Skip to content

PROMPT.md

Latest commit

 

History

History
93 lines (70 loc) · 4.87 KB

PROMPT.md

File metadata and controls

93 lines (70 loc) · 4.87 KB

GitHub Copilot Enterprise — Analysis Context

Each record = one user's activity on one day. Fields:

Field Meaning
user_login GitHub username
day YYYY-MM-DD
code_generation_activity_count Copilot triggers (suggestions, chat prompts, agent calls — including dismissed)
code_acceptance_activity_count Completions accepted. Only meaningful for code_completion; agent mode has no acceptances.
loc_suggested_to_add_sum Lines offered as ghost text / chat suggestions (user-initiated only, not agent writes)
loc_added_sum Lines actually applied, all modes including agent
loc_deleted_sum Lines removed by Copilot (agent refactors). Positive value, not a cost.
used_agent true if user used any agent feature that day
totals_by_ide Array: {ide, code_generation_activity_count, code_acceptance_activity_count}
totals_by_feature Array: {feature, code_generation_activity_count, code_acceptance_activity_count, loc_added_sum, loc_deleted_sum}
totals_by_language_feature Array: {language, code_generation_activity_count, code_acceptance_activity_count}
totals_by_language_model Array: {model, code_generation_activity_count}

Agent feature keys (used in agent contribution calc): agent, agent_edit, chat_panel_agent_mode, chat_panel_plan_mode, chat_panel_custom_mode

Other feature keys: code_completion, chat, chat_inline, inline_chat, chat_panel_ask_mode

Aggregation

Per user (sum all days): generations, acceptances, linesAdded, linesDeleted, locSuggested, daysActive (distinct days), usedAgent (true if any record has used_agent=true)

Per day (sum all users): generations, linesAdded, linesDeleted, activeUsers (distinct user_login count)

Per week (Monday start): activeUsers = distinct users across all days in that week

Per feature (from totals_by_feature, all records): sum generations, linesAdded, linesDeleted

Per model (from totals_by_language_model): sum generations

Metrics & Formulas

Metric Formula
Acceptance Rate sum(acceptances) / sum(generations) × 100 — completions only
Lines Changed with AI linesAdded + linesDeleted — GitHub's headline metric
Net Lines linesAdded − linesDeleted
Agent Contribution % sum(linesAdded+linesDeleted for agent features only) / (linesAdded+linesDeleted total) × 100
Agent Adoption % users with usedAgent=true / totalUsers × 100
Top Model model with highest total generation count
Avg Daily Users mean of activeUsers per day
Estimated Value (lines / 30) × $90 — assumes 30 lines/hr manual speed, $90/hr blended rate
Value (net) (netLines / 30) × $90
Value (total activity) ((linesAdded + linesDeleted) / 30) × $90

Charts to produce

  1. Daily Lines Changed — grouped bar per day: Lines Added (green) + Lines Deleted (red)
  2. Daily Active Users — line chart, unique users per day
  3. Weekly Active Users — bar chart, unique users per week (Monday start, deduplicated)
  4. Activity Timeline — line chart, total generations per day
  5. Suggested vs Applied — dual line: loc_suggested_to_add_sum vs loc_added_sum per day
  6. Top 15 Users by Generations — horizontal bar
  7. Top 15 Users by Lines Added — horizontal bar
  8. IDE Market Share — pie/bar from totals_by_ide, sized by generations
  9. Language Distribution — bar, top 10 languages by generations from totals_by_language_feature
  10. Feature Usage — bar, all features sorted by generations descending
  11. Model Distribution — pie/bar from totals_by_language_model
  12. Feature Adoption — % of users who used each feature (user used feature if any record has that feature with generations > 0)

Insights to flag

Check Rule
Power Users Top 10% by generations — list up to 5
High Efficiency Users with ≥50 generations AND acceptance rate ≥70%
Spotlight Users Top 5 by linesAdded
Quota Exceeded Any user-day where generations > 500
Week-over-Week (last 7 days total gens − prior 7 days) / prior 7 days × 100%
Zero Acceptance Days Count of user-days where generations > 0 AND acceptances = 0 (normal in agent mode)

Critical caveats

  • Do not divide loc_added_sum by loc_suggested_to_add_sum — result will exceed 100% because agents write code without ghost text
  • Acceptance rate near 0% is normal if most activity is agent mode (agent mode has no accept/reject)
  • Lines deleted = positive value — count it alongside lines added when estimating savings
  • active_time_minutes is often 0 in recent exports — do not use for time analysis
  • Deduplicate before aggregating if using multiple overlapping exports: for same user_login + day, take max of all numeric fields
⚡