# All 16 repos in scope — always check all of them
Hemut-Prod
Voice-AI-Element
zones
RFP-Module
Web_Scrapers
hemut-sonar
tracking_backend
QuoteFlow_Backend
hemutchat
Hemut-News
SharedInfraLayer
MobileAppUpdated
Helm
Hemut-MCP
Hemut-Library
Hemut-Website

# These are bots/service accounts — never count as engineer activity
hemut-swe          # Andrew FE Zhang's service account (zhang@hemut.com)
dependabot[bot]
use-tusk[bot]

# 1. Pull all merged PRs in window (run per repo)
gh pr list \
  --repo Hemut2025/[REPO] \
  --state merged \
  --author [GITHUB_HANDLE] \
  --json number,title,additions,deletions,mergedAt,body,labels,createdAt,headRefName \
  --limit 300 \
  | jq '[.[] | select(.mergedAt >= "[START_DATE]" and .mergedAt <= "[END_DATE]")]'

# 2. Per-PR detail: files changed
gh api repos/Hemut2025/[REPO]/pulls/[PR_NUMBER]/files

# 3. Per-PR detail: reviews received
gh api repos/Hemut2025/[REPO]/pulls/[PR_NUMBER]/reviews

# 4. Per-PR detail: commits
gh api repos/Hemut2025/[REPO]/pulls/[PR_NUMBER]/commits

# 5. CI runs for this engineer
gh api "repos/Hemut2025/[REPO]/actions/runs?actor=[GITHUB_HANDLE]&per_page=100"

# 6. Direct commits (not via PR)
gh api "repos/Hemut2025/[REPO]/commits?author=[GITHUB_HANDLE]&since=[START_DATE]T00:00:00Z&until=[END_DATE]T23:59:59Z&per_page=100"

# 7. Pre-window activity check (CRITICAL for tenure correction)
gh api "repos/Hemut2025/[REPO]/commits?author=[GITHUB_HANDLE]&until=[START_DATE]T00:00:00Z&per_page=1" \
  --jq '.[0].commit.author.date'

# 8. Reviews given by this engineer (not received)
gh api "repos/Hemut2025/[REPO]/pulls/[PR_NUMBER]/reviews" \
  | jq '[.[] | select(.user.login == "[GITHUB_HANDLE]")]'

# 9. Check rate limit before and after heavy batches
gh api rate_limit --jq '.resources.core | {remaining, reset}'

{
  "engineer": {
    "name": "[ENGINEER_NAME]",
    "github": "[GITHUB_HANDLE]",
    "role": "[ROLE]",
    "monthly_salary": [MONTHLY_SALARY],
    "known_data": {}
  },
  "repos_checked": [],          // list of all 16 repos, even those with 0 PRs
  "errors": {},                  // any repos that 404'd or errored
  "prs": [],                      // full PR objects with files/reviews/commits embedded
  "ci_runs": [],                 // CI run objects per repo
  "reviews_given": [],           // PRs this engineer reviewed (authored by others)
  "direct_commits": [],          // commits not attached to a PR
  "pre_window_first_commit": null, // ISO date string if exists; null = new hire
  "pull_timestamp": "",           // ISO datetime when Agent 1 ran
  "window": {
    "start": "[START_DATE]",
    "end": "[END_DATE]"
  }
}

# Agent 1 must end with this exact line:
AGENT 1 COMPLETE — [NUMBER] PRs across [REPOS] repos. Saved to data/[XX]_[Name]-raw-data.json

# INPUT
data/[XX]_[Name]-raw-data.json   # or .md for older format files

# OUTPUT
[XX]_[Name]-Deep-Dive.html       # saved to Outputs/2026-03-14_Engineer-Deep-Dives/

# Chart.js CDN (always use this exact URL)
https://cdn.jsdelivr.net/npm/chart.js

# The 12 required charts:
1.  Weekly velocity line (13 weeks + team avg + frontier reference lines)
2.  Additions vs deletions stacked area chart
3.  Work type donut (Feature / Bug / Refactor / Infra / Chore / Test / Docs)
4.  Time-of-day heatmap (7 days × 24 hours, derived from commit timestamps)
5.  90-day commit calendar (13×7 HTML grid — NOT Chart.js, pure HTML/CSS)
6.  AI spend vs PR output dual-axis bar+line
7.  PR size histogram (bins: XS <50, S 50-200, M 200-500, L 500-2K, XL >2K lines)
8.  Depth/complexity trend over 13 weeks
9.  Cost efficiency trend ($/PR vs $9.15 frontier threshold)
10. Codebase coverage radar (one axis per repo)
11. PR lifecycle funnel (opened → reviewed → merged, with day counts)
12. Review quality scatter (reviews given vs review depth)

/* CSS variables — use these in every report */
--bg: #0f1724;
--surface: #1B2A4A;
--gold: #C9A84C;
--green: #2ecc71;
--red: #e74c3c;

/* Fonts (Google Fonts CDN) */
IBM Plex Sans   — body, UI
IBM Plex Mono   — code, metrics, chart labels

/* Google Fonts URL */
https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500;600&family=IBM+Plex+Sans:wght@300;400;500;600;700&display=swap

data/linear-data.md           — teams, cycles, issue counts, ticket linkage per engineer
data/sentry-data.md           — open issues per project, error counts, top issues
data/meetings-calendar-data.md — relevant Otter transcripts with engineer mentions
data/context-enrichment.md    — combined synthesis of all three sources above

# MCP tool: mcp__linear__list_teams
# → Get all team IDs and names

# MCP tool: mcp__linear__list_cycles (per team)
# → For each engineering team: SWAG, ASH, CRODIE, GUDA, ADT
# → Identify current active cycle

# MCP tool: mcp__linear__list_issues (per team, limit 50)
# → For each team: pull 50 most recent issues
# → Current cycle completion rate = completedIssueCount / totalIssueCount

# Ticket linkage per engineer:
# Count issues where gitBranchName CONTAINS engineer's github_handle
# This is the primary ticket-linkage metric

# Save as: data/linear-data.md

# MCP tool: mcp__sentry__search_issues

Organization: hemut

# Projects to check:
command-backend
command-frontend
hemut-quoteflow-backend

# Query: unresolved errors in the last 90 days
Query: "is:unresolved"
Date range: last 90 days

# Save: count of unresolved issues, top 10 by event count
# Save as: data/sentry-data.md

# MCP tool: mcp__slack__slack_search_public_and_private

# Primary channel query
Query: "in:#hemutdevs after:[DATE_30_DAYS_AGO]"

# What to look for and save:
# - Sprint retrospective updates or announcements
# - Engineer shoutouts or performance callouts
# - Blockers or incident reports
# - Any anomalies (sudden silence, repeated issues)
# - Messages mentioning specific engineers by @handle or name

# Save relevant messages with: sender, timestamp, channel, full text
# Save as: data/context-enrichment.md (Slack section)

# Step 1: List recent transcripts
# MCP tool: mcp__otter__otter_list_transcripts
# → Get last 20 transcripts

# Step 2: For each transcript with engineer names in summary
# MCP tool: mcp__otter__otter_get_transcript
# → Pull full transcript text

# Key patterns to extract and save:
# - Offer discussions (comp, equity, timeline mentions)
# - Performance mentions (positive or corrective)
# - Technical decisions or architectural discussions
# - Sprint planning decisions that affect workload

# Save as: data/meetings-calendar-data.md

# Raw data files (all 26)
data/01_sumedh-kane-raw.json
data/02_Ashish-Lathkar-raw-data.json
data/03_... through data/26_...

# Enrichment context
data/context-enrichment.md
data/linear-data.md
data/sentry-data.md
data/meetings-calendar-data.md

# Benchmark reference
data/elite-benchmarks.md         ← frontier benchmarks (update manually quarterly)

# Engineer configuration
/Users/lokicheema/Desktop/Hemut\ Files/Dashboard/engineers.json

Outputs/2026-03-14_Engineer-Deep-Dives/00_MASTER-Engineering-Synthesis.html

{
  "id": "sumedh",                         // short key, lowercase, no spaces
  "name": "Sumedh Kane",
  "github_handle": "sumedhkane03",       // primary handle (verify by checking commits)
  "github_handle_alt": null,              // secondary handle if exists
  "cursor_email": "sumedh@hemut.com",    // email for Cursor/Claude billing attribution
  "claude_key_prefix": "sk-ant-...",     // prefix in Anthropic billing to match spend
  "role": "Founding Engineer",
  "monthly_salary": 5000,                // integer USD, 0 if unpaid/terminated
  "tier": 1,                              // 1-5 performance tier (1 = top)
  "status": "active",                     // active | terminated | unresolved | external
  "notes": ""                              // alt emails, billing quirks, anything unusual
}

# Federico Lora
Commits from: lora.fed.03@gmail.com (personal) + federico@hemut.com
Both emails → handle: lorafed
# Must search BOTH emails in billing attribution

# Pranav Guda — TWO ACCOUNTS
Primary: PranavGuda (pranav@hemut.com)
Dev/merge: hemoot (developer@hemut.com)
# Pull data for BOTH handles and merge in Agent 1

# Andrew FE Zhang — service account confusion
hemut-swe = zhang@hemut.com = Andrew FE Zhang's service account
Personal handle: zhemut
# Add hemut-swe to skip_logins. Use zhemut for pull.

# TWO ANDREW ZHANGS — completely different people
zhemut    = Andrew FE Zhang (founding eng, $5K/mo, zhemut GitHub)
zhangandrew2 = A2Z2 (summer intern, $500/mo, separate person)
# Never merge their data. Always confirm by email.

# Darius Mahjoob — bundled account history
Pre-Oct 2025 activity: under shared Hemut account, unattributable
Individual GitHub history starts: Oct 2025 under DMahjoob
# Use Linear join date as true start date proxy for pre-Oct work

# Bao Tran
GitHub: BaoT1301
Email: bao@hemut.com
# Separate from Pranav — different person entirely

# SAFE: Run up to 6 Agent 1s simultaneously
# UNSAFE: Running 7+ — will hit rate limits mid-pull

# If rate limited:
# → Agent will receive HTTP 403 with "rate limit exceeded"
# → Do NOT cancel. Wait for retry. Limit resets every hour on the hour.
# → Check reset time: gh api rate_limit --jq '.resources.core.reset'
# → Convert unix timestamp to local time to know when to retry

# Stagger if needed: start each Agent 1 with a 30-second gap
# This distributes the burst load

# Check remaining limit before starting a batch:
gh api rate_limit --jq '.resources.core | {remaining, reset}'

Outputs/2026-03-14_AI-Spend-Analytics/Sprint-Dashboard.html
Outputs/2026-03-14_AI-Spend-Analytics/Engineer-90Day-Breakdown.html

# PRs merged this sprint (run per repo, replace SPRINT_START with ISO date)
gh pr list \
  --repo Hemut2025/Hemut-Prod \
  --state merged \
  --json author,title,additions,deletions,mergedAt,labels \
  --limit 500 \
  | jq '[.[] | select(.mergedAt >= "SPRINT_START")]'

# Aggregate by author (total PRs per engineer this sprint)
  | jq 'group_by(.author.login) | map({login: .[0].author.login, count: length})'

# CI pass rate this sprint (per repo)
gh api "repos/Hemut2025/Hemut-Prod/actions/runs?per_page=100" \
  --jq '[.workflow_runs[] | {actor: .actor.login, conclusion: .conclusion, created_at: .created_at}]'

# CI pass rate formula:
# success_runs / (success_runs + failure_runs) — exclude skipped/cancelled

# Anthropic Console:
console.anthropic.com → Billing → Usage → Export CSV
# Match rows by cursor_email or claude_key_prefix from engineers.json

# Cursor seats:
cursor.sh → Team Settings → Billing → Export usage
# Match by cursor_email from engineers.json

// In Sprint-Dashboard.html, find the SPRINT_DATA constant and update:
const SPRINT_DATA = {
  engineers: [
    {
      id: "sumedh",
      sprint: {
        prs: 12,                // ← PRs merged this sprint
        aiSpend: 48.50,          // ← AI spend this sprint in USD
        ciPassRate: 0.94          // ← 0.0–1.0
      },
      ninety: {
        prs: 87,                // ← rolling 90-day PR total
        aiSpend: 312.00,         // ← rolling 90-day AI spend
        dpr: 6.7                 // ← dollars per PR (aiSpend / prs)
      },
      sprintNote: ""              // ← new signals or flags this sprint
    },
    // ... repeat for all engineers ...
  ]
}

# Paste this to a Claude Haiku agent (one per batch)
Use Haiku model. Read:
1. public/engineers/sumedh.html
2. public/engineers/ashish.html
3. public/engineers/a2z2.html
4. data/github_scored.json

For each engineer, verify:
- Total PRs in HTML matches github_scored.json total_prs
- AI spend is formatted as $X,XXX.XX (no raw floats)
- Linear done/total shows non-zero values
- Weighted rank matches github_scored.json weighted_rank
- Model mix percentages are not all 0%

Output: one line per engineer in format:
[NAME]: PASS or FAIL — [brief note if FAIL]

# Fix rendering bug → fast redeploy (no re-scoring needed)
./run.sh --skip-github --deploy

# Fix data bug → check source files in data/ first, then redeploy
# If score_prs.py was changed: --rebuild-cache required
./run.sh --skip-github --rebuild-cache --deploy

public/
├── index.html                           ← dashboard home / nav hub
├── engineers/
│   ├── sumedh.html                      ← per-engineer sprint page (25 files)
│   ├── ashish.html
│   └── ... (one per engineer) ...
├── sprint.html                          ← team-wide sprint view
├── team.html                            ← org tier map, spend treemap
├── spend.html                           ← AI spend analytics
├── sentry.html                          ← production error health
├── linear.html                          ← issue completion rates
├── code-quality.html                    ← PR quality metrics
├── WEEKLY-RUNBOOK.html                  ← this document (also in Dashboard root)
└── archive/                             ← superseded legacy files

NOTE: public/**/*.html is gitignored — regenerated on every pipeline run.
Save committed docs (runbook, playbooks) to Dashboard/ root, not public/.

Dashboard/
├── 00_MASTER-Engineering-Synthesis.html   ← quarterly org narrative
├── 01_Sumedh-Kane-Deep-Dive.html          ← quarterly deep-dive (14 files)
├── 02_Ashish-Lathkar-Deep-Dive.html
└── ... (one per engineer) ...

# Deploy is handled by run.sh --deploy flag
# Manual deploy if needed:
cd "/Users/lokicheema/Desktop/Hemut Files/Dashboard"
vercel --prod --yes

# Live URL:
https://public-delta-tawny.vercel.app/

# Runbook URL:
https://public-delta-tawny.vercel.app/WEEKLY-RUNBOOK.html

# After updating WEEKLY-RUNBOOK.html — copy to public/ and deploy:
cp WEEKLY-RUNBOOK.html public/WEEKLY-RUNBOOK.html
vercel --prod --yes