version=2025-09-01`.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-quantization",
"evidence_quote": "POST https://[servicename].search.windows.net/indexes?api-version=2026-04-01",
"reason": "Både indeksoppretting og vektorspørringer i gjeldende dokumentasjon bruker api-version=2026-04-01, som migreringssiden kaller «the latest stable REST API version»; 2025-09-01 er en eldre stabil versjon og ikke den gjeldende raden.",
"op": "O1",
"code": "PROVEN",
"proposal": {
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 266,
"token": "2025-09-01",
"replacement": "2026-04-01",
"type": "iso_date",
"before": "POST https://[service].search.windows.net/indexes?api-version=2025-09-01",
"after": "POST https://[service].search.windows.net/indexes?api-version=2026-04-01",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/vector-search-how-to-quantization",
"evidence_quote": "POST https://[servicename].search.windows.net/indexes?api-version=2026-04-01"
}
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#23",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 319,
"rule": "R4",
"claim": "Azure OpenAI embeddings med dimensions-parameter (MRL) brukes med api_version `2024-02-01`.",
"evidence_url": "https://learn.microsoft.com/azure/ai-foundry/openai/reference",
"evidence_quote": "POST https://{endpoint}/openai/deployments/{deployment-id}/embeddings?api-version=2024-10-21 ... | dimensions | integer | The number of dimensions the resulting output embeddings should have. Only supported in `text-embedding-3` and later models.",
"reason": "dimensions-parameteren er dokumentert, men gjeldende api-version for embeddings-endepunktet er 2024-10-21 i den stabile referansen (og v1/preview i den nye Foundry-API-en); 2024-02-01 er en avløst versjon.",
"op": "O1",
"code": "PROVEN",
"proposal": {
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 318,
"token": "2024-02-01",
"replacement": "2024-10-21",
"type": "iso_date",
"before": " api_version=\"2024-02-01\"",
"after": " api_version=\"2024-10-21\"",
"evidence_url": "https://learn.microsoft.com/azure/ai-foundry/openai/reference",
"evidence_quote": "POST https://{endpoint}/openai/deployments/{deployment-id}/embeddings?api-version=2024-10-21 ... | dimensions | integer | The number of dimensions the resulting output embeddings should have. Only supported in `text-embedding-3` and later models."
}
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#25",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 348,
"rule": "R8",
"claim": "Vector extension i Azure SQL Database er i preview, støtter float32-vektorer, men ikke native quantization.",
"evidence_url": "https://learn.microsoft.com/sql/t-sql/functions/vector-search-transact-sql?view=sql-server-ver17",
"evidence_quote": "**Advanced quantization**: Vector quantization techniques have been integrated to provide better storage efficiency and faster query performance, with these optimizations being transparent to users",
"reason": "Preview-statusen og float32-lagringen (single-precision 4-byte elementer) stemmer, men den bærende delen «ikke native quantization» motsies: nyeste vektorindekser i Azure SQL Database har integrert kvantisering, og float16-vektorer støttes bak PREVIEW_FEATURES.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "32",
"type": "number"
}
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#26",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 367,
"rule": "R2",
"claim": "Azure AI Search støtter regionene Norway East og Norway West med full data residency.",
"evidence_url": "https://learn.microsoft.com/en-us/azure/search/search-region-support",
"evidence_quote": "Europe: France Central | Germany West Central | Italy North | Norway East | North Europe | Poland Central | Spain Central | Sweden Central | Switzerland North | Switzerland West | UK South | UK West | West Europe",
"reason": "Den kanoniske regionlisten for Azure AI Search har Norway East, men Norway West forekommer ikke i noen europeisk rad — fraværet i den uttømmende oppregningen motbeviser eksistenspåstanden.",
"op": "O3",
"code": "NO_VALUE_TOKEN"
},
{
"id": "ms-ai-security/cost-optimization/vector-storage-cost-optimization.md#34",
"file": "skills/ms-ai-security/references/cost-optimization/vector-storage-cost-optimization.md",
"line": 454,
"rule": "R2",
"claim": "Microsoft Foundry unified billing er i preview (februar 2026).",
"evidence_url": "https://learn.microsoft.com/en-us/azure/foundry/concepts/manage-costs",
"evidence_quote": "Common billing approaches include: - **Pay-as-you-go (Serverless API):** ... - **Commitment tiers:** ... ## Chargeback with project-level cost attribution (Preview)",
"reason": "Den kanoniske Foundry-kostnadssiden oppregner faktureringsmodellene og den ene preview-funksjonen (project-level cost attribution); noen «unified billing» i preview finnes ikke der eller i søk på learn.microsoft.com.",
"op": "O3",
"code": "NOT_VERBATIM",
"detail": {
"token": "2026",
"type": "number"
}
}
]
}
Reference in a new issue View git blame Copy permalink