Inspectable software recipes for real jobs

Pick a job. Run the recipe. Check the result.

Start free with an executable workflow and downloadable run kit. When the decision matters, get an independent critique, a personalized shortlist, or implementation help.

Use the free cost kit Build my free shortlist

No sign-in or email required for either free result.

Start with a real job

Leave with a checked artifact, not another tool list.

Write and review a research paper

Obsidian → Claude Code + Claude Opus 5 → LaTeX → OpenAI Codex

You produce: A compiled paper, source ledger, review findings, and a resolved-claims checklist

Run this recipe →

Ship a focused software change

Claude Code or Codex → Ponytail → Project checks → Different model reviewer

You produce: A reviewed diff with test, build, and runtime evidence

Run this recipe →

Audit an MCP server before connecting it

Git → OSV-Scanner → OpenSSF Scorecard → Disposable MCP client

You produce: A versioned MCP preflight record with provenance, permission ledger, dependency evidence, negative tests, and an approve, pilot, or reject decision

Run this recipe →

Read the flagship research-paper guide →

FOR BUYERS · $39 ONE-TIME

Decision report

A private shortlist, open-source path, implementation plan, and decision checks. No page-count theater.

Start the questionnaire →See the sample

FREE LIBRARY

Recipes and comparisons

Job-first workflows, exact models and harnesses, add-on boundaries, and reproducible comparison protocols.

Browse the AI library →

IMPLEMENTATION

Integrator matching

Request help with migration, automation, managed hosting, or self-hosted operations. Both sides approve an introduction.

Request a match →

Latest evidence

Decisions with the hidden work included.

Ponytail changed two harnesses differently

Twelve blinded runs passed. Claude got faster with equal code size; Codex produced smaller code but took more time and tokens on the same tiny fixture.

Compare both experiments →

Slack vs. Mattermost cost kit

A fixed 25-person scenario counting migration, infrastructure, backups, and operator labor.

Use the decision kit →

Run the SaaS exit test

Five steps for proving whether an export can reconstruct useful working state before you buy.

Use the exit test →

Free tools

Put your own numbers in.

Both run in your browser. No sign-in, no account, and nothing you type is sent to us or stored.

Lifecycle cost estimator

Price buying it, running it, and leaving it. Returns first-year and steady-state cost, an itemized exit cost, and a switching ratio that says how much the exit should weigh in the decision.

Estimate a stack →

Exit test scorer

Score what a product's export actually gives you back, get the one check that is weakest, and carry the verdict straight into the cost estimator.

Score an export →

More decisions with the work shown

One task across four coding agents Zapier, Activepieces, or n8n Audit an MCP server before connecting it

Start with the job

Needs, current tools, budget, team, and constraints come before vendor logos.

Compare both ownership models

Managed SaaS and open-source options include lock-in, operation, and migration tradeoffs.

Make it work

Use the implementation plan yourself or request a scoped integrator introduction.

Trust policy

Vendors can pay for claim verification. They cannot buy a recommendation.

Affiliate availability, verification status, and sponsorship never improve editorial inclusion, assessment, or ranking.

Read the methodology Vendor verification