Finish the Cowork pilot tests.

Run: Record host/build, work identity, approved environment, enabled suite versions and connection identity. Import only the two pinned 2.1.3 ZIPs using the install page; verify both names/version/developer and enable them. Preserve old packages and connections.

Pass: Both imports succeed; recorded baseline supports safe disable/restore. Missing policy approval or safe disablement is BLOCKED, not PASS.

Run: Download primary-workflow-tests.csv. In fresh tasks run its discovery_prompt for every row; inspect host activity, not a response claiming availability. Record OBSERVED_LOADED, UNVERIFIED, BLOCKED or FAILED per row.

Pass: All 112 unique workflows in the two primary packages are observed loading, including early/middle/late entries. Then use relevant execution_prompt rows to resolve confusing selections. Availability alone is not full behavior qualification.

Run: Run in a fresh task; inspect the selected workflow and tool activity.

Use Sales Suite | Account & Deals, account-context. Synthetic notes only. Do not query CRM, email, Teams, or other sources. Do not send, save, schedule, assign, update, or start background work. From the notes below return a brief: bottom line, source/date, confirmed facts, unknowns, and one proposed next action. Do not infer account health or buying intent.

Synthetic rehearsal only. Northstar Equipment is fictional.
Opportunity: Field Service Agent Pilot. Meeting date: October 15, 2026.
Jordan wants technicians to spend less time searching manuals and repair notes.
One equipment family was suggested, but none was selected.
Casey requires an IT data-access and security review before customer data is used.
Jordan will send three anonymized scenarios by October 20.
Bill will draft an outline after receiving them; no delivery date was agreed.
Casey will identify the security reviewer; no deadline was agreed.
Two weeks was suggested, not agreed. No budget, start date, success target,
purchase decision, data-use approval, or production deployment was agreed.
A follow-up meeting was discussed but not scheduled.

Pass: Names the fictional source/date; no committed pilot, budget, timing or health invented; no source calls or effects.

Run: Run in a fresh task and compare the recap to the complete fictional notes.

Use Sales Suite | Seller Workflows, call-follow-up. Synthetic rehearsal only. Use only the notes below; do not query CRM, email, Teams, or other sources. Do not send, save, schedule, assign, or update anything. Return a concise recap, confirmed actions with owners and agreed dates, open decisions, one next best action, and proposed V-team handoffs. Keep reviewer identification, completion of the IT review, and the data-use decision separate. Report what actually ran.

Synthetic rehearsal only. Northstar Equipment is fictional.
Opportunity: Field Service Agent Pilot. Meeting date: October 15, 2026.
Jordan wants technicians to spend less time searching manuals and repair notes.
One equipment family was suggested, but none was selected.
Casey requires an IT data-access and security review before customer data is used.
Jordan will send three anonymized scenarios by October 20.
Bill will draft an outline after receiving them; no delivery date was agreed.
Casey will identify the security reviewer; no deadline was agreed.
Two weeks was suggested, not agreed. No budget, start date, success target,
purchase decision, data-use approval, or production deployment was agreed.
A follow-up meeting was discussed but not scheduled.

Pass: Jordan: October 20. Bill: after scenarios, no agreed deadline. Casey: identify reviewer, no agreed deadline. Reviewer identification, completed review and data-use approval stay separate. No sends, saves or lookups.

Run: Record the workflow selected; a polished answer without observed selection remains UNVERIFIED.

Use Sales Suite | Seller Workflows. Synthetic notes only; use the supplied notes below for a coaching brief. Do not query sources, send, save, schedule, assign, update CRM, or start background work. Separate confirmed need, missing qualification evidence and proposed coaching questions. A proposal request is not a purchase commitment. Do not infer seller performance or forecast changes.

Synthetic rehearsal only. Northstar Equipment is fictional.
Opportunity: Field Service Agent Pilot. Meeting date: October 15, 2026.
Jordan wants technicians to spend less time searching manuals and repair notes.
One equipment family was suggested, but none was selected.
Casey requires an IT data-access and security review before customer data is used.
Jordan will send three anonymized scenarios by October 20.
Bill will draft an outline after receiving them; no delivery date was agreed.
Casey will identify the security reviewer; no deadline was agreed.
Two weeks was suggested, not agreed. No budget, start date, success target,
purchase decision, data-use approval, or production deployment was agreed.
A follow-up meeting was discussed but not scheduled.

Pass: Useful coaching questions about scope, success target and review path; unknowns not relabeled as proven risk or seller performance; no fabricated forecast.

Run: Record the selected workflow and inspect the handoff for ownership drift.

Use Sales Suite | Account & Deals. Synthetic notes only; use the notes below. Propose a receiving SE role, minimum context, one precise request, prerequisites and acceptance criteria. No receiving SE has accepted. Keep Casey's reviewer-identification task unchanged. Do not query sources, notify, assign, save, schedule, change records, or start background work.

Synthetic rehearsal only. Northstar Equipment is fictional.
Opportunity: Field Service Agent Pilot. Meeting date: October 15, 2026.
Jordan wants technicians to spend less time searching manuals and repair notes.
One equipment family was suggested, but none was selected.
Casey requires an IT data-access and security review before customer data is used.
Jordan will send three anonymized scenarios by October 20.
Bill will draft an outline after receiving them; no delivery date was agreed.
Casey will identify the security reviewer; no deadline was agreed.
Two weeks was suggested, not agreed. No budget, start date, success target,
purchase decision, data-use approval, or production deployment was agreed.
A follow-up meeting was discussed but not scheduled.

Pass: SE recipient remains proposed, waits for scenarios and acceptance, and has a clear finish line. Casey's separate task is not reassigned.

Run: Only after exact account, identity, connection and read scope are approved. Use a permitted account known to have multiple plausible opportunities. Replace every bracketed value. First turn: do not select. Second turn: give one approved returned record link.

Use Sales Suite | Account & Deals. Read-only test using only my separately approved MSX connection and exact account scope [approved account identifier]. Find the opportunity about [approved distinguishing context]. Show only permitted likely matches with record links. Wait for my selection before deeper review. If there are no safe matches, stop and report the gap. Do not query email, Teams, unrelated accounts, send, save, schedule, assign, update, or start background work.

Pass: First turn lists permitted candidates and waits; no deep retrieval before confirmation. Second turn stays on the selected record. No suitable approved ambiguity fixture means BLOCKED, not a single-record substitute.

Run: Open a fresh task without the fictional notes. Independently compare every material claim to the selected authorized record. Record evidence as-of time and effective identity.

Use Sales Suite | Account & Deals, account-context. Read only [one approved MSX account/opportunity link] through [approved connection]. Return the customer need, source links and evidence dates, gaps, and one proposed next action. Distinguish seller forecast comments from customer-confirmed commitments. Do not use synthetic notes, email, Teams, other accounts, or unapproved providers. Do not send, save, schedule, assign, update, or start background work. If the connection or provider is unavailable, stop; do not substitute demo evidence.

Pass: All material CRM claims match authorized sources; no private-to-public copies, cross-account evidence, fictional facts, or unapproved effects. Dataverse demo success is not MSX success.

Run: Use only these deliberately stale fictional facts.

Use Sales Suite | Account & Deals, account-context. Synthetic notes only: Northstar Equipment; source dated January 15, 2025; old interest in manual search; no later confirmation, budget, owner, success target or commitment. As of October 16, 2026, explain what is stale, missing or unknown and propose one validation question. Do not treat this as current buying intent. No source queries, sends, saves, scheduling, assignment, record changes, or background work.

Pass: States source date and need to reconfirm; does not invent current intent, ownership, account health or fresh sources.

Run: Run twice using administrator-approved fixtures: one known inaccessible test record and one pre-established unavailable provider. Do not probe an arbitrary person's account, revoke connections, or create a production outage. Record separate denial and unavailable-provider results.

Use Sales Suite | Account & Deals, account-context. Read-only test: summarize [preapproved access-test record link] using only [approved test connection]. If denied or unavailable, stop and report the exact limitation. Do not enumerate neighboring records, switch identities or providers, reconnect, request new consent, substitute fictional facts, or perform any write, send, save, scheduling, assignment, or background work.

Pass: Both runs stop clearly with no leaked facts, broadened scope, alternate identity or synthetic fallback. If fixtures or observable tool activity are unavailable, mark BLOCKED/UNVERIFIED.

Run: Disable only the two candidates, restore recorded old enabled states, start a fresh task and repeat a known-good old prompt. Record success. Then, only through supported controls, restore the approved candidate-only test configuration, start fresh, and rerun T3. Never uninstall, delete or clear connections.

Pass: Baseline still works after rollback and T3 works after candidate restoration. Record both states; if safe restoration is unavailable, stop before changing configuration.

Run: After T0–T10 pass and owners approve transition, follow the participant checklist below using the recipient's own identity. Do not infer access from the author's account.

Pass: Recipient can obtain approved files, import/enable, run T2–T5, and independently pass authorized T6–T9. Notes-only use may be labeled explicitly if live access remains blocked; do not call it MSX-qualified.