Free resources
Short, clear pieces you can use today. Each one marked Tested ran a real test of ours.
For developers
Is it really done?
Does Claude say “done” when the work is wrong?
TestedSix small coding tasks, 100 runs: did Claude say "done" while the code was wrong?
Does Claude follow rules better when it can run a check?
TestedA check script Claude can run got 59 of 60 replies to follow three rules. The same script, never mentioned, got 0 of 60.
Make it follow your rules
Does a longer CLAUDE.md get obeyed less?
TestedThe same three rules in CLAUDE.md files of 5, 61 and 201 lines, 180 replies: the short file was followed more often.
Does a check that runs on every reply beat a written rule?
TestedA rule in CLAUDE.md held for 43 of 60 replies. With a Stop hook it held for 58 of 60, by a second grader.
Test and audit
How many test cases do you need to tell two setups apart?
TestedA simulation on 240 real graded replies shows how many cases it took to tell two setups apart. 20 to 50 tasks is the usual advice.
Does a team of Claude reviewers find more bugs in code than one?
TestedOne auditor or three found 7 of 8 planted bugs in all 10 audits, and all 10 missed the same race condition.
Not yet tested
How do you test a Claude workflow, not just one answer?
Not retestedTwenty to fifty real failures, graded the same way every time. Why one passing run proves little.
What goes in CLAUDE.md, what in a skill, what in a hook?
Not retestedWhat goes where, when to add each, and how to keep CLAUDE.md short enough to be obeyed.
For businesses
Give it your facts and your reasons
Does Claude do better when you give it your business notes?
TestedWith the notes of a made-up bakery, Claude used the right facts in 30 of 30 replies. Without them, in 0 of 30.
Does telling Claude the reason for a rule make it follow the rule better?
TestedAdding the reason to a rule cut speech-unfriendly writing, against the rule alone, by 28 percent on one model and 69 percent on another. An irrelevant reason did nothing.
Check its work
Does asking Claude to quote your document first make its answers more accurate?
TestedA prompt that makes Claude quote your document before it answers. It did not change accuracy in a test of 144 replies, but every answer came with a findable sentence.
Can a second chat catch mistakes in an email you wrote with AI?
TestedA second Claude chat, given your sheet, caught each of 10 planted mistakes in every review. How it was told who wrote the draft made no difference.
Can a second chat catch wrong facts in a client summary if it has the contract?
TestedA reviewer with the contract caught 30 of 30 planted changes on one model and 29 of 30 on the other. A reviewer without it caught 6 and 0.
Not yet tested
Why does Claude forget your business, and how do you fix it?
Not retestedEvery new chat starts blank. What Claude remembers, what it does not, and the one setup that fixes it.
Which task should you give to AI first?
Not retestedList the repetitive work in your week and pick the one that costs the most time.
Which Claude app do you need: terminal, desktop or phone?
Not retestedMost jobs are solved in the app. Which jobs need the terminal, and why it is less scary than it looks.