tradeit.gg

Week 15 · 2026

Engineering Operations

March 30 – April 5

Highlight of the Week

6 operations shipped — marketplace search architecture,
automated QA design & $450/mo cumulative savings

What We Shipped

Infrastructure & Performance

Faster trades and leaner infrastructure

OpenSearch Right-Sizing
Search servers were memory-heavy but CPU-starved. Swapped all 4 to compute-optimized instances and trimmed excess memory allocation causing GC pauses.
$390/mo saved (~$4,680/yr)  ·  CPU saturated → ~50%, GC pauses eliminated
Ehud · DEV-4732
Internal API Routing
Internal calls between legacy and main backends were routing through Cloudflare's global network. Rerouted to talk directly over our internal network.
~90–240ms faster per trade (3 checks × 30–80ms saved each)
Ehud · OPS-126

What We Shipped

Marketplace-Aware Search

Foundation for multi-marketplace growth

Marketplace-Aware OpenSearch: Complete & Testing
Rebuilt inventory search so each external marketplace (HaloSkins, etc.) has its own independent index. Can enable/disable marketplaces without rebuilding anything. Search queries automatically combine results across all active marketplaces.
New marketplaces (UUSkins, IGXE) = just a DB row, no code changes  ·  10+ bugs caught in code review before production
Ehud · DEV-4759, DEV-4629–4654

Now deployed to staging for end-to-end testing with production-scale data

What We Shipped

Tools & Automation

Smarter tooling, cheaper reviews, automated testing ahead

Claude Code Review — Deployed Across 19 Repos
Replaced GitHub Copilot with Claude Code as our automated PR reviewer. Fixed a max-turns limit causing failures on large PRs. Built a reusable skill to propagate config changes.
$60/mo saved (~$114 → ~$54/mo)  ·  Future config updates = one command across all repos
Ehud
Automated QA Platform — Designed & Planned
Designed a full automated testing system (Playwright). Includes login, trades, payments, and even Steam trade confirmation via bot — no human needed. 21 Linear issues created.
Phase 1 (~50 tests) eliminates manual regression testing  ·  Daily canary payments catch Stripe issues before users do
Ehud · OPS-127–147

Reliability

System Health

Investigating — 2 Active Issues
OpenSearch Data Node CPU Spikes
Intermittent 1-minute CPU spikes on OpenSearch data nodes. Root cause under investigation — being actively worked on.
Status: Investigating
Backend CPU Spikes
Related to expanded Nginx limits and increased traffic from Faceit. Being investigated and worked on.
Status: Investigating

Cost & Efficiency

This Week's Savings

OpenSearch Right-Sizing
$390/mo
r5 → c6a instances, heap 14→4GB
Claude Code Review
$60/mo
GitHub Copilot → Claude (~$114 → ~$54)

Plus ~90–240ms faster trades from internal API routing — hard to price, easy to feel

Cumulative Savings

What We've Saved So Far

$450/mo

$5,400 annualized in recurring savings

Last Week
$390
$450
This Week

+$60/mo from replacing GitHub Copilot with Claude Code Review

Team Focus

This Week's Themes

Infrastructure

OpenSearch right-sizing, internal API routing

Architecture

Multi-marketplace search system built and in testing

Tooling

Claude PR reviews, QA platform design

Busiest ops week yet — 6 operations across infra, architecture & tooling

Developer Quality of Life

Claude Code Integration

AI-powered development is now part of our daily workflow

Team Workshop Prepared
Full training deck built — 10 horizontal topics with 4–8 sub-slides each. Covers core capabilities, hooks, skills, MCP integrations, memory system, and real workflow examples from our codebase.
Already Live
Automated PR reviews across 19 repos, custom skills for deploy & cost monitoring, workflow propagation across all repos in one command.
Persistent Context
Memory system retains project architecture, business rules, and preferences across sessions — no re-explaining needed.

Developer Quality of Life

Ghostty & Modern Shell Stack

Full migration guide prepared for the team

Ghostty Terminal
GPU-accelerated (Metal), 2–3x faster rendering, 5–8x less memory than iTerm2. Native macOS tabs, sub-millisecond keypress-to-screen, plain-text config.
Starship Prompt
Rust-based, ~5ms render. Git status, node version, python env resolve in parallel.
Zinit Plugins
Autosuggestions, syntax highlighting, fzf-tab for fuzzy completions.
CLI Tools
zoxide (smart cd), Atuin (shell history search), eza (modern ls), fzf.

Team Activity

Development Trends (Swarmia)

Review Rate
77% ← 58%
+19pp improvement week-over-week
Time to First Review
6.5h ← 58h
9x faster — Claude Code reviews are instant
Cycle Time
3.1d ← 3.9d
Faster from code to production
Deploys & PRs
9 deploys · 15 PRs
5 contributors · steady deploy pace

Change failure rate up (67% vs 44%) — fewer PRs but more hotfixes. MTTR improved to 13h (from 27h).

Looking Ahead

What's Next

Marketplace Search → Production
Finish staging validation with production-scale data, then ship the multi-marketplace search system
Automated QA — Phase 1 Kickoff
Start building the first batch of Playwright tests — login, inventory search, and core trade flows
Continue Infrastructure Optimization
More right-sizing opportunities across the stack — every dollar saved goes toward building features

Did You Know?

90–240ms

shaved off every single trade this week

Our internal API calls were taking a world tour through Cloudflare
before coming back to the server next door.
Now they just knock on the wall. 🏠