Start with the decision
We begin with a job to ship and the constraints around it, not a category page filled with tool logos.
Evidence before recommendations.
Build a stack, not a directory.
Start with the work, fill six required workflow stages, then inspect gaps, handoffs, and the evidence behind every selected tool.
Profile evidence is not hands-on evidence.40 sourced tool profiles
STACK / REFERENCELIVE CATALOG DATA
Illustrative stack — not a tested recommendation
06Evidence-led content site
01ResearchPerplexity, AhrefsCheck required
02DesignLovartCheck required
03BuildCursorCheck required
04BackendNo tool selectedMissing
05DeployCloudflareCovered
06Measure & MaintainCloudflare, PostHog, AhrefsCheck required
50submitted
30measurable bands
20advanced queries
30 of 50 queries showed measurable demand in Google Keyword Planner; 20 advanced to Phase 0B.
00 / DIFFERENCE
Not another AI directory
Discovery is cheap. A defensible decision is not. We evaluate the task, the complete workflow, and the evidence required to ship.
Show the work
Recommendations must point to prompts, artifacts, verification output, checked dates, and clearly labeled unknowns.
Evaluate the whole path
A tool is useful only inside a repeatable path from brief to code, review, deployment, maintenance, and ownership.
Count the rescue work
We record broken attempts, human intervention, elapsed work, and attributable cost instead of showing only the polished demo.
01 / COMPARE
Decisions worth testing
Search demand is strongest where builders are choosing or switching tools. We turn those decisions into controlled tests.
Codex vs Claude Code
One web feature, one repository, one acceptance test — measured for completion quality, intervention cost, and token spend.
10K–100K primary query band · YoY +900%
Cursor vs Claude Code
Editor-native flow versus terminal agent: compare codebase orientation, multi-file edits, review burden, and recovery from mistakes.
1K–10K query band · strong recent Trends signal
Cursor vs Windsurf
A controlled IDE comparison for context retrieval, multi-file changes, terminal work, and the cost of correcting an agent.
1K–10K band in both query directions
02 / METHOD
What we record
Costs, failures, artifacts, and checked dates — enough evidence for another builder to audit the recommendation.
Read the methodologySame brief
Each tool receives the same repository, constraints, and acceptance criteria.
Visible evidence
We keep prompts, artifacts, failures, costs, and the final verification output.
No invented winner
A recommendation remains planned until the test has actually been run and checked.
Freshness built in
Every claim carries a checked date because these products change quickly.
03 / WORKFLOWS
From prompt to production
Workflow content supports the evidence: how to build, test, deploy, and maintain the thing after the demo works.
How to vibe code a website — and ship it safely
Move from a prompt to a deployable site with checkpoints for structure, accessibility, SEO, testing, backups, and ownership.
100–1K band · YoY +900%
AI coding workflow for controlled changes
Turn a request into a scoped plan, failing test, minimal implementation, reviewable diff, and fresh verification evidence.
10–100 band · YoY +900%
Production-readiness check for AI-built sites
A repeatable gate for functionality, security basics, ownership, analytics, indexing controls, and recovery before launch.
Supporting evidence format
NEXT TEST
Which tool actually ships the cleaner result?
The benchmark brief will be public before the recommendation is.
Inspect the test bench