Skip to content

configuration.md

Latest commit

 

History

History
99 lines (74 loc) · 6.55 KB

configuration.md

File metadata and controls

99 lines (74 loc) · 6.55 KB
Raw

Configuration

All thresholds live in app/domain/config/constants.js. Change values there, not inside the algorithms.

CONFIG object

export const CONFIG = {
  // ── Business thresholds ──────────────────────────────────────────────────
  DAILY_GENERATION_QUOTA:    500,   // generations/user/day above this = Quota Exceeded insight
  LOW_ACCEPTANCE_THRESHOLD:  0.20,  // reserved — not used (see note below)
  HIGH_ACCEPTANCE_THRESHOLD: 0.70,  // acceptance rate above this = High Efficiency insight

  // ── User classification ──────────────────────────────────────────────────
  POWER_USER_PERCENTILE:     0.90,  // top 10% by generations = Power User insight
  MIN_GENERATIONS_FOR_RATE:  50,    // min generations before acceptance rate is meaningful

  // ── Trend analysis ───────────────────────────────────────────────────────
  TREND_COMPARISON_DAYS:     7,     // week-over-week window size

  // ── UI limits ────────────────────────────────────────────────────────────
  TABLE_PAGE_SIZE:           100,
  CHART_ANIMATION_DURATION:  750,   // ms
  MAX_TOP_USERS_SHOWN:       15,
  MAX_LANGUAGES_SHOWN:       10,

  // ── Parser performance ───────────────────────────────────────────────────
  CHUNK_SIZE:                10000, // lines processed per setTimeout(0) batch
  MAX_FILE_SIZE_MB:          100,

  // ── Value calculation defaults (user-overridable in the UI) ─────────────
  BLENDED_RATE_PER_HOUR:     90,    // $/hr average developer cost
  MANUAL_LINES_PER_HOUR:     30,    // lines a developer writes manually per hour

  // ── AI Usage report (credit/cost) ───────────────────────────────────────
  NEAR_QUOTA_THRESHOLD:      0.80,  // flag users/orgs at >= 80% of their credit budget
  TOP_SPENDERS_SHOWN:        5,     // users listed in the Top Spenders insight
  MIN_PROJECTION_DAYS:       7,     // below this many observed days, projections are "preliminary"
  CREDIT_USD:                0.01,  // dollar value of one AI credit (GitHub's rate)
  PRICING_DOCS_URL: 'https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing',
  BILLING_DOCS_URL: 'https://docs.github.com/en/copilot/concepts/billing/usage-based-billing-for-organizations-and-enterprises',
};

BILLING_DOCS_URL is linked from the budget panel footnote as the source for the pooling rule: included credits pool at the billing entity level, so overage accrues only once the shared pool is exhausted (1,900 credits/seat on Copilot Business, 3,900 on Copilot Enterprise).

Runtime value overrides

BLENDED_RATE_PER_HOUR and MANUAL_LINES_PER_HOUR can be changed in the dashboard's "Value Calculation Configuration" panel. The domain functions that compute value estimates receive a ValueConfig object as an argument rather than reading CONFIG directly — keeping them testable with any values.

The AI Usage view has a second runtime panel, "License & Budget Configuration", where you enter licenses per tier (seats × per-seat quota) per org. computeAIUsageBudget(records, licenseConfig) takes that config as an argument; when it's enabled, org and enterprise budgets come from the configured seats instead of the active users in the file. This config is held in memory only (never localStorage) and is cleared on reset or a fresh upload, since seat counts are specific to one dataset.

Notes on the AI Usage constants

  • CREDIT_USD (0.01) converts credits to dollars. Model multipliers are already folded into the credit counts in the export, so credits × CREDIT_USD gives gross dollars. PRICING_DOCS_URL is linked from the budget panel as the source.
  • MIN_PROJECTION_DAYS (7) gates the burn-rate projection: with fewer observed days, enterprise.confidence is 'low', the projection is labelled preliminary, and the over-budget alarm is suppressed.
  • NEAR_QUOTA_THRESHOLD (0.80) drives the "at risk" status and the near-quota insights at both the per-user and per-org level.

Why LOW_ACCEPTANCE_THRESHOLD isn't used

The Low Acceptance Rate alert was removed. Around 94% of Copilot activity in enterprise exports is agent mode (Chat · Agent, Edit Mode), which does not track acceptance events. A user with 2,000 agent interactions and 0 acceptances looks identical to a user who never accepts anything — making the metric useless as a signal. The constant is kept in case a future API change makes it meaningful again.

Feature Labels

FEATURE_LABELS in the same file maps raw Copilot API feature keys to human-readable metadata used across the KPI cards and Feature Usage chart.

Current known feature keys (as of late 2025)

API key Label Description
code_completion Code Completion Passive inline ghost-text suggestions, accepted with Tab
agent_edit Edit Mode Copilot Edits panel — agentic multi-file editing with diff review
chat_panel_agent_mode Chat · Agent Autonomous agent via the Chat panel
chat_panel_ask_mode Chat · Ask Q&A mode in the Chat panel — explains but doesn't edit
chat_panel_plan_mode Chat · Plan Plan mode — Copilot drafts a step-by-step plan before acting
chat_panel_custom_mode Chat · Custom Custom-instructions mode with user-defined system prompts
chat_panel_unknown_mode Chat · Other Unrecognised mode flag — likely beta/preview
agent Agent Mode General agentic activity catch-all
chat_inline / inline_chat Inline Chat ⌘I / Ctrl+I inline editor chat
chat Copilot Chat Legacy catch-all key (pre-Ask/Agent split)

When a key appears that isn't in FEATURE_LABELS, humanizeFeature() in common/utils/format.js falls back to a capitalised, underscore-stripped version of the key.

To add a new feature:

// app/domain/config/constants.js
export const FEATURE_LABELS = {
  // ... existing entries ...
  my_new_feature: {
    label: 'My New Feature',
    short: 'New',
    icon: 'cpu',
    desc: 'What this feature does in plain language.'
  }
};