dojo.md 0.1.0

package/courses/bug-report-triage/scenarios/level-1/bug-acknowledgment-writing.yaml

@@ -0,0 +1,63 @@
+meta:
+  id: bug-acknowledgment-writing
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Write professional bug acknowledgment responses that set expectations and maintain reporter confidence"
+  tags: [bugs, acknowledgment, communication, response, basics]
+
+state: {}
+
+trigger: |
+  You're handling incoming bug reports at "FormBuilder" (form creation
+  SaaS). Write acknowledgment responses for each of these 4 bug
+  reports. Each requires a different tone and approach.
+
+  Bug 1 (from enterprise customer, angry tone):
+  "Your form builder crashed in the middle of our annual compliance
+  survey and we lost 200 responses. This is the third time this
+  month. We're paying $2,000/month for this? Fix this NOW or we're
+  leaving."
+  Filed: 2 hours ago
+
+  Bug 2 (from free-tier user, helpful tone):
+  "Hey! I found a small bug — when I set a form to require email
+  validation, it accepts 'test@test' without a TLD. Not a big deal
+  but thought you'd want to know! Love the product otherwise."
+  Filed: 1 day ago
+
+  Bug 3 (from internal developer, technical):
+  "The form submission handler is throwing a NullPointerException
+  when the conditional logic field reference points to a deleted
+  field. Stack trace attached. Reproduction: create form with
+  conditional logic, delete the referenced field, submit form."
+  Filed: 4 hours ago
+
+  Bug 4 (from customer, confused):
+  "I don't know if this is a bug or if I'm doing something wrong,
+  but my forms aren't showing up on my website even though I
+  published them. I've tried everything."
+  Filed: 30 minutes ago
+
+  For each, write the acknowledgment response and explain:
+  - What tone did you use and why?
+  - What expectations did you set?
+  - What's the next step you're committing to?
+
+assertions:
+  - type: llm_judge
+    criteria: "Enterprise customer response leads with empathy — acknowledges the data loss and repeated incidents first before discussing next steps, matching the urgency of their frustration"
+    weight: 0.25
+    description: "Empathy-first enterprise response"
+  - type: llm_judge
+    criteria: "Free-tier user response is warm and appreciative — thanks them for the helpful report, doesn't dismiss the bug, and matches their positive energy"
+    weight: 0.25
+    description: "Warm free-tier response"
+  - type: llm_judge
+    criteria: "Internal developer response is technical and efficient — acknowledges the stack trace, confirms the reproduction steps, and discusses the likely code path without unnecessary pleasantries"
+    weight: 0.25
+    description: "Technical internal response"
+  - type: llm_judge
+    criteria: "Confused customer response is reassuring — doesn't assume it's user error or a bug, offers to help troubleshoot, and makes the customer feel comfortable reporting"
+    weight: 0.25
+    description: "Reassuring confused-customer response"

package/courses/bug-report-triage/scenarios/level-1/bug-deduplication.yaml

@@ -0,0 +1,83 @@
+meta:
+  id: bug-deduplication
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Identify duplicate and related bug reports and learn to link them effectively for efficient triage"
+  tags: [bugs, deduplication, organization, triage, basics]
+
+state: {}
+
+trigger: |
+  Your bug tracker at "TaskManager" (task management SaaS) has 10
+  new reports from the last 48 hours. Several look like they might
+  be duplicates or related. Your job is to sort them out.
+
+  Reports:
+
+  #201: "Drag and drop doesn't work on the board view"
+    Reporter: QA, Chrome 120, Windows 11
+
+  #202: "Can't move tasks between columns on Kanban board"
+    Reporter: Customer, Safari 17, macOS
+
+  #203: "Board view tasks are stuck, can't rearrange them"
+    Reporter: Customer, Firefox 121, Ubuntu
+
+  #204: "The search function returns no results for tasks I know exist"
+    Reporter: Customer, Chrome 120, macOS
+
+  #205: "Search is broken — nothing comes up when I search"
+    Reporter: Internal team, Chrome 120, Windows 11
+
+  #206: "I searched for 'Q4 report' and got zero results even though
+    the task exists"
+    Reporter: Customer, mobile app (iOS)
+
+  #207: "Task notifications aren't being sent"
+    Reporter: Customer, N/A (email issue)
+
+  #208: "I'm not getting email alerts when tasks are assigned to me"
+    Reporter: Customer, N/A (email issue)
+
+  #209: "The API returns 500 when creating tasks with special
+    characters in the title (e.g., &, <, >)"
+    Reporter: Developer, API v2
+
+  #210: "Drag and drop is broken but only in the timeline view,
+    works fine on the board"
+    Reporter: QA, Chrome 120, Windows 11
+
+  Tasks:
+
+  1. Group these reports into clusters of duplicates and related
+     issues. Which reports describe the same bug?
+
+  2. For each cluster, identify the "canonical" report (the one
+     with the best quality/most information) that should be the
+     primary tracking issue.
+
+  3. #210 mentions drag-and-drop but in a different view than
+     #201-#203. Is this a duplicate or a separate issue? Explain
+     your reasoning.
+
+  4. Write the comment you'd add to a duplicate issue before closing
+     it, linking to the canonical issue.
+
+assertions:
+  - type: llm_judge
+    criteria: "Board drag-and-drop cluster is identified — #201, #202, #203 are grouped as the same bug across different browsers, while #210 is a separate issue in a different view"
+    weight: 0.25
+    description: "Board drag-drop cluster identified"
+  - type: llm_judge
+    criteria: "Search cluster is identified — #204, #205, #206 are grouped as the same search bug, with #206 noting it also affects mobile"
+    weight: 0.25
+    description: "Search cluster identified"
+  - type: llm_judge
+    criteria: "#210 is correctly separated from #201-203 — recognizes that timeline view drag-and-drop is likely a different code path than board view, making it a related but separate bug"
+    weight: 0.25
+    description: "Timeline bug correctly separated"
+  - type: llm_judge
+    criteria: "Duplicate close comment is professional — explains why the issue is being closed, links to the canonical issue, and assures the reporter their report is valued"
+    weight: 0.25
+    description: "Professional duplicate comment"

package/courses/bug-report-triage/scenarios/level-1/bug-prioritization-basics.yaml

@@ -0,0 +1,64 @@
+meta:
+  id: bug-prioritization-basics
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Learn the difference between severity and priority, and how to assign priority levels based on business impact, user reach, and available workarounds"
+  tags: [bugs, prioritization, severity, priority, fundamentals]
+
+state: {}
+
+trigger: |
+  You're a junior QA engineer at "TaskFlow" (project management SaaS,
+  5,000 users). Your team lead explains: "Severity describes how broken
+  something is technically. Priority describes how urgently we need to
+  fix it. They're different — a cosmetic bug on the login page might be
+  low severity but high priority because every user sees it."
+
+  Here are 6 bugs that just came in. For each one, assign both a
+  severity (Critical / Major / Minor / Trivial) and a priority
+  (P0-Immediate / P1-High / P2-Medium / P3-Low), then explain your
+  reasoning:
+
+  Bug A: The CEO's profile photo appears pixelated on the "About Us"
+  public marketing page. No functionality is affected.
+
+  Bug B: When users export a project to PDF, the last page is always
+  blank. The actual content exports correctly.
+
+  Bug C: The password reset flow sends the email but the reset link
+  returns a 404 error. Users cannot reset their passwords.
+
+  Bug D: On Safari mobile, the dropdown menu for project status overlaps
+  with the save button, making it hard (but not impossible) to save.
+
+  Bug E: When exactly 100 tasks exist in a project, the task counter
+  displays "99+" instead of "100". Over 100 shows correctly.
+
+  Bug F: The search function returns no results when a query contains
+  an apostrophe (e.g., searching for "John's tasks" returns nothing,
+  but "Johns tasks" works fine).
+
+  Tasks:
+  1. Classify each bug with severity AND priority, with reasoning
+  2. Explain a situation where severity and priority would diverge
+     (high severity but low priority, or low severity but high priority)
+  3. Rank these 6 bugs in the order you'd fix them
+
+assertions:
+  - type: llm_judge
+    criteria: "Severity and priority are treated as distinct dimensions — Bug C should be high severity AND high priority, while Bug A should be low severity but potentially medium priority due to visibility"
+    weight: 0.25
+    description: "Distinct severity vs priority"
+  - type: llm_judge
+    criteria: "Classifications are reasonable — Bug C is Critical/P0, Bug F is Major/P1-P2 (functionality broken), Bug D is Minor/P2, Bug E is Trivial/P3, Bug B is Minor/P3, Bug A is Trivial but elevated priority"
+    weight: 0.25
+    description: "Reasonable classifications"
+  - type: llm_judge
+    criteria: "Divergence example is clear — explains a scenario where a severe bug has low priority (e.g., crash in admin-only feature used by 2 people) or a trivial bug has high priority (e.g., typo on pricing page)"
+    weight: 0.25
+    description: "Clear divergence example"
+  - type: llm_judge
+    criteria: "Fix ordering prioritizes user impact — Bug C (password reset broken) comes first, followed by bugs affecting core functionality, with cosmetic issues last"
+    weight: 0.25
+    description: "Impact-based fix ordering"

package/courses/bug-report-triage/scenarios/level-1/bug-report-quality.yaml

@@ -0,0 +1,64 @@
+meta:
+  id: bug-report-quality
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Evaluate and improve bug report quality — identify what information is missing and rewrite incomplete bug reports"
+  tags: [bugs, report-quality, writing, communication, basics]
+
+state: {}
+
+trigger: |
+  Your engineering team at "ChatBase" (messaging platform) has been
+  complaining that bug reports from the QA team and customers are
+  too vague to act on. Your lead asks you to evaluate 3 bug reports
+  and improve them.
+
+  Bug Report A (from QA):
+  Title: "Chat doesn't work"
+  Description: "The chat feature is broken. Please fix."
+
+  Bug Report B (from customer):
+  Title: "Can't send messages"
+  Description: "I've been trying to send messages all morning but
+  it's not working. This is really frustrating. I need this fixed
+  ASAP because we have a client meeting."
+
+  Bug Report C (from a developer):
+  Title: "WebSocket connection drops after 30min idle on Firefox 115"
+  Description: "Steps: 1) Open chat room, 2) Leave idle for 30 min,
+  3) Attempt to send message. Expected: message sends. Actual:
+  'Connection lost' error. Tested on Firefox 115.0.2, macOS 14.2.
+  Does not reproduce on Chrome 120. Server logs show no heartbeat
+  response. Likely related to Firefox's aggressive socket timeout."
+
+  Tasks:
+
+  1. Rate each bug report's quality (1-5 scale) and explain why.
+
+  2. What are the essential components of a good bug report?
+     List them and explain why each matters.
+
+  3. Rewrite Bug Reports A and B to meet quality standards. Add
+     reasonable assumptions for missing details.
+
+  4. What questions would you ask the reporter of Bug Report A
+     and Bug Report B to get the information you need?
+
+assertions:
+  - type: llm_judge
+    criteria: "Bug Report C is rated highest — recognizes it as a well-written report with steps to reproduce, environment details, expected/actual behavior, and initial diagnosis"
+    weight: 0.25
+    description: "Quality ratings are correct"
+  - type: llm_judge
+    criteria: "Essential components include at minimum: title, steps to reproduce, expected behavior, actual behavior, environment/browser/OS, and severity"
+    weight: 0.25
+    description: "Essential components identified"
+  - type: llm_judge
+    criteria: "Rewritten reports include steps to reproduce — both Report A and B rewrites add specific reproduction steps, not just more descriptive summaries"
+    weight: 0.25
+    description: "Rewrites include reproduction steps"
+  - type: llm_judge
+    criteria: "Follow-up questions are specific — asks about browser, OS, steps taken, error messages, screenshots, and whether the issue is reproducible, not just 'can you give more detail?'"
+    weight: 0.25
+    description: "Specific follow-up questions"

package/courses/bug-report-triage/scenarios/level-1/bug-reproduction.yaml

@@ -0,0 +1,64 @@
+meta:
+  id: bug-reproduction
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Learn basic bug reproduction strategies — how to reproduce reported issues and what to do when you can't reproduce them"
+  tags: [bugs, reproduction, debugging, troubleshooting, basics]
+
+state: {}
+
+trigger: |
+  You're a support engineer at "ScheduleApp" (scheduling SaaS).
+  Three bugs were reported this week and you need to attempt
+  reproduction before passing them to engineering.
+
+  Bug A: "Calendar events disappear after refreshing the page"
+  Reporter: Sarah, Marketing team, Chrome on Windows 11
+  Additional info: "It happens sometimes. Not every time."
+
+  Bug B: "Can't create events on February 29, 2028"
+  Reporter: DevOps team during testing
+  Additional info: "Tried to schedule a recurring event that would
+  fall on Feb 29. Got error: 'Invalid date.'"
+
+  Bug C: "The app is really slow"
+  Reporter: Jake, Sales team, Safari on macOS
+  Additional info: "Pages take forever to load. Started about a
+  week ago."
+
+  Tasks:
+
+  1. For each bug, write a reproduction plan:
+     - What specific steps would you follow?
+     - What environment/conditions do you need to set up?
+     - What are you looking for to confirm the bug?
+
+  2. Bug A says "happens sometimes." What techniques can you use
+     to reproduce intermittent bugs? List at least 4 approaches.
+
+  3. You tried Bug A's reproduction steps 10 times and it never
+     happened. Write the response back to Sarah asking for more
+     information. What specific questions do you ask?
+
+  4. Bug C is vague ("really slow"). How do you turn this into
+     a measurable, actionable bug report? What data would you
+     collect?
+
+assertions:
+  - type: llm_judge
+    criteria: "Reproduction plans are step-by-step — each bug gets specific steps to follow with environment setup, not just 'try to reproduce the issue'"
+    weight: 0.25
+    description: "Step-by-step reproduction plans"
+  - type: llm_judge
+    criteria: "Intermittent bug techniques are listed — mentions approaches like varying timing, different data sets, network conditions, concurrent users, or cache states"
+    weight: 0.25
+    description: "Intermittent bug techniques"
+  - type: llm_judge
+    criteria: "Follow-up questions to Sarah are specific — asks about what she was doing before refresh, how many events, browser extensions, network, and whether specific event types disappear"
+    weight: 0.25
+    description: "Specific follow-up questions"
+  - type: llm_judge
+    criteria: "Performance bug is quantified — suggests collecting page load times, network tab data, browser dev tools metrics, or user count on the account to make 'slow' measurable"
+    weight: 0.25
+    description: "Quantified performance bug"

package/courses/bug-report-triage/scenarios/level-1/bug-severity-basics.yaml

@@ -0,0 +1,72 @@
+meta:
+  id: bug-severity-basics
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Understand the difference between bug severity and priority, and learn to classify bugs correctly using standard frameworks"
+  tags: [bugs, severity, priority, classification, basics]
+
+state: {}
+
+trigger: |
+  You're a new QA engineer at "AppFlow" (project management SaaS).
+  Your lead asks you to classify the severity and priority for 6
+  incoming bug reports before the weekly triage meeting.
+
+  Bug reports:
+
+  1. "App crashes when clicking Save on projects with more than
+     100 tasks. We've had 3 enterprise customers report this today."
+
+  2. "The 'Export to PDF' button text is misspelled as 'Exprot to PDF'
+     on the dashboard page."
+
+  3. "Users can access admin settings by modifying the URL parameter.
+     Any logged-in user can change billing and permissions."
+
+  4. "The loading spinner on the task list page is slightly off-center
+     by about 2 pixels on Safari."
+
+  5. "Email notifications for task assignments are being sent 4-6
+     hours late instead of immediately."
+
+  6. "The date picker shows incorrect dates when the user's timezone
+     is set to anything other than UTC."
+
+  Tasks:
+
+  1. Define the difference between severity and priority. Why are
+     they different, and give an example of a high-severity,
+     low-priority bug.
+
+  2. Classify each bug using this severity scale:
+     - Critical: System crash, data loss, security vulnerability
+     - Major: Core feature broken, significant user impact
+     - Minor: Feature works but with issues, workaround available
+     - Trivial: Cosmetic, no functional impact
+
+  3. Assign priority (P0-P3) to each bug:
+     - P0: Fix immediately, all hands on deck
+     - P1: Fix in current sprint
+     - P2: Fix in next sprint
+     - P3: Fix when time allows
+
+  4. For each classification, explain your reasoning in 1-2 sentences.
+
+assertions:
+  - type: llm_judge
+    criteria: "Severity vs priority distinction is clear — explains that severity is about technical impact while priority is about business urgency, with a valid example of their divergence"
+    weight: 0.25
+    description: "Clear severity-priority distinction"
+  - type: llm_judge
+    criteria: "Security bug (#3) is classified as Critical severity and P0 priority — unauthorized admin access is a security vulnerability that needs immediate attention"
+    weight: 0.25
+    description: "Security bug correctly classified"
+  - type: llm_judge
+    criteria: "Crash bug (#1) is high severity and high priority — enterprise customers affected with data loss risk puts this at Critical/P0 or P1"
+    weight: 0.25
+    description: "Crash bug correctly classified"
+  - type: llm_judge
+    criteria: "Cosmetic bugs (#2, #4) are low severity — typo and pixel alignment are Minor/Trivial with P2-P3 priority, not treated as urgent"
+    weight: 0.25
+    description: "Cosmetic bugs correctly classified"

package/courses/bug-report-triage/scenarios/level-1/customer-bug-communication.yaml

@@ -0,0 +1,67 @@
+meta:
+  id: customer-bug-communication
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Write clear, professional responses to customers who report bugs — from acknowledgment through resolution"
+  tags: [bugs, customer, communication, response, basics]
+
+state: {}
+
+trigger: |
+  You handle customer-reported bugs at "FileSync" (file storage
+  SaaS). Write responses for each stage of the bug lifecycle.
+
+  The bug: A customer reports "Files uploaded yesterday are showing
+  as 0 bytes. I can see them in my file list but they're empty.
+  This happened to about 20 files I uploaded between 3-5 PM EST."
+
+  Customer: Maria, Business plan ($49/mo), 2-year customer
+
+  Write 4 responses for each stage:
+
+  1. Initial acknowledgment (within 1 hour):
+     - Acknowledge the issue
+     - Show you understand the impact
+     - Set expectations for next update
+
+  2. Investigation update (4 hours later):
+     Engineering found that a storage migration caused the issue.
+     Files aren't lost — they're in a temporary cache. Recovery
+     is possible but will take 6-8 hours.
+     - Communicate what you found
+     - Set expectations for resolution
+     - Reassure about data safety
+
+  3. Resolution notification (next morning):
+     Files have been fully restored.
+     - Confirm resolution
+     - Explain what happened (non-technical)
+     - Explain what you're doing to prevent recurrence
+
+  4. Follow-up (3 days later):
+     Check in to make sure everything is working.
+     - Verify the fix is holding
+     - Rebuild confidence
+     - Thank them for reporting
+
+  For each response, note the key communication principle you're
+  applying.
+
+assertions:
+  - type: llm_judge
+    criteria: "Acknowledgment includes a timeline — tells Maria when she'll hear back next, not just 'we're looking into it' with no timeframe"
+    weight: 0.25
+    description: "Timeline in acknowledgment"
+  - type: llm_judge
+    criteria: "Investigation update leads with good news — 'files aren't lost' appears early in the message, not buried after technical explanation"
+    weight: 0.25
+    description: "Good news first in update"
+  - type: llm_judge
+    criteria: "Resolution explains non-technically — describes the storage migration issue in terms Maria can understand without being condescending"
+    weight: 0.25
+    description: "Non-technical explanation"
+  - type: llm_judge
+    criteria: "Follow-up demonstrates care — checking in 3 days later shows proactive customer service, not just closing the ticket"
+    weight: 0.25
+    description: "Proactive follow-up"

package/courses/bug-report-triage/scenarios/level-1/first-triage-shift.yaml

@@ -0,0 +1,66 @@
+meta:
+  id: first-triage-shift
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Handle your first day of bug triage — process a queue of reports, classify, respond, and escalate appropriately"
+  tags: [bugs, simulation, first-day, triage, basics]
+
+state: {}
+
+trigger: |
+  It's your first day handling bug triage at "DataDash" (analytics
+  dashboard SaaS). You have 2 hours before the daily standup and
+  10 bugs in the queue. Process them all.
+
+  Your toolkit:
+  - Bug tracker (Jira)
+  - Severity scale: Critical, Major, Minor, Trivial
+  - Priority: P0 (immediate), P1 (this sprint), P2 (next sprint),
+    P3 (backlog)
+  - You can: classify, acknowledge, request info, mark as duplicate,
+    escalate, or assign
+
+  The queue:
+
+  1. "Dashboard blank page after login" — Critical alert from
+     monitoring, affecting 15% of users, started 20 minutes ago
+  2. "Chart colors don't match brand guide" — From design team
+  3. "Data export times out for date ranges > 1 year" — Customer
+  4. "SQL injection possible via custom filter input" — From
+     security audit
+  5. "Tooltip shows wrong date format (MM/DD vs DD/MM)" — Customer
+     in UK
+  6. "Same as #1 — blank dashboard" — From customer support team
+  7. "Can you add a 'compare periods' feature?" — Customer request
+  8. "Charts render incorrectly on Internet Explorer 11" — Customer
+  9. "Refresh rate on real-time dashboard is 60s, should be 10s" —
+     Enterprise customer
+  10. "Memory leak in chart rendering — page crashes after 4 hours
+      open" — From internal monitoring
+
+  Tasks:
+  1. Process all 10 items: classify severity/priority, decide action,
+     and write a one-sentence plan for each.
+  2. Write full acknowledgment responses for your top 3 priorities.
+  3. Which items require escalation, and to whom?
+  4. After processing, write the standup update summarizing today's
+     triage queue.
+
+assertions:
+  - type: llm_judge
+    criteria: "SQL injection (#4) and blank page (#1) are top priorities — security vulnerability and active incident affecting 15% of users are both P0 requiring immediate escalation"
+    weight: 0.25
+    description: "Critical items prioritized"
+  - type: llm_judge
+    criteria: "Feature request (#7) and IE11 bug (#8) are handled appropriately — feature request redirected out of bug queue, IE11 assessed for support status (likely won't fix if EOL)"
+    weight: 0.25
+    description: "Non-bugs handled correctly"
+  - type: llm_judge
+    criteria: "Duplicate (#6) is linked to #1 — identified as the same blank dashboard issue and linked rather than treated as separate"
+    weight: 0.25
+    description: "Duplicate identified"
+  - type: llm_judge
+    criteria: "Standup update is concise and actionable — summarizes the queue status, highlights the active incident and security issue, and communicates what needs attention"
+    weight: 0.25
+    description: "Concise standup update"

package/courses/bug-report-triage/scenarios/level-1/triage-meeting-basics.yaml

@@ -0,0 +1,70 @@
+meta:
+  id: triage-meeting-basics
+  level: 1
+  course: bug-report-triage
+  type: output
+  description: "Understand how bug triage meetings work and make basic triage decisions on a queue of incoming bugs"
+  tags: [bugs, triage, meeting, process, basics]
+
+state: {}
+
+trigger: |
+  You're attending your first bug triage meeting at "InvoicePro"
+  (invoicing SaaS). The team reviews new bugs and decides what to
+  do with each one. Your lead explains the possible outcomes:
+
+  - Fix now: Assign to current sprint
+  - Fix next: Add to next sprint backlog
+  - Fix later: Add to backlog, not prioritized
+  - Need more info: Send back for clarification
+  - Won't fix: Close with explanation
+  - Duplicate: Link to existing issue and close
+
+  Here are 8 new bugs to triage:
+
+  1. "Invoice PDF generation fails for invoices over $1M"
+     Severity: Major | Reported by: Enterprise customer
+  2. "Typo in the footer of invoice emails: 'Thankyou' instead
+     of 'Thank you'"
+     Severity: Trivial | Reported by: Internal QA
+  3. "Dashboard shows revenue as negative when refunds exceed
+     sales in a given month"
+     Severity: Major | Reported by: Customer support
+  4. "Can't reproduce: 'App freezes when editing line items'"
+     Severity: Unknown | Reported by: Customer (no details)
+  5. "Tax calculation shows 0% for Canadian provinces"
+     Severity: Critical | Reported by: Accounting team
+  6. "Same as #5 — Canadian tax issue"
+     Severity: Critical | Reported by: Customer
+  7. "Would be nice to have dark mode on invoices"
+     Severity: N/A (feature request) | Reported by: Customer
+  8. "Password reset emails take 15+ minutes to arrive"
+     Severity: Minor | Reported by: Multiple customers
+
+  Tasks:
+
+  1. Assign a triage outcome to each of the 8 items. Explain
+     your reasoning.
+  2. What information do you need for item #4 before making a
+     triage decision?
+  3. How do you handle item #7 (feature request in the bug queue)?
+  4. Write a brief summary of triage decisions that could be shared
+     with the team after the meeting.
+
+assertions:
+  - type: llm_judge
+    criteria: "Tax calculation bug (#5) is triaged as 'Fix now' — incorrect tax calculations for an entire country is a critical business issue affecting compliance"
+    weight: 0.25
+    description: "Critical tax bug prioritized"
+  - type: llm_judge
+    criteria: "Duplicate (#6) is correctly identified — linked to #5 and closed as duplicate, not treated as a separate issue to fix"
+    weight: 0.25
+    description: "Duplicate correctly identified"
+  - type: llm_judge
+    criteria: "Feature request (#7) is redirected — moved out of the bug queue to a feature request tracker or product backlog, not triaged as a bug"
+    weight: 0.25
+    description: "Feature request redirected"
+  - type: llm_judge
+    criteria: "Unreproducible bug (#4) gets 'Need more info' — sent back with specific questions rather than closed or assigned without sufficient information"
+    weight: 0.25
+    description: "Unreproducible bug handled"