Read the test. Try the question.
grill-with-docs
The prompts and scoring guides used to test grill-with-docs, with each report’s findings and limits.
mattpocock/skills
A relentless interview to sharpen a plan or design, which also creates docs (ADR's and glossary) as we go.
The task
decision
Link to this taskWritten responses only. No tool execution, installed dependencies, media generation or cloud services are tested.
What we asked
Do not use tools, access the network, write files, run commands or install anything. Answer only in text using the supplied facts. Interview me about a reservation design and propose documentation to preserve decisions. Facts: a hold lasts 10 minutes; capture happens only at confirmation. The term booking is currently used for both a hold and a confirmed reservation.
The scoring guide
Pass threshold: 0.75.
Driftproof-authored from the pinned SKILL.md and README; no separate drafting model call.
Try it yourself
Paste the prompt into your own assistant, with or without the skill. Use the scoring guide to read the answer.
One try scored by hand is not a measurement.
Report 017: several answers per arm, each scored 5 times by a judge model. The report’s findings and caveats are below.
Report 018: several answers per arm, each scored 5 times by a judge model. The report’s findings and caveats are below.
Copying stays in your browser. Copies are never stored, counted or sent anywhere by Driftproof.
What happened on this task
Open a report to read its comparison and limits together. Beating the fake is not beating no skill.
Claude Sonnet 5.5Report 017 · Read the task result
Against the same-length fake
decision (clearly lower): with the skill 0.75 plus or minus 0; with the same-length placebo 0.9375 plus or minus 0.075.
Against nothing
decision (clearly lower): with the skill 0.75 plus or minus 0; without it 0.91 plus or minus 0.082158.
About grill-me and grill-with-docs: each skill's whole text is an instruction to call another skill. These text tasks forbid tools, so each reading reflects that one-line hand-off, not the skill it calls. Both stay in the pre-registered counts.
- What we tested. Report 015's twenty skills on Sonnet 5.5, each against no skill and against a same-length text from a fixed corpus, judged by Haiku 5.5.
- Beating the fake is not beating no skill. Where the fake scored below no skill, beating it does not mean beating no skill; each row shows both readings.
- What it does not show. It does not show how these skills do beyond three text tasks each, written by Driftproof for Report 015's twenty skills from the Skills.sh snapshot of 2026-10-17, on Claude Sonnet 5.5 judged by Claude Haiku 5.5.
Driftproof collected several answers per arm. A judge model scored each answer 5 times.
Claude Haiku 5.5Report 018 · Read the task result
Against the same-length fake
decision (too few answers to tell): with the skill 0.88 plus or minus 0.120416; with the same-length placebo 0.966667 plus or minus 0.028868.
Against nothing
decision (too few answers to tell): with the skill 0.88 plus or minus 0.120416; without it 0.87 plus or minus 0.115109.
About grill-me and grill-with-docs: each skill's whole text is an instruction to call another skill. These text tasks forbid tools, so each reading reflects that one-line hand-off, not the skill it calls. Both stay in the pre-registered counts.
- What we tested. Twenty skills on Haiku 5.5, each against no skill and against a same-length text from a fixed corpus, judged by Haiku 5.5, the same model.
- The judge is the target's model. The judge uses the same model as the target. Self-grading may favor its own model's answers.
- Beating the fake is not beating no skill. Where the fake scored below no skill, beating it does not mean beating no skill; each row shows both readings.
- What it does not show. It does not show how these skills do beyond three text tasks each, written by Driftproof for the twenty skills from the Skills.sh snapshot of 2026-10-17, on Claude Haiku 5.5 judged by Claude Haiku 5.5.
Driftproof collected several answers per arm. A judge model scored each answer 5 times.
The task
adr
Link to this taskWritten responses only. No tool execution, installed dependencies, media generation or cloud services are tested.
What we asked
Do not use tools, access the network, write files, run commands or install anything. Answer only in text using the supplied facts. We settled that cancelled reservations stay as tombstones because audit exports reference them. Write the short decision note you would preserve and one remaining question about retention.
The scoring guide
Pass threshold: 0.75.
Driftproof-authored from the pinned SKILL.md and README; no separate drafting model call.
Try it yourself
Paste the prompt into your own assistant, with or without the skill. Use the scoring guide to read the answer.
One try scored by hand is not a measurement.
Report 017: several answers per arm, each scored 5 times by a judge model. The report’s findings and caveats are below.
Report 018: several answers per arm, each scored 5 times by a judge model. The report’s findings and caveats are below.
Copying stays in your browser. Copies are never stored, counted or sent anywhere by Driftproof.
What happened on this task
Open a report to read its comparison and limits together. Beating the fake is not beating no skill.
Claude Sonnet 5.5Report 017 · Read the task result
Against the same-length fake
adr (no clear difference): with the skill 1 plus or minus 0; with the same-length placebo 1 plus or minus 0.
Against nothing
adr (no clear difference): with the skill 1 plus or minus 0; without it 1 plus or minus 0.
About grill-me and grill-with-docs: each skill's whole text is an instruction to call another skill. These text tasks forbid tools, so each reading reflects that one-line hand-off, not the skill it calls. Both stay in the pre-registered counts.
- What we tested. Report 015's twenty skills on Sonnet 5.5, each against no skill and against a same-length text from a fixed corpus, judged by Haiku 5.5.
- Beating the fake is not beating no skill. Where the fake scored below no skill, beating it does not mean beating no skill; each row shows both readings.
- What it does not show. It does not show how these skills do beyond three text tasks each, written by Driftproof for Report 015's twenty skills from the Skills.sh snapshot of 2026-10-17, on Claude Sonnet 5.5 judged by Claude Haiku 5.5.
Driftproof collected several answers per arm. A judge model scored each answer 5 times.
Claude Haiku 5.5Report 018 · Read the task result
Against the same-length fake
adr (too few answers to tell): with the skill 0.9 plus or minus 0.05; with the same-length placebo 0.925 plus or minus 0.086603.
Against nothing
adr (too few answers to tell): with the skill 0.9 plus or minus 0.05; without it 0.9375 plus or minus 0.047871.
About grill-me and grill-with-docs: each skill's whole text is an instruction to call another skill. These text tasks forbid tools, so each reading reflects that one-line hand-off, not the skill it calls. Both stay in the pre-registered counts.
- What we tested. Twenty skills on Haiku 5.5, each against no skill and against a same-length text from a fixed corpus, judged by Haiku 5.5, the same model.
- The judge is the target's model. The judge uses the same model as the target. Self-grading may favor its own model's answers.
- Beating the fake is not beating no skill. Where the fake scored below no skill, beating it does not mean beating no skill; each row shows both readings.
- What it does not show. It does not show how these skills do beyond three text tasks each, written by Driftproof for the twenty skills from the Skills.sh snapshot of 2026-10-17, on Claude Haiku 5.5 judged by Claude Haiku 5.5.
Driftproof collected several answers per arm. A judge model scored each answer 5 times.
The task
revision
Link to this taskWritten responses only. No tool execution, installed dependencies, media generation or cloud services are tested.
What we asked
Do not use tools, access the network, write files, run commands or install anything. Answer only in text using the supplied facts. A prior note says holds expire after 10 minutes. I now propose 30 minutes to help slow checkouts, but have not accepted the capacity tradeoff. Give a written interview response and the status you would put on the proposed note.
The scoring guide
Pass threshold: 0.75.
Driftproof-authored from the pinned SKILL.md and README; no separate drafting model call.
Try it yourself
Paste the prompt into your own assistant, with or without the skill. Use the scoring guide to read the answer.
One try scored by hand is not a measurement.
Report 017: several answers per arm, each scored 5 times by a judge model. The report’s findings and caveats are below.
Report 018: several answers per arm, each scored 5 times by a judge model. The report’s findings and caveats are below.
Copying stays in your browser. Copies are never stored, counted or sent anywhere by Driftproof.
What happened on this task
Open a report to read its comparison and limits together. Beating the fake is not beating no skill.
Claude Sonnet 5.5Report 017 · Read the task result
Against the same-length fake
revision (too few answers to tell): with the skill 0.966667 plus or minus 0.028868; with the same-length placebo 0.8875 plus or minus 0.085391.
Against nothing
revision (too few answers to tell): with the skill 0.966667 plus or minus 0.028868; without it 0.857143 plus or minus 0.093223.
About grill-me and grill-with-docs: each skill's whole text is an instruction to call another skill. These text tasks forbid tools, so each reading reflects that one-line hand-off, not the skill it calls. Both stay in the pre-registered counts.
- What we tested. Report 015's twenty skills on Sonnet 5.5, each against no skill and against a same-length text from a fixed corpus, judged by Haiku 5.5.
- Beating the fake is not beating no skill. Where the fake scored below no skill, beating it does not mean beating no skill; each row shows both readings.
- What it does not show. It does not show how these skills do beyond three text tasks each, written by Driftproof for Report 015's twenty skills from the Skills.sh snapshot of 2026-10-17, on Claude Sonnet 5.5 judged by Claude Haiku 5.5.
Driftproof collected several answers per arm. A judge model scored each answer 5 times.
Claude Haiku 5.5Report 018 · Read the task result
Against the same-length fake
revision (too few answers to tell): with the skill 0.83 plus or minus 0.097468; with the same-length placebo 0.77 plus or minus 0.075829.
Against nothing
revision (too few answers to tell): with the skill 0.83 plus or minus 0.097468; without it 0.7875 plus or minus 0.085391.
About grill-me and grill-with-docs: each skill's whole text is an instruction to call another skill. These text tasks forbid tools, so each reading reflects that one-line hand-off, not the skill it calls. Both stay in the pre-registered counts.
- What we tested. Twenty skills on Haiku 5.5, each against no skill and against a same-length text from a fixed corpus, judged by Haiku 5.5, the same model.
- The judge is the target's model. The judge uses the same model as the target. Self-grading may favor its own model's answers.
- Beating the fake is not beating no skill. Where the fake scored below no skill, beating it does not mean beating no skill; each row shows both readings.
- What it does not show. It does not show how these skills do beyond three text tasks each, written by Driftproof for the twenty skills from the Skills.sh snapshot of 2026-10-17, on Claude Haiku 5.5 judged by Claude Haiku 5.5.
Driftproof collected several answers per arm. A judge model scored each answer 5 times.