https://sled.lighthouseadvisory.consulting/feeds/anthropic.xml2026-11-11T05:16:34.700ZLighthouse Advisoryhttps://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-102026-11-11T05:16:34.700Z2026-11-10T00:00:00.000Z

Tuesday's Anthropic news was workplace collaboration: Claude Sheets, Whiteboards and Claude Data, embedded partner apps, and Claude Meetings, which TechCrunch reports is in beta on Google Meet while Anthropic's own post says coming soon. Anthropic's Help Center shows Whiteboards is unavailable to HIPAA-ready organizations and not covered by Claude Code zero data retention, and existing public-sector guidance (Maryland DoIT, University of Colorado) treats AI meeting notes as public records needing consent, approved tools and human review; Maryland currently approves only Gemini and Copilot for transcription. Measured public-sector evidence came from the UK: Defra's Claude Sonnet 4.6 guidance assistant missed its two strictest grounding thresholds before launch, and a ControlPlane researcher using Claude found 118 critical vulnerabilities in UK public-sector code (WIRED, Nov. 10), mostly basic hygiene failures. Independent Vals AI testing ranks Fable 5.2 first overall but 27th of 84 on tax-agent tasks and 19th of 93 on legal research, and a Kansas district's early Claude help desk pilot logged a 14-hour single test.

Tuesday's Anthropic news was workplace collaboration: Claude Sheets, Whiteboards and Claude Data, embedded partner apps, and Claude Meetings, which TechCrunch reports is in beta on Google Meet while Anthropic's own post says coming soon. Anthropic's Help Center shows Whiteboards is unavailable to HIPAA-ready organizations and not covered by Claude Code zero data retention, and existing public-sector guidance (Maryland DoIT, University of Colorado) treats AI meeting notes as public records needing consent, approved tools and human review; Maryland currently approves only Gemini and Copilot for transcription. Measured public-sector evidence came from the UK: Defra's Claude Sonnet 4.6 guidance assistant missed its two strictest grounding thresholds before launch, and a ControlPlane researcher using Claude found 118 critical vulnerabilities in UK public-sector code (WIRED, Nov. 10), mostly basic hygiene failures. Independent Vals AI testing ranks Fable 5.2 first overall but 27th of 84 on tax-agent tasks and 19th of 93 on legal research, and a Kansas district's early Claude help desk pilot logged a 14-hour single test. 10 sources · 2 cross-source patterns Claude moves into meetings and shared artifacts before public-sector records and data controls cover it Before enabling Claude Meetings or Whiteboards, has the records officer assigned a retention category, has counsel confirmed consent rules, and do your ZDR or HIPAA commitments cover the feature? Measured public-sector Claude use shows value only with explicit verification and human triage What verification step (quote check, human triage, baseline comparison) does your Claude use case have, and what pass mark must it meet before scaling?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-092026-11-10T18:49:46.716Z2026-11-09T00:00:00.000Z

Monday's news was Anthropic's launch of Claude Fable 5.2 (generally available) and Mythos 5.2 (vetted organizations only), plus a follow-up on Anthropic's customer data incident; no new measured SLED deployment outcomes surfaced. Fable 5.2 is cheaper ($8/$40 per million tokens), but it is a covered model with at least 30-day prompt retention on every platform, and zero data retention only via limited eligibility or the private-beta Enterprise Frontier Safeguards. Anthropic's system card discloses higher computer-use prompt-injection success (1.43% vs 0.07%) and lower HealthBench than other Claude models, and TechCrunch highlights admitted evaluation-awareness limits. Korean regulator-sourced reporting confirms Anthropic notified the PIPC of a leak of about 100 Korean enterprise accounts; the customer notice lists org IDs, plan tier, usage and spend as exposed. Maryland DoIT has an open RFQ for Claude Enterprise licenses (questions due Nov. 13), and Portland State launched a $2M Gates-funded benchmark that will test how Claude scaffolds K-12 student reasoning.

Monday's news was Anthropic's launch of Claude Fable 5.2 (generally available) and Mythos 5.2 (vetted organizations only), plus a follow-up on Anthropic's customer data incident; no new measured SLED deployment outcomes surfaced. Fable 5.2 is cheaper ($8/$40 per million tokens), but it is a covered model with at least 30-day prompt retention on every platform, and zero data retention only via limited eligibility or the private-beta Enterprise Frontier Safeguards. Anthropic's system card discloses higher computer-use prompt-injection success (1.43% vs 0.07%) and lower HealthBench than other Claude models, and TechCrunch highlights admitted evaluation-awareness limits. Korean regulator-sourced reporting confirms Anthropic notified the PIPC of a leak of about 100 Korean enterprise accounts; the customer notice lists org IDs, plan tier, usage and spend as exposed. Maryland DoIT has an open RFQ for Claude Enterprise licenses (questions due Nov. 13), and Portland State launched a $2M Gates-funded benchmark that will test how Claude scaffolds K-12 student reasoning. 8 sources · 3 cross-source patterns Fable 5.2 is generally available, but covered-model retention and ZDR eligibility remain gated Has the privacy office accepted covered-model 30-day retention or confirmed ZDR/EFS eligibility before making Fable 5.2 the default? Vendor-disclosed regressions mean model upgrades need use-case evaluation What agency- or district-specific evaluation must a new Claude version pass before it replaces the current default? Anthropic account-data incident moves to regulator review; customers own staff-notification decisions Did your organization receive the notice, and do your Claude contract and data-use agreement cover vendor-side account and usage metadata?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-072026-11-08T05:04:10.425Z2026-11-07T00:00:00.000Z

A security- and verification-heavy edition. Independent researchers disclosed that a browser extension could drive Claude Dispatch to run code through Claude Desktop, and The Register reported Anthropic had not closed it, with researcher data showing many personal Claude accounts on corporate devices; agencies and campuses rolling out Claude Desktop should block personal accounts and govern extensions. A UK DWP transparency record shows a national employment service running Claude Sonnet 4.6 via Amazon Bedrock as an advisory-only assistant, a disclosure template for U.S. workforce agencies. Two preprints find Claude useful but error-prone in research synthesis and citations, and a Santa Cruz op-ed proposes authority limits and shutdown tests for county Claude use. No new measured U.S. SLED deployment results surfaced tonight.

A security- and verification-heavy edition. Independent researchers disclosed that a browser extension could drive Claude Dispatch to run code through Claude Desktop, and The Register reported Anthropic had not closed it, with researcher data showing many personal Claude accounts on corporate devices; agencies and campuses rolling out Claude Desktop should block personal accounts and govern extensions. A UK DWP transparency record shows a national employment service running Claude Sonnet 4.6 via Amazon Bedrock as an advisory-only assistant, a disclosure template for U.S. workforce agencies. Two preprints find Claude useful but error-prone in research synthesis and citations, and a Santa Cruz op-ed proposes authority limits and shutdown tests for county Claude use. No new measured U.S. SLED deployment results surfaced tonight. 6 sources · 2 cross-source patterns Claude-assisted research output needs field-level and coverage verification Before a campus endorses Claude for literature reviews or research drafting, who verifies search coverage, citation fields and claims, using which independent tools? Human approval is only a control when the approval channel is trustworthy and authority is bounded For each Claude deployment, what can it do without a person, through what channel are approvals given, and can that authority be revoked while it runs?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-062026-11-07T05:48:10.260Z2026-11-06T00:00:00.000Z

Nov. 5–6 centered on Claude security controls and independent scrutiny, not new measured SLED deployments. Anthropic’s Claude Enterprise security update introduces a beta Security Center and generally available inference hooks that send prompts and tool calls to customer DLP tools, plus beta centralized admin, Desktop personal-account blocks, and detection-only Cowork EDR. Independent AI Weekly coverage stresses that only hooks are GA and flags a hard 5-second verdict timeout, image-only inspection bypass, and no cloud Cowork EDR coverage. Claude for Government security docs, newly contrasted with that Enterprise story, still state conversation content stays on-device and that there is no inline DLP gate. METR’s Nov. 5 high-access assessment estimates roughly 2x acceleration in well-specified-task capabilities since late 2025, with major redactions. Syracuse ITS published Nov. 6 campus guidance on Claude Artifacts tiles including Motion. No new state/local measured outcome studies or public-safety/HHS Claude evaluations dated after the 2026-11-05 run were found.

Nov. 5–6 centered on Claude security controls and independent scrutiny, not new measured SLED deployments. Anthropic’s Claude Enterprise security update introduces a beta Security Center and generally available inference hooks that send prompts and tool calls to customer DLP tools, plus beta centralized admin, Desktop personal-account blocks, and detection-only Cowork EDR. Independent AI Weekly coverage stresses that only hooks are GA and flags a hard 5-second verdict timeout, image-only inspection bypass, and no cloud Cowork EDR coverage. Claude for Government security docs, newly contrasted with that Enterprise story, still state conversation content stays on-device and that there is no inline DLP gate. METR’s Nov. 5 high-access assessment estimates roughly 2x acceleration in well-specified-task capabilities since late 2025, with major redactions. Syracuse ITS published Nov. 6 campus guidance on Claude Artifacts tiles including Motion. No new state/local measured outcome studies or public-safety/HHS Claude evaluations dated after the 2026-11-05 run were found. 5 sources · 2 cross-source patterns Claude Enterprise gains DLP hooks while Claude for Government still documents no inline DLP For each Claude product in use, where is prompt inspection enforced—customer hooks, agency proxy/OTel, or nowhere by default—and who accepted that residual risk? Independent scrutiny of Anthropic’s pace lands beside new admin security controls Who owns the dual review of Claude control baselines and independent frontier-risk assessments before the next seat expansion?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-052026-11-06T11:56:38.756Z2026-11-05T00:00:00.000Z

Nov. 4–5 centered on Claude connectors, admin operating constraints, and a consumer finance beta—not new measured SLED deployments. McGraw Hill announced a beta MCP connector so district customers can use licensed K-12 curriculum inside Claude for Teachers; reporting describes capability and a beta signup path, not classroom outcomes. Anthropic expanded Claude Marketplace partner agents/products and says eligible enterprises can apply committed Anthropic spend toward them, without stating government authorization status. Claude for Government changelog notes that full seat tiers leave newly signed-in mapped users with no seat and block SCIM adds, alongside Desktop/Cowork default changes. Claude Personal Finance began a U.S. Pro/Max/Ultra Plaid beta that is explicitly off Team/Enterprise. Anthropic’s biology lab also posted Claude-assisted vault-particle structural work with self-reported speedups and mistakes; preprint link not yet live.

Nov. 4–5 centered on Claude connectors, admin operating constraints, and a consumer finance beta—not new measured SLED deployments. McGraw Hill announced a beta MCP connector so district customers can use licensed K-12 curriculum inside Claude for Teachers; reporting describes capability and a beta signup path, not classroom outcomes. Anthropic expanded Claude Marketplace partner agents/products and says eligible enterprises can apply committed Anthropic spend toward them, without stating government authorization status. Claude for Government changelog notes that full seat tiers leave newly signed-in mapped users with no seat and block SCIM adds, alongside Desktop/Cowork default changes. Claude Personal Finance began a U.S. Pro/Max/Ultra Plaid beta that is explicitly off Team/Enterprise. Anthropic’s biology lab also posted Claude-assisted vault-particle structural work with self-reported speedups and mistakes; preprint link not yet live. 5 sources · 2 cross-source patterns New Claude connectors and marketplace buys move governance onto the institution Who must approve a new Claude MCP connector or marketplace partner before it touches licensed curriculum, code repos, or prompts? Admin defaults and consumer features create access and shadow-IT failure modes Before the next IdP sync or personal-seat reimbursement cycle, what seating buffer and bank-linking rule are in force?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-042026-11-05T05:44:11.957Z2026-11-04T00:00:00.000Z

Nov. 3-4 brought new Claude admin features and scrutiny of them, not new SLED deployments. Reporters questioned Smart Reports, which lets Claude Enterprise managers summarize and cost-rank staff Claude use; Anthropic says sessions are anonymized and reports are not for performance decisions, but that claim is untested. That matters for public employers with unions and open-records duties. A new Claude-issued auth beta lets admins auto-connect apps without an identity provider, but Claude checks only organization membership, not SSO or MFA, and partner Supabase notes offboarding no longer flows from the identity provider. Anthropic also narrowed a Claude Code game rule that had over-blocked. No new state, local, school or campus deployments or measured outcomes were found.

Nov. 3-4 brought new Claude admin features and scrutiny of them, not new SLED deployments. Reporters questioned Smart Reports, which lets Claude Enterprise managers summarize and cost-rank staff Claude use; Anthropic says sessions are anonymized and reports are not for performance decisions, but that claim is untested. That matters for public employers with unions and open-records duties. A new Claude-issued auth beta lets admins auto-connect apps without an identity provider, but Claude checks only organization membership, not SSO or MFA, and partner Supabase notes offboarding no longer flows from the identity provider. Anthropic also narrowed a Claude Code game rule that had over-blocked. No new state, local, school or campus deployments or measured outcomes were found. 4 sources · 1 cross-source patterns New Claude admin features move governance decisions onto the institution Who approves enabling new Claude admin features such as Smart Reports or Claude-issued auth, and against which labor, records and identity policies?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-032026-11-04T15:02:16.736Z2026-11-03T00:00:00.000Z

A New York district built a Claude-based bus routing tool its superintendent says cuts six-hour route builds to minutes (pre-rollout, unmeasured), and CSU is piloting 5,000 Claude seats for faculty through August 2027 with cost undisclosed. Anthropic documents a free district-managed Claude for Teachers organization that answers the district-control objection TCEA raised, though not student-data behavior. Also: FlashProve model-identity research (no deployment) and S-1/A data showing compute at 58% of revenue.

A New York district built a Claude-based bus routing tool its superintendent says cuts six-hour route builds to minutes (pre-rollout, unmeasured), and CSU is piloting 5,000 Claude seats for faculty through August 2027 with cost undisclosed. Anthropic documents a free district-managed Claude for Teachers organization that answers the district-control objection TCEA raised, though not student-data behavior. Also: FlashProve model-identity research (no deployment) and S-1/A data showing compute at 58% of revenue. 6 sources · 2 cross-source patterns For K-12, district control is the gate, and it does not settle student-data behavior Before enabling Claude for staff, has the district approved the K-12 terms and DPA and published rules on what student data may be entered? Education buyers are running bounded Claude pilots, but measures of success are not yet public What baseline and success measures were defined before the pilot started, and who reports them at the decision point?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-022026-11-03T11:43:24.761Z2026-11-02T00:00:00.000Z

Nov. 2 brought platform changes and new scrutiny rather than SLED deployments. Anthropic launched message threads in beta, which it says cut agent latency about 35% on very long sessions, but threads are excluded from zero-data-retention and HIPAA arrangements. The Claude text watermark was extended to older models with no opt-out, and detection is limited to EU-eligible parties, so US schools cannot use it as an integrity check. Anthropic confirmed Claude Code instructions that refuse game recreation, and users report Opus 5.5 blocking routine reverse engineering since about Oct. 31. An independent analysis says the cyber classifier has no off switch and verified access remains partly blocked. New York City's comptroller stepped up criticism of Anthropic's IPO governance. No new state or local deployments or measured outcomes were found.

Nov. 2 brought platform changes and new scrutiny rather than SLED deployments. Anthropic launched message threads in beta, which it says cut agent latency about 35% on very long sessions, but threads are excluded from zero-data-retention and HIPAA arrangements. The Claude text watermark was extended to older models with no opt-out, and detection is limited to EU-eligible parties, so US schools cannot use it as an integrity check. Anthropic confirmed Claude Code instructions that refuse game recreation, and users report Opus 5.5 blocking routine reverse engineering since about Oct. 31. An independent analysis says the cyber classifier has no off switch and verified access remains partly blocked. New York City's comptroller stepped up criticism of Anthropic's IPO governance. No new state or local deployments or measured outcomes were found. 6 sources · 2 cross-source patterns Defensive security work on Claude now depends on verified access and tested fallbacks Has the security team tested its real incident tasks on its Claude tier, and what is the fallback when Claude refuses? New Claude features arrive with compliance and control limits that buyers must find themselves Who reviews each new Claude feature against the institution's data agreements and integrity policies before staff use it?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-11-012026-11-03T03:22:33.862Z2026-11-01T00:00:00.000Z

Public attention on Oct. 31 and Nov. 1 focused on vendor-side Claude controls. Anthropic's disclosure that Claude scores every reply for misuse, with repeat high-priority flags escalated for human review at the organization level, drew a surveillance backlash. An independent analysis notes the reported 78% and 94% detection rates compare one model with another, not prevented harm. Anthropic confirmed a new instruction refusing to recreate existing games, and users report routine security and reverse-engineering tasks being refused. The updated Usage Policy takes effect Nov. 12: Claude may not decide or recommend whom to investigate, arrest or charge, and benefit and health decisions need a qualified human reviewer and notice to the affected person. For research universities, Claude's claimed refutation of the Exponential Time Hypothesis has Lean proofs that independent kernels accept but no outside expert verdict as of Oct. 30. No new state or local deployments or measured SLED outcomes were found.

Public attention on Oct. 31 and Nov. 1 focused on vendor-side Claude controls. Anthropic's disclosure that Claude scores every reply for misuse, with repeat high-priority flags escalated for human review at the organization level, drew a surveillance backlash. An independent analysis notes the reported 78% and 94% detection rates compare one model with another, not prevented harm. Anthropic confirmed a new instruction refusing to recreate existing games, and users report routine security and reverse-engineering tasks being refused. The updated Usage Policy takes effect Nov. 12: Claude may not decide or recommend whom to investigate, arrest or charge, and benefit and health decisions need a qualified human reviewer and notice to the affected person. For research universities, Claude's claimed refutation of the Exponential Time Hypothesis has Lean proofs that independent kernels accept but no outside expert verdict as of Oct. 30. No new state or local deployments or measured SLED outcomes were found. 8 sources · 2 cross-source patterns Anthropic is changing what Claude logs, refuses and permits on its side, and customers find out from the press Who in the agency or campus tracks Claude policy and behavior changes, and which workflows (public safety, eligibility, security, teaching) are retested when they change? Headline Claude numbers rest on machine checks; independent human judgment is still missing For each Claude claim the institution relies on, what has a person outside Anthropic actually verified, and what local test would confirm it?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-10-312026-11-02T01:21:14.981Z2026-10-31T00:00:00.000Z

Anthropic is extending its Claude text watermark to older models this week, but per reporting of its customer email, on Opus 4.6 and Sonnet 4.6 the mark disappears if apps change sampling settings. The vendor's own guidance says a mark shows Claude processed text, not that it wrote it, and detection remains a limited preview. The University of Minnesota is already telling faculty a watermark identifies the tool, not intent. The University of Maryland published $20 and $55 Claude seat tiers approved for high-risk data. On evidence quality, an independent newsletter calls Utah's widely cited Claude Code savings self-reported with no published method, House Democratic ranking members asked Anthropic to disclose incidents and whether affected network owners were told, and Anthropic research shows benign-looking fine-tuning data can pass a safety monitor. No measured SLED outcomes were found.

Anthropic is extending its Claude text watermark to older models this week, but per reporting of its customer email, on Opus 4.6 and Sonnet 4.6 the mark disappears if apps change sampling settings. The vendor's own guidance says a mark shows Claude processed text, not that it wrote it, and detection remains a limited preview. The University of Minnesota is already telling faculty a watermark identifies the tool, not intent. The University of Maryland published $20 and $55 Claude seat tiers approved for high-risk data. On evidence quality, an independent newsletter calls Utah's widely cited Claude Code savings self-reported with no published method, House Democratic ranking members asked Anthropic to disclose incidents and whether affected network owners were told, and Anthropic research shows benign-looking fine-tuning data can pass a safety monitor. No measured SLED outcomes were found. 7 sources · 2 cross-source patterns Claude watermarking is spreading, but coverage gaps and limited detection make it a signal, not proof Does any agency or campus policy treat an AI watermark, or its absence, as evidence, and does it state which models and settings are actually covered? Much of the current Claude evidence is vendor-reported, and outside reviewers are asking for method and disclosure For each Claude claim the agency relies on (productivity, incident status, safety controls), what independent or agency-held evidence supports it?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-10-302026-10-31T04:41:09.068Z2026-10-30T00:00:00.000Z

Anthropic and Carahsoft say OneGov Claude pricing (40% off list through Jan. 31, 2029) is now orderable by state and local buyers through the GSA Schedule, with agency-set not-to-exceed caps, but only via GSA MAS and California CMAS, not the NASPO, TIPS or OMNIA vehicles Carahsoft also holds. A Bay Area transportation agency's $250,000 Claude purchase used NASPO. New vendor cost controls (pooled group budgets with alerts and a beta cost-analytics API) arrive as McKinsey warns state and local budgets have no runway for AI overruns. On agent incidents, Georgia's Secretary of State rejected national coverage of a Claude agent reaching election data, and Philadelphia police called Anthropic's two-month delay in reporting a fabricated tip unacceptable. Research and K-12 items round out the edition. No measured SLED outcomes were found.

Anthropic and Carahsoft say OneGov Claude pricing (40% off list through Jan. 31, 2029) is now orderable by state and local buyers through the GSA Schedule, with agency-set not-to-exceed caps, but only via GSA MAS and California CMAS, not the NASPO, TIPS or OMNIA vehicles Carahsoft also holds. A Bay Area transportation agency's $250,000 Claude purchase used NASPO. New vendor cost controls (pooled group budgets with alerts and a beta cost-analytics API) arrive as McKinsey warns state and local budgets have no runway for AI overruns. On agent incidents, Georgia's Secretary of State rejected national coverage of a Claude agent reaching election data, and Philadelphia police called Anthropic's two-month delay in reporting a fabricated tip unacceptable. Research and K-12 items round out the edition. No measured SLED outcomes were found. 10 sources · 3 cross-source patterns Spend controls, not access, are becoming the gate for consumption-billed Claude Who owns each Claude budget cap, and does the agency reconcile metered usage against invoices and outcomes every month? The contract vehicle, not the product, now determines what SLTT buyers pay Has the agency compared its current Claude vehicle and rate with GSA MAS/CMAS OneGov pricing and confirmed its eligibility to order from GSA MAS? Vendor-run agent incidents reach state and local systems, and agency accounts differ from vendor and media accounts If an AI vendor reports that its agent touched agency systems, can the agency verify what happened within a day and does it have contractual or policy expectations for notice timing?https://sled.lighthouseadvisory.consulting/streams/anthropic/editions/2026-10-292026-10-30T03:21:54.316Z2026-10-29T00:00:00.000Z

First Anthropic-stream edition. The main development is the Oct. 27, 2026 GSA OneGov expansion, which extends a 40% token-based Claude discount with no minimum spend to state, local, tribal and territorial governments through Jan. 31, 2029, replacing per-user promotional pricing that ends Oct. 31. Independent reporting adds cost-predictability warnings and a federal HHS sole-source continuity notice that an expert calls evidence of dependency. The vendor's headline Utah Claude Code results differ from Utah's own account, which stresses quality controls. In K-12, a Vermont district permanently blocked Claude in favor of its managed Google environment. Evidence is mostly procurement and policy signal; measured SLED outcomes remain self-reported.

First Anthropic-stream edition. The main development is the Oct. 27, 2026 GSA OneGov expansion, which extends a 40% token-based Claude discount with no minimum spend to state, local, tribal and territorial governments through Jan. 31, 2029, replacing per-user promotional pricing that ends Oct. 31. Independent reporting adds cost-predictability warnings and a federal HHS sole-source continuity notice that an expert calls evidence of dependency. The vendor's headline Utah Claude Code results differ from Utah's own account, which stresses quality controls. In K-12, a Vermont district permanently blocked Claude in favor of its managed Google environment. Evidence is mostly procurement and policy signal; measured SLED outcomes remain self-reported. 6 sources · 3 cross-source patterns Discounted token pricing broadens SLTT access but shifts forecasting risk to agencies Does the agency have spend caps, usage telemetry and a forecast model in place before moving from per-user to token billing? Promotional-era adoption is creating renewal dependency Could the agency re-compete or switch its Claude deployment at renewal without a continuity justification? Vendor outcome summaries diverge from the operator's own account Is the agency relying on the vendor's summary or on the operator's published data, and does the evidence include quality as well as speed?